diff --git a/README.md b/README.md index 2bc4d8a3879fc657fd3f8fa6c8b63bcf9d76e031..968e4de5831c2c9de686a9e7b372eef33cde59cd 100644 --- a/README.md +++ b/README.md @@ -9,6 +9,7 @@ python_version: "3.12" suggested_hardware: zero-a10g license: cc-by-nc-4.0 pinned: false +header: mini --- # BioLM-NET Workbench diff --git a/app.py b/app.py index 58a8ea03b1ab09a96ab8be3d443da51391267c6d..6ef954d74b6716bfd244705ec95b28ec691a439d 100644 --- a/app.py +++ b/app.py @@ -34,7 +34,8 @@ from biolmnet.data import ( validate_and_align_omics, ) from biolmnet.training import ( - DEFAULT_VALIDATION_FRACTION, + DEFAULT_TEST_FRACTION, + EARLY_STOPPING_FRACTION, Hyperparameters, ModelBundle, pathway_importance, @@ -173,7 +174,7 @@ def _run_state_rows(active: str, workspace, bundle, run_meta: dict, align: dict, ("Dataset", dataset, None), ("Architecture", arch_text, arch_tone), ("Model", model_text, model_tone), - ("Val accuracy", f"{bundle.metrics['accuracy']:.3f}" if bundle else "—", None), + (f"{_score_split(bundle)} accuracy" if bundle else "Test accuracy", f"{bundle.metrics['accuracy']:.3f}" if bundle else "—", None), ("Artifact", "Ready" if bundle else "None", "accent" if bundle else "muted"), ] if active == "predict": @@ -202,8 +203,8 @@ def _run_state_rows(active: str, workspace, bundle, run_meta: dict, align: dict, f"{bundle.metrics['epochs_completed']}" if bundle else "—", None, ), - ("Val accuracy", f"{bundle.metrics['accuracy']:.3f}" if bundle else "—", "accent" if bundle else None), - ("Macro F1", f"{bundle.metrics['f1_macro']:.3f}" if bundle else "—", None), + (f"{_score_split(bundle)} accuracy" if bundle else "Test accuracy", f"{bundle.metrics['accuracy']:.3f}" if bundle else "—", "accent" if bundle else None), + (f"{_score_split(bundle)} macro F1" if bundle else "Test macro F1", f"{bundle.metrics['f1_macro']:.3f}" if bundle else "—", None), ( "Predictions", f"{predicted['n_samples']:,} scored" if predicted.get("n_samples") else "None yet", @@ -500,7 +501,7 @@ def _loss_plot(history: list[dict[str, float]]) -> go.Figure: ) figure.add_trace( go.Scatter( - x=frame["epoch"], y=frame["validation_loss"], mode="lines", name="Validation", + x=frame["epoch"], y=frame["validation_loss"], mode="lines", name="Validation (early stopping)", line={"color": "#2c455d", "width": 2.5}, ) ) @@ -517,11 +518,14 @@ def _loss_plot(history: list[dict[str, float]]) -> go.Figure: return figure -def _validation_split_note(validation_fraction: float) -> str: - validation_percent = round(float(validation_fraction) * 100) +def _split_note(test_fraction: float) -> str: + test_percent = round(float(test_fraction) * 100) + validation_percent = round((1 - float(test_fraction)) * EARLY_STOPPING_FRACTION * 100) return ( - f'
{validation_percent}% of samples are held out for validation ' - f"(stratified by class). The remaining {100 - validation_percent}% are used for training.
" + f'
{test_percent}% of samples are held out as an untouched test set ' + f"(stratified by class); every reported score comes from it. Of the remaining " + f"{100 - test_percent}%, {validation_percent}% of all samples choose the early-stopping checkpoint " + f"and {100 - test_percent - validation_percent}% train the model.
" ) @@ -534,7 +538,7 @@ def estimate_training_duration( dropout: float, projection_dim: int, fusion_dim: int, - validation_fraction: float, + test_fraction: float, optimizer: str, class_weighting: bool, progress=None, @@ -587,14 +591,14 @@ def _architecture_audit(bundle: ModelBundle) -> pd.DataFrame: return pd.DataFrame(rows) -def _macro_auc(validation_predictions: pd.DataFrame, label_names: list[str]) -> float | None: - """Macro-averaged one-vs-rest AUC, computed here from the stored validation +def _macro_auc(test_predictions: pd.DataFrame, label_names: list[str]) -> float | None: + """Macro-averaged one-vs-rest AUC, computed here from the stored test-set probabilities — presentation-layer only; does not change what train() computes or returns.""" try: probability_columns = [f"P({label})" for label in label_names] - y_true = validation_predictions["observed"] - y_score = validation_predictions[probability_columns].to_numpy() + y_true = test_predictions["observed"] + y_score = test_predictions[probability_columns].to_numpy() y_true_indices = y_true.map({label: index for index, label in enumerate(label_names)}).to_numpy() if len(label_names) < 2 or len(set(y_true_indices)) < 2: return None @@ -615,7 +619,7 @@ def train_workspace( dropout: float, projection_dim: int, fusion_dim: int, - validation_fraction: float, + test_fraction: float, optimizer: str, class_weighting: bool, progress=gr.Progress(track_tqdm=False), @@ -637,7 +641,7 @@ def train_workspace( parameters = Hyperparameters( epochs=int(epochs), batch_size=int(batch_size), learning_rate=float(learning_rate), weight_decay=float(weight_decay), dropout=float(dropout), projection_dim=int(projection_dim), - fusion_dim=int(fusion_dim), validation_fraction=float(validation_fraction), optimizer=optimizer, + fusion_dim=int(fusion_dim), test_fraction=float(test_fraction), optimizer=optimizer, class_weighting=bool(class_weighting), ) @@ -650,12 +654,13 @@ def train_workspace( bundle = result.bundle artifact_path = save_bundle(bundle) metrics = bundle.metrics - macro_auc = _macro_auc(result.validation_predictions, bundle.label_names) + macro_auc = _macro_auc(result.test_predictions, bundle.label_names) minutes, seconds = divmod(int(elapsed_seconds), 60) metrics_html = ui.simple_status_html( - f"Training complete. Validation accuracy {metrics['accuracy']:.3f} · " - f"elapsed {minutes:02d}:{seconds:02d}. Best validation checkpoint restored; the " + f"Training complete. Test accuracy {metrics['accuracy']:.3f} · " + f"test macro F1 {metrics['f1_macro']:.3f} · elapsed {minutes:02d}:{seconds:02d}. Scores come from the " + "untouched test set; the early-stopping checkpoint was chosen on a separate validation slice. The " "downloadable artifact includes architecture, preprocessing, weights and metrics." ) loss_figure = _loss_plot(bundle.history) @@ -673,7 +678,7 @@ def train_workspace( "train_loss": bundle.history[-1]["training_loss"], } return ( - bundle, metrics_html, loss_figure, result.validation_predictions, artifact_path, + bundle, metrics_html, loss_figure, result.test_predictions, artifact_path, importance, architecture_audit, run_meta_update, confusion_html, ) except Exception as exc: @@ -756,7 +761,7 @@ def _export_panel(bundle: ModelBundle | None, artifact_path: str | None, run_met f'
Trained bundle ready
' f'
' f'Produced by the run that finished at {html.escape(run_meta.get("finished_at", "—"))} with ' - f'validation accuracy {bundle.metrics["accuracy"]:.3f} over {len(bundle.label_names)} classes. ' + f'{_score_split(bundle).lower()} accuracy {bundle.metrics["accuracy"]:.3f} over {len(bundle.label_names)} classes. ' f'Bundle size {ui.human_bytes(total_size)}.
', extra_class="", ) @@ -944,12 +949,28 @@ def _pathway_attention_table(bundle: ModelBundle) -> pd.DataFrame: return importance[["branch", "pathway", "genes", "peak_gene_attention", "attention_entropy"]] +def _for_display(frame: pd.DataFrame | None) -> pd.DataFrame: + """Round float columns to 3 decimals for on-screen tables only; the + downloadable prediction file keeps full precision.""" + if frame is None or frame.empty: + return frame if frame is not None else pd.DataFrame() + float_columns = frame.select_dtypes("float").columns + # Model outputs are float32; widen before rounding so 0.02 shows as 0.02, not 0.0199999995. + return frame.astype({column: "float64" for column in float_columns}).round({column: 3 for column in float_columns}) + + +def _score_split(bundle: ModelBundle) -> str: + """Which partition a bundle's metrics come from. Artifacts saved before the + 80/20 test split carry validation scores and must stay labelled as such.""" + return "Test" if bundle.metrics.get("evaluation_split") == "test" else "Validation" + + def refresh_results(bundle: ModelBundle | None, run_meta: dict): run_meta = run_meta or {} if bundle is None: stats = ( '
' - + "".join(ui.stat_plate("—", label) for label in ("Validation accuracy", "Macro F1", "Macro AUC", "Retained pathways")) + + "".join(ui.stat_plate("—", label) for label in ("Test accuracy", "Test macro F1", "Test macro AUC", "Retained pathways")) + "
" ) return ( @@ -963,14 +984,15 @@ def refresh_results(bundle: ModelBundle | None, run_meta: dict): ) macro_auc = run_meta.get("macro_auc") + split = _score_split(bundle) retained = len(bundle.config["gene_pathways"]) + len(bundle.config["dna_pathways"]) stats = ( '
' + "".join( [ - ui.stat_plate(f"{bundle.metrics['accuracy']:.3f}", "Validation accuracy"), - ui.stat_plate(f"{bundle.metrics['f1_macro']:.3f}", "Macro F1"), - ui.stat_plate(f"{macro_auc:.3f}" if macro_auc is not None else "—", "Macro AUC"), + ui.stat_plate(f"{bundle.metrics['accuracy']:.3f}", f"{split} accuracy"), + ui.stat_plate(f"{bundle.metrics['f1_macro']:.3f}", f"{split} macro F1"), + ui.stat_plate(f"{macro_auc:.3f}" if macro_auc is not None else "—", f"{split} macro AUC"), ui.stat_plate(f"{retained:,}", "Retained pathways"), ] ) @@ -1050,7 +1072,7 @@ with gr.Blocks(title="BioLM-NET Workbench", fill_width=True) as demo: with gr.Column(min_width=0): gr.HTML('
Repository dataset
') example_dataset = gr.Dropdown( - ["BRCA", "COAD", "GBM", "scTrioseq2"], value="BRCA", + ["scTrioseq2", "BRCA", "COAD", "GBM"], value="scTrioseq2", show_label=False, container=False, elem_id="dataset-select", ) with gr.Column(visible=False) as github_group: @@ -1168,7 +1190,7 @@ with gr.Blocks(title="BioLM-NET Workbench", fill_width=True) as demo: gr.HTML( ui.title_block_html( "Train Model", - "Fit the prepared architecture with a stratified validation split and balanced " + "Fit the prepared architecture on a stratified 80/20 train/test split with balanced " "loss options.", ) ) @@ -1201,7 +1223,7 @@ with gr.Blocks(title="BioLM-NET Workbench", fill_width=True) as demo: with gr.Column(): dropout = gr.Slider(0, 0.8, value=0.3, step=0.05, show_label=False, container=False) - gr.HTML('
B · Validation & loss
') + gr.HTML('
B · Split & loss
') with gr.Row(elem_classes=["srow-row", "first"]): with gr.Column(min_width=140): gr.HTML(ui.esc("Branch projection")) @@ -1214,15 +1236,13 @@ with gr.Blocks(title="BioLM-NET Workbench", fill_width=True) as demo: fusion_dim = gr.Dropdown([8, 12, 16, 32], value=12, show_label=False, container=False) with gr.Row(elem_classes=["srow-row"]): with gr.Column(min_width=140): - gr.HTML(ui.esc("Validation fraction")) + gr.HTML(ui.esc("Test fraction")) with gr.Column(): - validation_fraction = gr.Slider( - 0.1, 0.4, value=DEFAULT_VALIDATION_FRACTION, step=0.05, show_label=False, container=False - ) - validation_split_note = gr.HTML(_validation_split_note(DEFAULT_VALIDATION_FRACTION)) - validation_fraction.change( - _validation_split_note, inputs=[validation_fraction], outputs=[validation_split_note] + test_fraction = gr.Slider( + 0.1, 0.4, value=DEFAULT_TEST_FRACTION, step=0.05, show_label=False, container=False ) + split_note = gr.HTML(_split_note(DEFAULT_TEST_FRACTION)) + test_fraction.change(_split_note, inputs=[test_fraction], outputs=[split_note]) with gr.Row(elem_classes=["srow-row"]): with gr.Column(min_width=140): gr.HTML(ui.esc("Optimizer")) @@ -1294,6 +1314,11 @@ with gr.Blocks(title="BioLM-NET Workbench", fill_width=True) as demo: ) with gr.Row(elem_classes=["actbar-row"]): gr.HTML('
Bundle written to a temp file on export · not persisted between Space restarts
') + with gr.Row(elem_classes=["actbar-buttons"]): + continue_predict_button = gr.Button( + "Continue to Predict", size="sm", variant="primary", + elem_id="continue-predict-btn", elem_classes=["actbar-primary-btn"], + ) # ── Predict ────────────────────────────────────────────── with gr.Column(visible=False, elem_id="page-predict", elem_classes=["workspace-page"]) as predict_page: @@ -1371,7 +1396,7 @@ with gr.Blocks(title="BioLM-NET Workbench", fill_width=True) as demo: ui.title_block_html( "Results", "Everything the run produced, read in one place: architecture audit, " - "validation outputs, pathway attention and the scored cohort.", + "test-set outputs, pathway attention and the scored cohort.", ) ) results_empty_note = gr.HTML(ui.empty_note_html("Build the biological architecture, train a model, or run prediction to populate this page.")) @@ -1384,7 +1409,7 @@ with gr.Blocks(title="BioLM-NET Workbench", fill_width=True) as demo: gr.HTML('
Training history
') results_loss_plot = gr.Plot(show_label=False) with gr.Column(scale=6, min_width=0): - gr.HTML('
Confusion matrix · validation split
') + gr.HTML('
Confusion matrix · test set
') results_confusion = gr.HTML("") with gr.Row(elem_classes=["results-grid"], equal_height=False): with gr.Column(scale=6, min_width=0): @@ -1393,9 +1418,9 @@ with gr.Blocks(title="BioLM-NET Workbench", fill_width=True) as demo: with gr.Column(scale=6, min_width=0): gr.HTML('
Pathway attention · top retained
') results_pathway_table = gr.Dataframe(max_height=TABLE_MAX_HEIGHT, interactive=False, wrap=False, value=pd.DataFrame(), headers=[""], column_count=(1, "dynamic")) - with gr.Column(visible=False) as results_tables_bottom: - gr.HTML('
Validation predictions
') - results_validation_table = gr.Dataframe(max_height=TABLE_MAX_HEIGHT, interactive=False, wrap=False, value=pd.DataFrame(), headers=[""], column_count=(1, "dynamic")) + with gr.Column(visible=False, elem_classes=["results-bottom"]) as results_tables_bottom: + gr.HTML('
Test-set predictions
') + results_test_table = gr.Dataframe(max_height=TABLE_MAX_HEIGHT, interactive=False, wrap=False, value=pd.DataFrame(), headers=[""], column_count=(1, "dynamic")) gr.HTML('
Predictions and class probabilities
') results_prediction_table = gr.Dataframe(max_height=TABLE_MAX_HEIGHT, interactive=False, wrap=False, value=pd.DataFrame(), headers=[""], column_count=(1, "dynamic")) with gr.Row(elem_classes=["actbar-row"]): @@ -1433,7 +1458,7 @@ with gr.Blocks(title="BioLM-NET Workbench", fill_width=True) as demo: intro_next_buttons.append(next_btn) else: back_btn = gr.Button("Back", size="sm", variant="secondary", elem_classes=["btn-ghost"]) - start_btn = gr.Button("Start with the BRCA example", size="sm", variant="primary") + start_btn = gr.Button("Start with the scTrioseq2 example", size="sm", variant="primary") intro_back_buttons.append(back_btn) intro_start_buttons.append(start_btn) intro_cards.append(intro_card) @@ -1452,10 +1477,21 @@ with gr.Blocks(title="BioLM-NET Workbench", fill_width=True) as demo: return _handler + nav_outputs = [active_page_state, *page_columns, *chrome_outputs] + for button, key in zip(rail_row_buttons, STAGE_KEYS): - button.click( - _nav(key), inputs=chrome_inputs_tail, outputs=[active_page_state, *page_columns, *chrome_outputs] - ) + button.click(_nav(key), inputs=chrome_inputs_tail, outputs=nav_outputs) + + def _advance_to(page_key): + """Move to the next workflow page once a step has succeeded; on failure + stay put so the error message stays in view.""" + def _handler(result, workspace, bundle, run_meta, align, predicted, session_id): + succeeded = not result.empty if isinstance(result, pd.DataFrame) else result is not None + if not succeeded: + return tuple(gr.skip() for _ in nav_outputs) + return _enter_page(page_key, workspace, bundle, run_meta, align, predicted, session_id) + + return _handler def _intro_updates(index: int, visible: bool = True): index = max(0, min(len(ui.INTRO_CARDS) - 1, int(index))) @@ -1468,13 +1504,13 @@ with gr.Blocks(title="BioLM-NET Workbench", fill_width=True) as demo: def _intro_close(dont_show): return 0, gr.update(visible=False), bool(dont_show) - def _intro_start_brca(dont_show): + def _intro_start_example(dont_show): return ( 0, gr.update(visible=False), bool(dont_show), "BioLM-NET examples", - "BRCA", + "scTrioseq2", *_source_visibility("BioLM-NET examples"), ) @@ -1505,7 +1541,7 @@ with gr.Blocks(title="BioLM-NET Workbench", fill_width=True) as demo: ) for button in intro_start_buttons: button.click( - _intro_start_brca, inputs=[intro_dont_show], + _intro_start_example, inputs=[intro_dont_show], outputs=[ intro_step_state, intro_group, intro_dismissed_state, source_mode, example_dataset, example_group, github_group, upload_group, @@ -1570,6 +1606,9 @@ with gr.Blocks(title="BioLM-NET Workbench", fill_width=True) as demo: _after_prepare, inputs=[workspace_state, run_meta_state, session_id_state], outputs=[prepare_button, revalidate_button, *chrome_outputs], show_progress="hidden", + ).then( + _advance_to("train"), inputs=[workspace_state, *chrome_inputs_tail], outputs=nav_outputs, + show_progress="hidden", ) revalidate_button.click( @@ -1586,6 +1625,9 @@ with gr.Blocks(title="BioLM-NET Workbench", fill_width=True) as demo: _after_prepare, inputs=[workspace_state, run_meta_state, session_id_state], outputs=[prepare_button, revalidate_button, *chrome_outputs], show_progress="hidden", + ).then( + _advance_to("train"), inputs=[workspace_state, *chrome_inputs_tail], outputs=nav_outputs, + show_progress="hidden", ) show_symbols_btn.click(_reveal_symbols, inputs=[run_meta_state], outputs=[unmatched_reveal]) @@ -1631,14 +1673,14 @@ with gr.Blocks(title="BioLM-NET Workbench", fill_width=True) as demo: # passed into/out of `train_workspace` itself. train_inputs = [ workspace_state, epochs, batch_size, learning_rate, weight_decay, dropout, projection_dim, - fusion_dim, validation_fraction, optimizer, class_weighting, + fusion_dim, test_fraction, optimizer, class_weighting, ] def _after_train(bundle, run_meta, session_id): chrome = _refresh_chrome("train", bundle is not None and True, bundle, run_meta, {}, {}, session_id) return _idle("Train BioLM-NET on ZeroGPU", ["actbar-primary-btn"]), *chrome - validation_predictions_state = gr.State(pd.DataFrame()) + test_predictions_state = gr.State(pd.DataFrame()) importance_state = gr.State(pd.DataFrame()) architecture_audit_state = gr.State(pd.DataFrame()) confusion_html_state = gr.State("") @@ -1650,7 +1692,7 @@ with gr.Blocks(title="BioLM-NET Workbench", fill_width=True) as demo: ).then( train_workspace, inputs=train_inputs, outputs=[ - model_state, training_status, loss_plot, validation_predictions_state, artifact_path_state, + model_state, training_status, loss_plot, test_predictions_state, artifact_path_state, importance_state, architecture_audit_state, run_meta_update_state, confusion_html_state, ], show_progress="hidden", @@ -1677,13 +1719,13 @@ with gr.Blocks(title="BioLM-NET Workbench", fill_width=True) as demo: outputs=[results_stats, results_stats_row, results_empty_note, results_content, results_loss_plot], show_progress="hidden", ).then( - lambda df: gr.update(value=df), inputs=[validation_predictions_state], outputs=[results_validation_table], + lambda df: gr.update(value=_for_display(df)), inputs=[test_predictions_state], outputs=[results_test_table], show_progress="hidden", ).then( - lambda df: gr.update(value=df), inputs=[architecture_audit_state], outputs=[results_architecture_table], + lambda df: gr.update(value=_for_display(df)), inputs=[architecture_audit_state], outputs=[results_architecture_table], show_progress="hidden", ).then( - lambda bundle: gr.update(value=_pathway_attention_table(bundle)) if bundle else gr.update(value=pd.DataFrame()), + lambda bundle: gr.update(value=_for_display(_pathway_attention_table(bundle))) if bundle else gr.update(value=pd.DataFrame()), inputs=[model_state], outputs=[results_pathway_table], show_progress="hidden", ).then( @@ -1692,6 +1734,9 @@ with gr.Blocks(title="BioLM-NET Workbench", fill_width=True) as demo: ).then( lambda bundle: gr.update(visible=bundle is not None), inputs=[model_state], outputs=[results_tables_bottom], show_progress="hidden", + ).then( + _advance_to("export"), inputs=[model_state, *chrome_inputs_tail], outputs=nav_outputs, + show_progress="hidden", ) # training_status is NOT re-set here: train_workspace's own return # already carries the final "Training complete · accuracy · elapsed" @@ -1746,14 +1791,19 @@ with gr.Blocks(title="BioLM-NET Workbench", fill_width=True) as demo: outputs=chrome_outputs, show_progress="hidden", ).then( - lambda df: gr.update(value=df), inputs=[prediction_table], outputs=[results_prediction_table], + lambda df: gr.update(value=_for_display(df)), inputs=[prediction_table], outputs=[results_prediction_table], show_progress="hidden", ).then( refresh_results, inputs=[model_state, run_meta_state], outputs=[results_stats, results_stats_row, results_empty_note, results_content, results_loss_plot], show_progress="hidden", + ).then( + _advance_to("results"), inputs=[prediction_table, *chrome_inputs_tail], outputs=nav_outputs, + show_progress="hidden", ) + continue_predict_button.click(_nav("predict"), inputs=chrome_inputs_tail, outputs=nav_outputs) + if __name__ == "__main__": demo.queue(default_concurrency_limit=1).launch(theme=THEME, css=CSS) diff --git a/biolmnet/training.py b/biolmnet/training.py index 82ed4a0787f91c46e1bd61fede1d99a0a383c1ce..bc88f567b75ff98a9c973f684549b5d1eeb3885b 100644 --- a/biolmnet/training.py +++ b/biolmnet/training.py @@ -26,9 +26,12 @@ from .model import BioLMNet ProgressCallback = Callable[[float, str], None] -# 75% training / 25% validation. The validation partition drives early -# stopping and best-checkpoint selection, so it is not an untouched test set. -DEFAULT_VALIDATION_FRACTION = 0.25 +# 80% training / 20% test, as in the BioLM-NET paper. The test partition is never +# used during fitting; every reported score comes from it. +DEFAULT_TEST_FRACTION = 0.2 +# Share of the training partition held out to choose the early-stopping checkpoint +# (1/8 of 80% = 10% of all samples), so checkpoint selection never sees the test set. +EARLY_STOPPING_FRACTION = 0.125 # Activations BioLMNet accepts. ACTIVATIONS = ("relu", "sigmoid", "tanh") @@ -47,7 +50,8 @@ class Hyperparameters: dropout: float = 0.3 projection_dim: int = 64 fusion_dim: int = 12 - validation_fraction: float = DEFAULT_VALIDATION_FRACTION + test_fraction: float = DEFAULT_TEST_FRACTION + early_stopping_fraction: float = EARLY_STOPPING_FRACTION optimizer: str = "Adam" class_weighting: bool = True early_stopping_patience: int = 12 @@ -66,8 +70,10 @@ class Hyperparameters: raise ValueError("Learning rate must be in (0, 1].") if not 0 <= self.dropout < 1: raise ValueError("Dropout must be in [0, 1).") - if not 0.05 <= self.validation_fraction <= 0.5: - raise ValueError("Validation fraction must be between 0.05 and 0.5.") + if not 0.05 <= self.test_fraction <= 0.5: + raise ValueError("Test fraction must be between 0.05 and 0.5.") + if not 0.05 <= self.early_stopping_fraction <= 0.5: + raise ValueError("Early-stopping fraction must be between 0.05 and 0.5.") if self.optimizer.lower() not in {"adam", "sgd"}: raise ValueError("Optimizer must be Adam or SGD.") for name in ("biological_activation", "projection_activation", "fusion_activation"): @@ -95,7 +101,7 @@ class ModelBundle: @dataclass class TrainingResult: bundle: ModelBundle - validation_predictions: pd.DataFrame + test_predictions: pd.DataFrame confusion: np.ndarray @@ -107,34 +113,49 @@ def set_reproducible_seed(seed: int) -> None: torch.cuda.manual_seed_all(seed) -def split_train_validation( - labels: np.ndarray, validation_fraction: float, seed: int +def stratified_split( + labels: np.ndarray, holdout_fraction: float, seed: int, holdout_name: str = "test" ) -> tuple[np.ndarray, np.ndarray]: - """Stratified sample-index split shared by both omics and the labels.""" + """Stratified (kept, held-out) index split shared by both omics and the labels.""" classes, class_counts = np.unique(labels, return_counts=True) if class_counts.min() < 2: raise ValueError( - "Each class needs at least two samples for a stratified train/validation split." + f"Each class needs at least two samples for a stratified {holdout_name} split." ) sample_count = len(labels) - validation_count = math.ceil(validation_fraction * sample_count) - train_count = sample_count - validation_count - if min(validation_count, train_count) < len(classes): + holdout_count = math.ceil(holdout_fraction * sample_count) + kept_count = sample_count - holdout_count + if min(holdout_count, kept_count) < len(classes): raise ValueError( - f"A validation fraction of {validation_fraction:.2f} on {sample_count} samples " - f"gives {train_count} training and {validation_count} validation samples; " + f"A {holdout_name} fraction of {holdout_fraction:.2f} on {sample_count} samples " + f"gives {kept_count} kept and {holdout_count} {holdout_name} samples; " f"each partition needs at least one sample per class ({len(classes)} classes) " - "for a stratified split. Adjust the validation fraction or add samples." + f"for a stratified split. Adjust the {holdout_name} fraction or add samples." ) indices = np.arange(sample_count) - train_index, validation_index = train_test_split( + kept_index, holdout_index = train_test_split( indices, - test_size=validation_fraction, + test_size=holdout_fraction, random_state=seed, stratify=labels, ) - return train_index, validation_index + return kept_index, holdout_index + + +def split_train_validation_test( + labels: np.ndarray, test_fraction: float, early_stopping_fraction: float, seed: int +) -> tuple[np.ndarray, np.ndarray, np.ndarray]: + """Training, early-stopping validation and untouched test indices. + + The test set is split off first; the validation set is then taken from the + training part only, so neither checkpoint selection nor fitting sees the test set. + """ + development_index, test_index = stratified_split(labels, test_fraction, seed, "test") + fit_position, validation_position = stratified_split( + labels[development_index], early_stopping_fraction, seed, "early-stopping validation" + ) + return development_index[fit_position], development_index[validation_position], test_index def _fit_scaler(values: np.ndarray) -> tuple[np.ndarray, np.ndarray]: @@ -207,8 +228,11 @@ def train( hyperparameters.validate() set_reproducible_seed(hyperparameters.seed) labels = workspace.labels - train_index, validation_index = split_train_validation( - labels, hyperparameters.validation_fraction, hyperparameters.seed + train_index, validation_index, test_index = split_train_validation_test( + labels, + hyperparameters.test_fraction, + hyperparameters.early_stopping_fraction, + hyperparameters.seed, ) gene_mean, gene_scale = _fit_scaler( workspace.gene_expression[train_index] @@ -226,8 +250,11 @@ def train( dna_validation = _scale( workspace.dna_methylation[validation_index], dna_mean, dna_scale ) + gene_test = _scale(workspace.gene_expression[test_index], gene_mean, gene_scale) + dna_test = _scale(workspace.dna_methylation[test_index], dna_mean, dna_scale) y_train = labels[train_index] y_validation = labels[validation_index] + y_test = labels[test_index] dataset = TensorDataset( torch.from_numpy(gene_train), @@ -331,55 +358,47 @@ def train( if best_state is not None: model.load_state_dict(best_state) - validation_loss, probabilities, predictions = _evaluate( + # Scores come only from the untouched test set. + test_loss, probabilities, predictions = _evaluate( model, - gene_validation, - dna_validation, - y_validation, + gene_test, + dna_test, + y_test, device, ) confusion = confusion_matrix( - y_validation, + y_test, predictions, labels=np.arange(len(workspace.label_names)), ) metrics = { - "validation_loss": validation_loss, - "accuracy": float(accuracy_score(y_validation, predictions)), + "evaluation_split": "test", + "test_loss": test_loss, + "best_validation_loss": best_loss, + "accuracy": float(accuracy_score(y_test, predictions)), "f1_macro": float( - f1_score( - y_validation, predictions, average="macro", zero_division=0 - ) + f1_score(y_test, predictions, average="macro", zero_division=0) ), "f1_weighted": float( - f1_score( - y_validation, predictions, average="weighted", zero_division=0 - ) + f1_score(y_test, predictions, average="weighted", zero_division=0) ), "precision_macro": float( - precision_score( - y_validation, predictions, average="macro", zero_division=0 - ) + precision_score(y_test, predictions, average="macro", zero_division=0) ), "recall_macro": float( - recall_score( - y_validation, predictions, average="macro", zero_division=0 - ) + recall_score(y_test, predictions, average="macro", zero_division=0) ), "epochs_completed": len(history), "training_samples": int(len(train_index)), "validation_samples": int(len(validation_index)), + "test_samples": int(len(test_index)), "device": str(device), } predictions_frame = pd.DataFrame( { - "sample_row": validation_index, - "observed": [ - workspace.label_names[value] for value in y_validation - ], - "predicted": [ - workspace.label_names[value] for value in predictions - ], + "sample_row": test_index, + "observed": [workspace.label_names[value] for value in y_test], + "predicted": [workspace.label_names[value] for value in predictions], } ) for index, label in enumerate(workspace.label_names): @@ -422,7 +441,7 @@ def train( ) return TrainingResult( bundle=bundle, - validation_predictions=predictions_frame, + test_predictions=predictions_frame, confusion=confusion, ) diff --git a/biolmnet/ui/components.py b/biolmnet/ui/components.py index 07c5c853dc8ea8af6b11bee82ebf8fa94b5d55d2..4ed909305d2d314ee45e1eac69a274c371d32dc1 100644 --- a/biolmnet/ui/components.py +++ b/biolmnet/ui/components.py @@ -240,7 +240,7 @@ INTRO_CARDS = [ "KEGG pathways, transcription-factor targets, protein interactions — so trained weights can be read " "back as pathways rather than as a black box." ), - "This workbench runs the model end to end in five stages. On the bundled BRCA example it takes a few minutes.", + "This workbench runs the model end to end in five stages. On the bundled scTrioseq2 example it takes a few minutes.", "", ), ( @@ -280,7 +280,7 @@ INTRO_CARDS = [ "Train Model", ( "Hyperparameters sit on the left as a spec sheet. The run panel on the right reports epoch, train " - "and validation loss, accuracy and ETA as the fit progresses." + "and validation loss, accuracy and ETA as the fit progresses; final scores come from a held-out test set." ), ( "Controls lock for the duration of a run and the primary button reads Training on ZeroGPU " @@ -308,7 +308,7 @@ INTRO_CARDS = [ "Introduction · 06 / 06 · Stage 05", "Reading the results", ( - "Results collects everything a run produced: validation metrics, the training curve, the confusion " + "Results collects everything a run produced: test-set metrics, the training curve, the confusion " "matrix, an audit of how sparse each layer actually is, and the pathway attention weights — which is " "the part the architecture exists to give you." ), diff --git a/biolmnet/ui/styles.py b/biolmnet/ui/styles.py index 7f4840961ebac5e765a8a2bad002747c17aeefa8..06817b3d3312c501d593a0f8e9a7d557bf34b55f 100644 --- a/biolmnet/ui/styles.py +++ b/biolmnet/ui/styles.py @@ -315,6 +315,27 @@ body, .gradio-container { } .results-grid { padding: 26px 30px 0 !important; gap: 26px !important; align-items: flex-start !important; } +/* Results page reads as a report: white cards, body-font tables, full-strength text. */ +#page-results .results-grid > .column, +#page-results .results-bottom { + background: #fff !important; + border: 1px solid var(--color-divider) !important; + border-radius: 6px; + padding: 16px 18px 18px !important; +} +#page-results .results-bottom { margin: 26px 30px 0 !important; } +#page-results .sect { font-size: 11.5px !important; color: var(--color-text) !important; letter-spacing: .08em; } +#page-results .stat { background: #fff; } +#page-results .stat .v { font-size: 34px; color: var(--color-text); } +#page-results .stat .k { font-size: 11px; color: var(--color-muted); } +/* Gradio draws visible rows as a virtual list of divs, not s. */ +#page-results table, #page-results table *, #page-results .virtual-body, #page-results .virtual-body * { font-family: var(--font-body) !important; font-size: 14px !important; font-variant-numeric: tabular-nums; } +#page-results thead th, #page-results thead th * { font: 600 11px/1.3 var(--font-mono) !important; letter-spacing: .06em; } +#page-results tbody td, #page-results .body-cell { color: var(--color-text) !important; background: #fff !important; } +#page-results tbody tr:nth-child(even) td, #page-results .virtual-row:nth-child(even) .body-cell { background: color-mix(in srgb, var(--color-accent) 5%, #fff) !important; } +#page-results thead th { font-size: 11px !important; color: var(--color-muted) !important; background: #fff !important; } +#page-results .table-wrap, #page-results .dataframe { background: #fff !important; } + .sect { margin: 0 0 9px; padding-bottom: 8px; border-bottom: 1px solid var(--color-text); } .sect-spaced { margin-top: 26px !important; } diff --git a/diagnostics/2026-10-30-other-datasets/results/configs_GBM.json b/diagnostics/2026-10-30-other-datasets/results/configs_GBM.json new file mode 100644 index 0000000000000000000000000000000000000000..371b1164acc5df4dfa7e1f2929735a74ee818717 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/configs_GBM.json @@ -0,0 +1,59 @@ +{ + "D default (lab grid, tanh/relu/tanh, b32)": {}, + "PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03)": { + "optimizer": "SGD", + "learning_rate": 0.01, + "weight_decay": 0.01, + "batch_size": 16, + "projection_dim": 64, + "epochs": 200, + "sgd_momentum": 0.03, + "biological_activation": "relu", + "projection_activation": "sigmoid", + "fusion_activation": "tanh" + }, + "BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep)": { + "optimizer": "Adam", + "learning_rate": 0.001, + "weight_decay": 0.0001, + "batch_size": 16, + "projection_dim": 64, + "epochs": 100, + "biological_activation": "relu", + "projection_activation": "sigmoid", + "fusion_activation": "tanh" + }, + "G1 grid b16 p64 tanh/relu/tanh": { + "learning_rate": 0.01, + "weight_decay": 0.01, + "epochs": 200, + "optimizer": "Adam", + "batch_size": 16, + "projection_dim": 64, + "biological_activation": "tanh", + "projection_activation": "relu", + "fusion_activation": "tanh" + }, + "G2 grid b32 p64 tanh/relu/relu": { + "learning_rate": 0.01, + "weight_decay": 0.01, + "epochs": 200, + "optimizer": "Adam", + "batch_size": 32, + "projection_dim": 64, + "biological_activation": "tanh", + "projection_activation": "relu", + "fusion_activation": "relu" + }, + "G3 grid b16 p128 tanh/tanh/tanh": { + "learning_rate": 0.01, + "weight_decay": 0.01, + "epochs": 200, + "optimizer": "Adam", + "batch_size": 16, + "projection_dim": 128, + "biological_activation": "tanh", + "projection_activation": "tanh", + "fusion_activation": "tanh" + } +} \ No newline at end of file diff --git a/diagnostics/2026-10-30-other-datasets/results/configs_common.json b/diagnostics/2026-10-30-other-datasets/results/configs_common.json new file mode 100644 index 0000000000000000000000000000000000000000..b87e43247b17f450483b2a7fe248034079053e13 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/configs_common.json @@ -0,0 +1,5 @@ +{ + "D default (lab grid, tanh/relu/tanh, b32)": {}, + "PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03)": {"optimizer": "SGD", "learning_rate": 0.01, "weight_decay": 0.01, "batch_size": 16, "projection_dim": 64, "epochs": 200, "sgd_momentum": 0.03, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh"}, + "BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep)": {"optimizer": "Adam", "learning_rate": 0.001, "weight_decay": 0.0001, "batch_size": 16, "projection_dim": 64, "epochs": 100, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh"} +} diff --git a/diagnostics/2026-10-30-other-datasets/results/configs_scTrioseq2_top.json b/diagnostics/2026-10-30-other-datasets/results/configs_scTrioseq2_top.json new file mode 100644 index 0000000000000000000000000000000000000000..4e0fb98e27f2debcbefc907d09f73d7b4f4494bb --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/configs_scTrioseq2_top.json @@ -0,0 +1,35 @@ +{ + "S1 grid b32 p128 relu/tanh/tanh": { + "learning_rate": 0.01, + "weight_decay": 0.01, + "epochs": 200, + "optimizer": "Adam", + "batch_size": 32, + "projection_dim": 128, + "biological_activation": "relu", + "projection_activation": "tanh", + "fusion_activation": "tanh" + }, + "S2 grid b32 p128 tanh/relu/tanh": { + "learning_rate": 0.01, + "weight_decay": 0.01, + "epochs": 200, + "optimizer": "Adam", + "batch_size": 32, + "projection_dim": 128, + "biological_activation": "tanh", + "projection_activation": "relu", + "fusion_activation": "tanh" + }, + "S3 grid b32 p128 tanh/tanh/relu": { + "learning_rate": 0.01, + "weight_decay": 0.01, + "epochs": 200, + "optimizer": "Adam", + "batch_size": 32, + "projection_dim": 128, + "biological_activation": "tanh", + "projection_activation": "tanh", + "fusion_activation": "relu" + } +} \ No newline at end of file diff --git a/diagnostics/2026-10-30-other-datasets/results/grid_COAD_seed42.jsonl b/diagnostics/2026-10-30-other-datasets/results/grid_COAD_seed42.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..5ad9922d4da40cfcd85970a38512f1ebe085a4ab --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/grid_COAD_seed42.jsonl @@ -0,0 +1,108 @@ +{"key": {"projection_dim": 64, "act1": "relu", "act2": "relu", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3909893035888672, "best_val_loss": 1.240048885345459} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "relu", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.4272258877754211, "best_val_loss": 1.219422698020935} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "relu", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3968784809112549, "best_val_loss": 1.2141286134719849} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "relu", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 25, "best_epoch": 13, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3599404295285542, "best_val_loss": 1.2035648822784424} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "sigmoid", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.1875, "f1_macro": 0.07894736842105263, "f1_weighted": 0.05921052631578947, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [0, 64, 0, 0], "classes_predicted": 1, "recall": [0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.4072861075401306, "best_val_loss": 1.2022572755813599} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.4442652463912964, "best_val_loss": 1.1492609977722168} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "sigmoid", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.4321313897768657, "best_val_loss": 1.1414717435836792} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "tanh", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3983305891354878, "best_val_loss": 1.2026528120040894} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "relu", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 29, "best_epoch": 17, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.384100337823232, "best_val_loss": 1.178083896636963} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "tanh", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.4272340734799702, "best_val_loss": 1.2163499593734741} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "tanh", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.396897792816162, "best_val_loss": 1.1978163719177246} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "relu", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.379861573378245, "best_val_loss": 1.1927897930145264} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "relu", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.1875, "f1_macro": 0.07894736842105263, "f1_weighted": 0.05921052631578947, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [0, 64, 0, 0], "classes_predicted": 1, "recall": [0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.3861074844996135, "best_val_loss": 1.159651517868042} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "relu", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.4214906692504883, "best_val_loss": 1.1969072818756104} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "sigmoid", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.412907252709071, "best_val_loss": 1.1631500720977783} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "relu", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3805653254191081, "best_val_loss": 1.1614563465118408} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.4119123816490173, "best_val_loss": 1.148013949394226} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "sigmoid", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.1875, "f1_macro": 0.07894736842105263, "f1_weighted": 0.05921052631578947, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [0, 64, 0, 0], "classes_predicted": 1, "recall": [0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.3877576192220051, "best_val_loss": 1.1955989599227905} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.4395293394724529, "best_val_loss": 1.1547168493270874} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "sigmoid", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.43336021900177, "best_val_loss": 1.1464194059371948} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "tanh", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.1875, "f1_macro": 0.07894736842105263, "f1_weighted": 0.05921052631578947, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [0, 64, 0, 0], "classes_predicted": 1, "recall": [0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.3968048294385274, "best_val_loss": 1.1957687139511108} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "sigmoid", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 29, "best_epoch": 17, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.408077786366145, "best_val_loss": 1.1551365852355957} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "tanh", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 22, "best_epoch": 10, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3777307669321697, "best_val_loss": 1.2420912981033325} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "tanh", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3768468697865803, "best_val_loss": 1.2027405500411987} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "tanh", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 23, "best_epoch": 11, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.4129614631334941, "best_val_loss": 1.1951897144317627} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "relu", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.390831708908081, "best_val_loss": 1.2398327589035034} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "relu", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.4272301991780598, "best_val_loss": 1.2192124128341675} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "relu", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3094717065493267, "best_val_loss": 1.2129569053649902} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "tanh", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 29, "best_epoch": 17, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3841036260128021, "best_val_loss": 1.1781153678894043} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "sigmoid", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.1875, "f1_macro": 0.07894736842105263, "f1_weighted": 0.05921052631578947, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [0, 64, 0, 0], "classes_predicted": 1, "recall": [0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.4072534640630086, "best_val_loss": 1.2022759914398193} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.444263478120168, "best_val_loss": 1.1492773294448853} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "tanh", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 18, "best_epoch": 6, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.361961970726649, "best_val_loss": 1.2035990953445435} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "sigmoid", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.432081123193105, "best_val_loss": 1.141425371170044} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "tanh", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3989734649658203, "best_val_loss": 1.19907546043396} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "relu", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3715976774692535, "best_val_loss": 1.2145028114318848} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "tanh", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.4272460341453552, "best_val_loss": 1.2160921096801758} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "tanh", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.1591469248135884, "best_val_loss": 1.197121024131775} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "relu", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3704001903533936, "best_val_loss": 1.2720025777816772} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "relu", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.377534846464793, "best_val_loss": 1.2321189641952515} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "relu", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 29, "best_epoch": 17, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3864441911379497, "best_val_loss": 1.1905261278152466} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "relu", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3983726104100545, "best_val_loss": 1.2272427082061768} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "sigmoid", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3817058801651, "best_val_loss": 1.1318120956420898} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.402257223924001, "best_val_loss": 1.192915916442871} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "relu", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 18, "best_epoch": 6, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3573923707008362, "best_val_loss": 1.1654988527297974} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "sigmoid", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.4438231190045674, "best_val_loss": 1.0954335927963257} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "tanh", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3789408405621846, "best_val_loss": 1.2444725036621094} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "sigmoid", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 15, "best_epoch": 3, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.36286727587382, "best_val_loss": 1.1704843044281006} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "tanh", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3777499794960022, "best_val_loss": 1.2277497053146362} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "tanh", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.1875, "f1_macro": 0.07894736842105263, "f1_weighted": 0.05921052631578947, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [0, 64, 0, 0], "classes_predicted": 1, "recall": [0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.3973684112230937, "best_val_loss": 1.2392377853393555} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.4116776784261067, "best_val_loss": 1.1554261445999146} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "relu", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3817211985588074, "best_val_loss": 1.2915568351745605} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "relu", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.382285475730896, "best_val_loss": 1.2392115592956543} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "sigmoid", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 20, "best_epoch": 8, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3634015421072643, "best_val_loss": 1.1400322914123535} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "relu", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 18, "best_epoch": 6, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3922471006711323, "best_val_loss": 1.1967424154281616} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "sigmoid", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3745754957199097, "best_val_loss": 1.1845200061798096} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.4015939633051555, "best_val_loss": 1.2013922929763794} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "tanh", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.1875, "f1_macro": 0.07894736842105263, "f1_weighted": 0.05921052631578947, "epochs_completed": 23, "best_epoch": 11, "pred_counts": [0, 64, 0, 0], "classes_predicted": 1, "recall": [0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.3559860289096832, "best_val_loss": 1.1589688062667847} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "sigmoid", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.42834476629893, "best_val_loss": 1.0944057703018188} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "tanh", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 16, "best_epoch": 4, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.38552329937617, "best_val_loss": 1.0078891515731812} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "tanh", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 23, "best_epoch": 11, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3633587062358856, "best_val_loss": 1.1935319900512695} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "tanh", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 23, "best_epoch": 11, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.366386592388153, "best_val_loss": 1.2155482769012451} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "tanh", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 17, "best_epoch": 5, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.4204500516255696, "best_val_loss": 1.066306710243225} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "tanh", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 18, "best_epoch": 6, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.4099347293376923, "best_val_loss": 1.1388736963272095} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "relu", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3703823288281758, "best_val_loss": 1.272276520729065} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "relu", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3775306940078735, "best_val_loss": 1.2322691679000854} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "relu", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3062642415364583, "best_val_loss": 1.2243828773498535} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "sigmoid", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3817062775293987, "best_val_loss": 1.1318864822387695} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.4022567868232727, "best_val_loss": 1.1929216384887695} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "sigmoid", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.4420830806096394, "best_val_loss": 1.0954800844192505} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "tanh", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3793847759564717, "best_val_loss": 1.2465585470199585} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "tanh", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3777526021003723, "best_val_loss": 1.2280129194259644} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "relu", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.609375, "f1_macro": 0.19117647058823528, "f1_weighted": 0.5137867647058824, "epochs_completed": 52, "best_epoch": 40, "pred_counts": [59, 0, 5, 0], "classes_predicted": 2, "recall": [0.9069767441860465, 0.0, 0.0, 0.0], "final_train_loss": 1.21039746205012, "best_val_loss": 1.110617756843567} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "tanh", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.421875, "f1_macro": 0.2357142857142857, "f1_weighted": 0.4147321428571428, "epochs_completed": 29, "best_epoch": 17, "pred_counts": [27, 2, 32, 3], "classes_predicted": 4, "recall": [0.4418604651162791, 0.0, 1.0, 0.0], "final_train_loss": 1.0241879522800446, "best_val_loss": 1.168595790863037} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "relu", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 29, "best_epoch": 17, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3840636511643727, "best_val_loss": 1.1780965328216553} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "relu", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.609375, "f1_macro": 0.18932038834951456, "f1_weighted": 0.5087985436893204, "epochs_completed": 36, "best_epoch": 24, "pred_counts": [60, 0, 2, 2], "classes_predicted": 3, "recall": [0.9069767441860465, 0.0, 0.0, 0.0], "final_train_loss": 1.0449495712916057, "best_val_loss": 1.1139811277389526} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "sigmoid", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.413020024696986, "best_val_loss": 1.1631495952606201} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.4119150936603546, "best_val_loss": 1.148013710975647} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "sigmoid", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 29, "best_epoch": 17, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.408005913098653, "best_val_loss": 1.1541296243667603} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "tanh", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3613113860289257, "best_val_loss": 1.2032543420791626} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "tanh", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 29, "best_epoch": 17, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3841348787148793, "best_val_loss": 1.178093671798706} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "tanh", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.625, "f1_macro": 0.19230769230769232, "f1_weighted": 0.5168269230769231, "epochs_completed": 29, "best_epoch": 17, "pred_counts": [61, 0, 3, 0], "classes_predicted": 2, "recall": [0.9302325581395349, 0.0, 0.0, 0.0], "final_train_loss": 1.00029818713665, "best_val_loss": 1.1232004165649414} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "relu", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 18, "best_epoch": 6, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3678318361441295, "best_val_loss": 1.2158886194229126} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "relu", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3622398177782695, "best_val_loss": 1.194149136543274} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "relu", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 18, "best_epoch": 6, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.369064748287201, "best_val_loss": 1.203143835067749} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "sigmoid", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 16, "best_epoch": 4, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3527783652146657, "best_val_loss": 1.2085145711898804} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3967503309249878, "best_val_loss": 1.156850814819336} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "sigmoid", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.125, "f1_macro": 0.05555555555555555, "f1_weighted": 0.027777777777777776, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [0, 0, 64, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0], "final_train_loss": 1.403890570004781, "best_val_loss": 1.1782804727554321} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "tanh", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 20, "best_epoch": 8, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3543736040592194, "best_val_loss": 1.2200701236724854} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "tanh", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3622565070788066, "best_val_loss": 1.1938904523849487} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "tanh", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 18, "best_epoch": 6, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3698947727680206, "best_val_loss": 1.1976171731948853} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "relu", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3639320333798726, "best_val_loss": 1.2280261516571045} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "relu", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3910711805025737, "best_val_loss": 1.1997747421264648} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "relu", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 23, "best_epoch": 11, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3795968095461528, "best_val_loss": 1.1490718126296997} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "sigmoid", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 22, "best_epoch": 10, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3572869102160137, "best_val_loss": 1.1970808506011963} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3977110385894775, "best_val_loss": 1.155841588973999} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "sigmoid", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.125, "f1_macro": 0.05555555555555555, "f1_weighted": 0.027777777777777776, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [0, 0, 64, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0], "final_train_loss": 1.3980871538321178, "best_val_loss": 1.1731362342834473} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "tanh", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 14, "best_epoch": 2, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3688914676507313, "best_val_loss": 1.0352522134780884} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "tanh", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 16, "best_epoch": 4, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.380603164434433, "best_val_loss": 1.1692113876342773} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "tanh", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 20, "best_epoch": 8, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3785443206628163, "best_val_loss": 1.1369270086288452} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "relu", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 18, "best_epoch": 6, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3701550861199696, "best_val_loss": 1.2205222845077515} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "relu", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3622499605019887, "best_val_loss": 1.1941701173782349} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "relu", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.515625, "f1_macro": 0.19434269434269436, "f1_weighted": 0.48641636141636146, "epochs_completed": 54, "best_epoch": 42, "pred_counts": [48, 15, 1, 0], "classes_predicted": 3, "recall": [0.7441860465116279, 0.08333333333333333, 0.0, 0.0], "final_train_loss": 1.0227694809436798, "best_val_loss": 1.1140154600143433} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "sigmoid", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 16, "best_epoch": 4, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3528986970583599, "best_val_loss": 1.2085161209106445} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3967660466829936, "best_val_loss": 1.1568567752838135} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "sigmoid", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.125, "f1_macro": 0.05555555555555555, "f1_weighted": 0.027777777777777776, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [0, 0, 64, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0], "final_train_loss": 1.4031570255756378, "best_val_loss": 1.1782838106155396} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "tanh", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 20, "best_epoch": 8, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.353552758693695, "best_val_loss": 1.193418264389038} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "tanh", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.671875, "f1_macro": 0.20093457943925233, "f1_weighted": 0.5400116822429907, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [64, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.3622585535049438, "best_val_loss": 1.1938930749893188} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "tanh", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.421875, "f1_macro": 0.20723684210526316, "f1_weighted": 0.43585526315789475, "epochs_completed": 35, "best_epoch": 23, "pred_counts": [33, 28, 3, 0], "classes_predicted": 3, "recall": [0.5116279069767442, 0.4166666666666667, 0.0, 0.0], "final_train_loss": 1.0577431321144104, "best_val_loss": 1.0834901332855225} diff --git a/diagnostics/2026-10-30-other-datasets/results/grid_COAD_shard0.log b/diagnostics/2026-10-30-other-datasets/results/grid_COAD_shard0.log new file mode 100644 index 0000000000000000000000000000000000000000..08e91d6a09e31facde7e2a6ab5e6b38d02410304 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/grid_COAD_shard0.log @@ -0,0 +1,55 @@ +[0] 0 {'projection_dim': 64, 'act1': 'relu', 'act2': 'relu', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 2 {'projection_dim': 64, 'act1': 'relu', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 4 {'projection_dim': 64, 'act1': 'relu', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 6 {'projection_dim': 64, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 8 {'projection_dim': 64, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 10 {'projection_dim': 64, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 12 {'projection_dim': 64, 'act1': 'relu', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 14 {'projection_dim': 64, 'act1': 'relu', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 16 {'projection_dim': 64, 'act1': 'relu', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 18 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 20 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 22 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 24 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 26 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 28 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 30 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[0] 32 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 34 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 36 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'relu', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.191 classes=2 +[0] 38 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 40 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.189 classes=3 +[0] 42 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 44 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 46 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 48 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 50 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 52 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.192 classes=2 +[0] 54 {'projection_dim': 128, 'act1': 'relu', 'act2': 'relu', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 56 {'projection_dim': 128, 'act1': 'relu', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 58 {'projection_dim': 128, 'act1': 'relu', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 60 {'projection_dim': 128, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 62 {'projection_dim': 128, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 64 {'projection_dim': 128, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.056 classes=1 +[0] 66 {'projection_dim': 128, 'act1': 'relu', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 68 {'projection_dim': 128, 'act1': 'relu', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 70 {'projection_dim': 128, 'act1': 'relu', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 72 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 74 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 76 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 78 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 80 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 82 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.056 classes=1 +[0] 84 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 86 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 88 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 90 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'relu', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 92 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 94 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.194 classes=3 +[0] 96 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 98 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 100 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.056 classes=1 +[0] 102 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 104 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[0] 106 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.207 classes=3 +SHARD 0 DONE diff --git a/diagnostics/2026-10-30-other-datasets/results/grid_COAD_shard1.log b/diagnostics/2026-10-30-other-datasets/results/grid_COAD_shard1.log new file mode 100644 index 0000000000000000000000000000000000000000..1ff6d55c0b1d13b5742aa04484b9e79b68fc1f46 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/grid_COAD_shard1.log @@ -0,0 +1,55 @@ +[1] 1 {'projection_dim': 64, 'act1': 'relu', 'act2': 'relu', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 3 {'projection_dim': 64, 'act1': 'relu', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 5 {'projection_dim': 64, 'act1': 'relu', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 7 {'projection_dim': 64, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 9 {'projection_dim': 64, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 11 {'projection_dim': 64, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 13 {'projection_dim': 64, 'act1': 'relu', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 15 {'projection_dim': 64, 'act1': 'relu', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 17 {'projection_dim': 64, 'act1': 'relu', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 19 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 21 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 23 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 25 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 27 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 29 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 31 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 33 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 35 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 37 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'relu', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 39 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 41 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 43 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 45 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 47 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 49 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 51 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 53 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 55 {'projection_dim': 128, 'act1': 'relu', 'act2': 'relu', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 57 {'projection_dim': 128, 'act1': 'relu', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 59 {'projection_dim': 128, 'act1': 'relu', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 61 {'projection_dim': 128, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 63 {'projection_dim': 128, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 65 {'projection_dim': 128, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 67 {'projection_dim': 128, 'act1': 'relu', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 69 {'projection_dim': 128, 'act1': 'relu', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 71 {'projection_dim': 128, 'act1': 'relu', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 73 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 75 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 77 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 79 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 81 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 83 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 85 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 87 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 89 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 91 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'relu', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 93 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 95 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 97 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 99 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 101 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 103 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 105 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.201 classes=1 +[1] 107 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.236 classes=4 +SHARD 1 DONE diff --git a/diagnostics/2026-10-30-other-datasets/results/grid_COAD_summary.txt b/diagnostics/2026-10-30-other-datasets/results/grid_COAD_summary.txt new file mode 100644 index 0000000000000000000000000000000000000000..2763c4568bb25c65c0b3c8f0bb7324bff986ea29 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/grid_COAD_summary.txt @@ -0,0 +1,62 @@ +configs: 108 collapsed (1 class): 102 all 5 classes predicted: 0 macro F1 >= 0.70: 0 + +TOP 20 by macro F1 +optimizer batch_size projection_dim acts acc f1_macro f1_w min_recall best epochs pred + Adam 32 128 tanh/tanh/tanh 0.422 0.236 0.415 0.0 17 29 27/2/32/3 + Adam 16 128 tanh/tanh/tanh 0.422 0.207 0.436 0.0 23 35 33/28/3/0 + Adam 32 128 tanh/relu/sigmoid 0.672 0.201 0.540 0.0 1 13 64/0/0/0 + Adam 16 64 tanh/sigmoid/sigmoid 0.672 0.201 0.540 0.0 1 13 64/0/0/0 + Adam 16 64 tanh/sigmoid/relu 0.672 0.201 0.540 0.0 1 13 64/0/0/0 + Adam 16 64 tanh/relu/sigmoid 0.672 0.201 0.540 0.0 17 29 64/0/0/0 + Adam 32 128 tanh/tanh/sigmoid 0.672 0.201 0.540 0.0 1 13 64/0/0/0 + Adam 32 128 tanh/tanh/relu 0.672 0.201 0.540 0.0 1 13 64/0/0/0 + Adam 32 128 tanh/sigmoid/tanh 0.672 0.201 0.540 0.0 1 13 64/0/0/0 + Adam 32 128 tanh/sigmoid/sigmoid 0.672 0.201 0.540 0.0 1 13 64/0/0/0 + Adam 32 128 tanh/sigmoid/relu 0.672 0.201 0.540 0.0 1 13 64/0/0/0 + Adam 32 128 tanh/relu/tanh 0.672 0.201 0.540 0.0 1 13 64/0/0/0 + Adam 32 128 tanh/relu/relu 0.672 0.201 0.540 0.0 1 13 64/0/0/0 + Adam 32 128 sigmoid/relu/relu 0.672 0.201 0.540 0.0 1 13 64/0/0/0 + Adam 16 64 sigmoid/tanh/tanh 0.672 0.201 0.540 0.0 6 18 64/0/0/0 + Adam 32 128 sigmoid/tanh/tanh 0.672 0.201 0.540 0.0 5 17 64/0/0/0 + Adam 32 128 sigmoid/tanh/sigmoid 0.672 0.201 0.540 0.0 11 23 64/0/0/0 + Adam 16 64 sigmoid/tanh/sigmoid 0.672 0.201 0.540 0.0 11 23 64/0/0/0 + Adam 32 128 sigmoid/tanh/relu 0.672 0.201 0.540 0.0 4 16 64/0/0/0 + Adam 32 128 sigmoid/sigmoid/tanh 0.672 0.201 0.540 0.0 1 13 64/0/0/0 + +BY optimizer + n f1_mean f1_max collapsed working +optimizer +Adam 108 0.189 0.236 102 0 + +BY batch_size + n f1_mean f1_max collapsed working +batch_size +16 54 0.190 0.207 49 0 +32 54 0.188 0.236 53 0 + +BY projection_dim + n f1_mean f1_max collapsed working +projection_dim +64 54 0.187 0.201 51 0 +128 54 0.191 0.236 51 0 + +BY act1 + n f1_mean f1_max collapsed working +act1 +relu 36 0.190 0.201 36 0 +sigmoid 36 0.183 0.201 36 0 +tanh 36 0.194 0.236 30 0 + +BY act2 + n f1_mean f1_max collapsed working +act2 +relu 36 0.197 0.201 33 0 +sigmoid 36 0.179 0.201 36 0 +tanh 36 0.192 0.236 33 0 + +BY act3 + n f1_mean f1_max collapsed working +act3 +relu 36 0.180 0.201 35 0 +sigmoid 36 0.201 0.201 36 0 +tanh 36 0.186 0.236 31 0 diff --git a/diagnostics/2026-10-30-other-datasets/results/grid_GBM_seed42.jsonl b/diagnostics/2026-10-30-other-datasets/results/grid_GBM_seed42.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..231496efc5bb45bf84f5925ebf95209e5acdd412 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/grid_GBM_seed42.jsonl @@ -0,0 +1,108 @@ +{"key": {"projection_dim": 64, "act1": "relu", "act2": "relu", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.32786885245901637, "f1_macro": 0.2549019607843137, "f1_weighted": 0.1984892317582771, "epochs_completed": 29, "best_epoch": 17, "pred_counts": [0, 4, 50, 7, 0], "classes_predicted": 3, "recall": [0.0, 0.0, 0.8333333333333334, 1.0, 0.0], "final_train_loss": 1.1435838182115816, "best_val_loss": 1.3830714225769043} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "relu", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.3114754098360656, "f1_macro": 0.24315476190476187, "f1_weighted": 0.18264539422326306, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [49, 2, 0, 10, 0], "classes_predicted": 3, "recall": [0.8666666666666667, 0.08333333333333333, 0.0, 1.0, 0.0], "final_train_loss": 1.1442715349744579, "best_val_loss": 1.4128793478012085} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "relu", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6144538994043902, "best_val_loss": 1.5881210565567017} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "relu", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.3442622950819672, "f1_macro": 0.27715260017050297, "f1_weighted": 0.22553212304166145, "epochs_completed": 20, "best_epoch": 8, "pred_counts": [2, 1, 51, 7, 0], "classes_predicted": 4, "recall": [0.06666666666666667, 0.0, 0.8333333333333334, 1.0, 0.0], "final_train_loss": 1.134312358058867, "best_val_loss": 1.3640681505203247} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "relu", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.620258975550125, "best_val_loss": 1.5694557428359985} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "sigmoid", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6095422604044929, "best_val_loss": 1.5887278318405151} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6183931729832635, "best_val_loss": 1.5824211835861206} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "sigmoid", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 20, "best_epoch": 8, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.5866506738089472, "best_val_loss": 1.583168387413025} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "relu", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.4098360655737705, "f1_macro": 0.4216529296327446, "f1_weighted": 0.3767104652391917, "epochs_completed": 38, "best_epoch": 26, "pred_counts": [10, 17, 23, 7, 4], "classes_predicted": 5, "recall": [0.2, 0.5, 0.5555555555555556, 1.0, 0.09090909090909091], "final_train_loss": 0.9489645453098693, "best_val_loss": 1.3042616844177246} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "tanh", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.26229508196721313, "f1_macro": 0.2365079365079365, "f1_weighted": 0.1312776476710903, "epochs_completed": 26, "best_epoch": 14, "pred_counts": [0, 2, 0, 7, 52], "classes_predicted": 3, "recall": [0.0, 0.0, 0.0, 1.0, 1.0], "final_train_loss": 1.1652457596825772, "best_val_loss": 1.380925178527832} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "tanh", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6144397434641102, "best_val_loss": 1.5881556272506714} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "sigmoid", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 23, "best_epoch": 11, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6089167210573707, "best_val_loss": 1.5869073867797852} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "tanh", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.4426229508196721, "f1_macro": 0.4243478260869565, "f1_weighted": 0.37208579018985294, "epochs_completed": 46, "best_epoch": 34, "pred_counts": [7, 11, 37, 6, 0], "classes_predicted": 4, "recall": [0.13333333333333333, 0.5, 0.7777777777777778, 1.0, 0.0], "final_train_loss": 0.9258286298298445, "best_val_loss": 1.3016177415847778} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6170152627705225, "best_val_loss": 1.5710638761520386} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "relu", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 41, "best_epoch": 29, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6094496145925887, "best_val_loss": 1.6034870147705078} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "sigmoid", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6170247187379931, "best_val_loss": 1.5665905475616455} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "relu", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6128497820734327, "best_val_loss": 1.5870773792266846} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "relu", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 20, "best_epoch": 8, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.5926179729524206, "best_val_loss": 1.5838937759399414} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "sigmoid", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.19672131147540983, "f1_macro": 0.06575342465753424, "f1_weighted": 0.06467549966314844, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [0, 61, 0, 0, 0], "classes_predicted": 1, "recall": [0.0, 1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6095491830117064, "best_val_loss": 1.5882856845855713} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6180330619134538, "best_val_loss": 1.5822943449020386} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "tanh", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.36065573770491804, "f1_macro": 0.35501116427432217, "f1_weighted": 0.2638512824535258, "epochs_completed": 38, "best_epoch": 26, "pred_counts": [42, 13, 0, 6, 0], "classes_predicted": 3, "recall": [0.7333333333333333, 0.5, 0.0, 1.0, 0.0], "final_train_loss": 0.9833631785840936, "best_val_loss": 1.3438191413879395} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "sigmoid", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.19672131147540983, "f1_macro": 0.06575342465753424, "f1_weighted": 0.06467549966314844, "epochs_completed": 20, "best_epoch": 8, "pred_counts": [0, 61, 0, 0, 0], "classes_predicted": 1, "recall": [0.0, 1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.5987929196957031, "best_val_loss": 1.5732991695404053} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "tanh", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.19672131147540983, "f1_macro": 0.06575342465753424, "f1_weighted": 0.06467549966314844, "epochs_completed": 17, "best_epoch": 5, "pred_counts": [0, 61, 0, 0, 0], "classes_predicted": 1, "recall": [0.0, 1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.609694270488343, "best_val_loss": 1.5873526334762573} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "tanh", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6202207719042, "best_val_loss": 1.569453239440918} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "tanh", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 14, "best_epoch": 2, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6109487528358002, "best_val_loss": 1.586789846420288} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "tanh", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 20, "best_epoch": 8, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6045726257595208, "best_val_loss": 1.572192668914795} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "relu", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.5573770491803278, "f1_macro": 0.6026491155902921, "f1_weighted": 0.5550033380313033, "epochs_completed": 55, "best_epoch": 43, "pred_counts": [15, 14, 16, 5, 11], "classes_predicted": 5, "recall": [0.4, 0.6666666666666666, 0.4444444444444444, 0.8, 0.7272727272727273], "final_train_loss": 0.7026482068124364, "best_val_loss": 0.9648270010948181} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "relu", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6144514618024148, "best_val_loss": 1.5881175994873047} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "tanh", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.47540983606557374, "f1_macro": 0.4936209736209737, "f1_weighted": 0.44901841623153105, "epochs_completed": 68, "best_epoch": 56, "pred_counts": [20, 14, 19, 6, 2], "classes_predicted": 5, "recall": [0.5333333333333333, 0.5, 0.5, 1.0, 0.09090909090909091], "final_train_loss": 0.925390218125015, "best_val_loss": 1.2859132289886475} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "relu", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.5409836065573771, "f1_macro": 0.5952165331475676, "f1_weighted": 0.5327840302402201, "epochs_completed": 44, "best_epoch": 32, "pred_counts": [14, 15, 15, 6, 11], "classes_predicted": 5, "recall": [0.4, 0.6666666666666666, 0.3888888888888889, 1.0, 0.6363636363636364], "final_train_loss": 0.689169718919556, "best_val_loss": 0.9460645318031311} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "sigmoid", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6095423255461814, "best_val_loss": 1.5887360572814941} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "relu", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 31, "best_epoch": 19, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6098398719329001, "best_val_loss": 1.592193841934204} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.618389784964056, "best_val_loss": 1.5824193954467773} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "sigmoid", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 20, "best_epoch": 8, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.5867049440008696, "best_val_loss": 1.5831856727600098} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "relu", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6179417176324813, "best_val_loss": 1.572983741760254} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "relu", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 18, "best_epoch": 6, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6090717530641399, "best_val_loss": 1.587717056274414} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "tanh", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.4426229508196721, "f1_macro": 0.5046972635207929, "f1_weighted": 0.4265845051765978, "epochs_completed": 54, "best_epoch": 42, "pred_counts": [9, 16, 16, 5, 15], "classes_predicted": 5, "recall": [0.13333333333333333, 0.8333333333333334, 0.3888888888888889, 1.0, 0.2727272727272727], "final_train_loss": 0.7952200685693918, "best_val_loss": 1.112304925918579} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "tanh", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.614437150824917, "best_val_loss": 1.5881528854370117} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "sigmoid", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 31, "best_epoch": 19, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6137747751559064, "best_val_loss": 1.5918093919754028} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.616969569133279, "best_val_loss": 1.571669578552246} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "tanh", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.5409836065573771, "f1_macro": 0.5611111111111111, "f1_weighted": 0.5231329690346084, "epochs_completed": 72, "best_epoch": 60, "pred_counts": [15, 12, 26, 4, 4], "classes_predicted": 5, "recall": [0.5333333333333333, 0.75, 0.6111111111111112, 0.8, 0.09090909090909091], "final_train_loss": 0.7009447650179836, "best_val_loss": 1.0161054134368896} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "sigmoid", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 15, "best_epoch": 3, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6192130912197091, "best_val_loss": 1.558797001838684} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "relu", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 22, "best_epoch": 10, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6107265740795864, "best_val_loss": 1.6001604795455933} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "relu", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.619074611064515, "best_val_loss": 1.5918970108032227} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "tanh", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 18, "best_epoch": 6, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6103808534601347, "best_val_loss": 1.5706861019134521} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "tanh", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 15, "best_epoch": 3, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6180536890290473, "best_val_loss": 1.5744789838790894} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "tanh", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 15, "best_epoch": 3, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6122698471194408, "best_val_loss": 1.5565847158432007} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "relu", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.5245901639344263, "f1_macro": 0.5696894503346116, "f1_weighted": 0.528221029807492, "epochs_completed": 67, "best_epoch": 55, "pred_counts": [16, 10, 13, 7, 15], "classes_predicted": 5, "recall": [0.4, 0.6666666666666666, 0.4444444444444444, 1.0, 0.45454545454545453], "final_train_loss": 0.863623067980907, "best_val_loss": 1.244720458984375} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "sigmoid", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 20, "best_epoch": 8, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6098492985866109, "best_val_loss": 1.5973776578903198} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6153350964270003, "best_val_loss": 1.589383602142334} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "sigmoid", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.616691382856317, "best_val_loss": 1.5708644390106201} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "relu", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.4098360655737705, "f1_macro": 0.42869062720225515, "f1_weighted": 0.3598863203133123, "epochs_completed": 34, "best_epoch": 22, "pred_counts": [10, 21, 25, 5, 0], "classes_predicted": 4, "recall": [0.2, 0.6666666666666666, 0.5, 1.0, 0.0], "final_train_loss": 1.0046982973651157, "best_val_loss": 1.2295571565628052} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "tanh", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6092947330631193, "best_val_loss": 1.5992294549942017} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "relu", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6202599357386105, "best_val_loss": 1.5694561004638672} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "tanh", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6190709957008154, "best_val_loss": 1.5918669700622559} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "tanh", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.4262295081967213, "f1_macro": 0.47738951379925015, "f1_weighted": 0.41803807704283213, "epochs_completed": 63, "best_epoch": 51, "pred_counts": [9, 17, 16, 6, 13], "classes_predicted": 5, "recall": [0.26666666666666666, 0.5833333333333334, 0.3888888888888889, 1.0, 0.2727272727272727], "final_train_loss": 0.9748181830989858, "best_val_loss": 1.2949243783950806} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "relu", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.5901639344262295, "f1_macro": 0.6113227081473926, "f1_weighted": 0.5856015283546151, "epochs_completed": 36, "best_epoch": 24, "pred_counts": [16, 13, 14, 6, 12], "classes_predicted": 5, "recall": [0.6, 0.6666666666666666, 0.4444444444444444, 0.8, 0.6363636363636364], "final_train_loss": 0.6505402593664784, "best_val_loss": 0.9335175156593323} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "relu", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 22, "best_epoch": 10, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6097198496750795, "best_val_loss": 1.6002179384231567} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "relu", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.618849476829904, "best_val_loss": 1.5916205644607544} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "sigmoid", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.19672131147540983, "f1_macro": 0.06575342465753424, "f1_weighted": 0.06467549966314844, "epochs_completed": 28, "best_epoch": 16, "pred_counts": [0, 61, 0, 0, 0], "classes_predicted": 1, "recall": [0.0, 1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6088988670234472, "best_val_loss": 1.585577368736267} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "relu", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 33, "best_epoch": 21, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6108017638732828, "best_val_loss": 1.5905486345291138} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "sigmoid", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.19672131147540983, "f1_macro": 0.06575342465753424, "f1_weighted": 0.06467549966314844, "epochs_completed": 16, "best_epoch": 4, "pred_counts": [0, 61, 0, 0, 0], "classes_predicted": 1, "recall": [0.0, 1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6152916619035065, "best_val_loss": 1.5933729410171509} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6170157793441104, "best_val_loss": 1.5710649490356445} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.18032786885245902, "f1_macro": 0.061111111111111116, "f1_weighted": 0.055100182149362485, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [0, 0, 0, 0, 61], "classes_predicted": 1, "recall": [0.0, 0.0, 0.0, 0.0, 1.0], "final_train_loss": 1.6152529286556556, "best_val_loss": 1.589416265487671} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "sigmoid", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6171117304452782, "best_val_loss": 1.5666223764419556} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "sigmoid", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.19672131147540983, "f1_macro": 0.06575342465753424, "f1_weighted": 0.06467549966314844, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [0, 61, 0, 0, 0], "classes_predicted": 1, "recall": [0.0, 1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6170354990359863, "best_val_loss": 1.5747703313827515} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "tanh", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.19672131147540983, "f1_macro": 0.06575342465753424, "f1_weighted": 0.06467549966314844, "epochs_completed": 34, "best_epoch": 22, "pred_counts": [0, 61, 0, 0, 0], "classes_predicted": 1, "recall": [0.0, 1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6191146816712259, "best_val_loss": 1.5948519706726074} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "tanh", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.19672131147540983, "f1_macro": 0.06575342465753424, "f1_weighted": 0.06467549966314844, "epochs_completed": 14, "best_epoch": 2, "pred_counts": [0, 61, 0, 0, 0], "classes_predicted": 1, "recall": [0.0, 1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.618957109138614, "best_val_loss": 1.5863590240478516} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "tanh", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.45901639344262296, "f1_macro": 0.45606060606060606, "f1_weighted": 0.40126676602086436, "epochs_completed": 34, "best_epoch": 22, "pred_counts": [9, 18, 26, 6, 2], "classes_predicted": 5, "recall": [0.2, 0.8333333333333334, 0.5555555555555556, 1.0, 0.0], "final_train_loss": 0.95225381492917, "best_val_loss": 1.1211503744125366} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "tanh", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 32, "best_epoch": 20, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6117250710888638, "best_val_loss": 1.59309720993042} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "tanh", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6202226225795642, "best_val_loss": 1.5694526433944702} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "relu", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 22, "best_epoch": 10, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6107376058244967, "best_val_loss": 1.6002317667007446} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "relu", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6190745810993383, "best_val_loss": 1.5918970108032227} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "relu", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.4262295081967213, "f1_macro": 0.418008658008658, "f1_weighted": 0.36725569512454753, "epochs_completed": 38, "best_epoch": 26, "pred_counts": [5, 12, 38, 6, 0], "classes_predicted": 4, "recall": [0.2, 0.4166666666666667, 0.7222222222222222, 1.0, 0.0], "final_train_loss": 0.9166352361277804, "best_val_loss": 1.246969223022461} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "tanh", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.5573770491803278, "f1_macro": 0.5724358974358974, "f1_weighted": 0.5215426649852879, "epochs_completed": 44, "best_epoch": 32, "pred_counts": [11, 14, 30, 5, 1], "classes_predicted": 5, "recall": [0.4, 0.75, 0.7222222222222222, 1.0, 0.09090909090909091], "final_train_loss": 0.8052770531242662, "best_val_loss": 1.033097505569458} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "sigmoid", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 20, "best_epoch": 8, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6098517687594305, "best_val_loss": 1.597347617149353} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6153320960008382, "best_val_loss": 1.5893851518630981} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "sigmoid", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6170860131581624, "best_val_loss": 1.5712255239486694} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "relu", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 31, "best_epoch": 19, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6098565710046904, "best_val_loss": 1.59080970287323} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "tanh", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.45901639344262296, "f1_macro": 0.46420306513409954, "f1_weighted": 0.4178314804346461, "epochs_completed": 56, "best_epoch": 44, "pred_counts": [33, 13, 11, 4, 0], "classes_predicted": 4, "recall": [0.7333333333333333, 0.5833333333333334, 0.3333333333333333, 0.8, 0.0], "final_train_loss": 0.8745194908048286, "best_val_loss": 1.1733287572860718} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "relu", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6190402755320399, "best_val_loss": 1.5791617631912231} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "tanh", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6190676102872754, "best_val_loss": 1.5918607711791992} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "tanh", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.36065573770491804, "f1_macro": 0.36431586914473135, "f1_weighted": 0.3001537186824034, "epochs_completed": 37, "best_epoch": 25, "pred_counts": [8, 18, 29, 6, 0], "classes_predicted": 4, "recall": [0.06666666666666667, 0.5, 0.5555555555555556, 1.0, 0.0], "final_train_loss": 0.8967835701228491, "best_val_loss": 1.2483084201812744} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "relu", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.45901639344262296, "f1_macro": 0.4264264767490574, "f1_weighted": 0.38115612573571317, "epochs_completed": 44, "best_epoch": 32, "pred_counts": [5, 19, 31, 6, 0], "classes_predicted": 4, "recall": [0.13333333333333333, 0.5833333333333334, 0.7777777777777778, 1.0, 0.0], "final_train_loss": 1.0252954515277362, "best_val_loss": 1.3152095079421997} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "sigmoid", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 22, "best_epoch": 10, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6170530039104607, "best_val_loss": 1.5765880346298218} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 15, "best_epoch": 3, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6233703836065825, "best_val_loss": 1.588867425918579} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "sigmoid", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.622798799165611, "best_val_loss": 1.5504661798477173} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "tanh", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 31, "best_epoch": 19, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.60984157929655, "best_val_loss": 1.5915166139602661} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "tanh", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6190970165481984, "best_val_loss": 1.579171061515808} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "tanh", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.3442622950819672, "f1_macro": 0.24398699891657638, "f1_weighted": 0.1960463918441291, "epochs_completed": 17, "best_epoch": 5, "pred_counts": [0, 0, 53, 8, 0], "classes_predicted": 2, "recall": [0.0, 0.0, 0.8888888888888888, 1.0, 0.0], "final_train_loss": 1.1648317367001308, "best_val_loss": 1.3554661273956299} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "relu", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 31, "best_epoch": 19, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6098419343187509, "best_val_loss": 1.5913604497909546} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "relu", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 14, "best_epoch": 2, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6189456181448014, "best_val_loss": 1.5804989337921143} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "relu", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 19, "best_epoch": 7, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6247041023494115, "best_val_loss": 1.5830711126327515} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "sigmoid", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 26, "best_epoch": 14, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6229580614736172, "best_val_loss": 1.5805857181549072} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 25, "best_epoch": 13, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6094001889880238, "best_val_loss": 1.5787322521209717} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "sigmoid", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 35, "best_epoch": 23, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6275016473290698, "best_val_loss": 1.5733928680419922} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "tanh", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 22, "best_epoch": 10, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6101267370369916, "best_val_loss": 1.5844577550888062} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "tanh", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 23, "best_epoch": 11, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6240527063119607, "best_val_loss": 1.5801976919174194} +{"key": {"projection_dim": 128, "act1": "sigmoid", "act2": "tanh", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 14, "best_epoch": 2, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6243553265847794, "best_val_loss": 1.5606313943862915} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "relu", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 15, "best_epoch": 3, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.609915028504335, "best_val_loss": 1.5926764011383057} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "relu", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6190394078447519, "best_val_loss": 1.579162359237671} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "relu", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.39344262295081966, "f1_macro": 0.4003809523809524, "f1_weighted": 0.32702576112412174, "epochs_completed": 46, "best_epoch": 34, "pred_counts": [6, 18, 32, 5, 0], "classes_predicted": 4, "recall": [0.06666666666666667, 0.5833333333333334, 0.6111111111111112, 1.0, 0.0], "final_train_loss": 0.9710713129877393, "best_val_loss": 1.2266485691070557} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "sigmoid", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 22, "best_epoch": 10, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.610132048690254, "best_val_loss": 1.5811766386032104} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 15, "best_epoch": 3, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.623384402749317, "best_val_loss": 1.5888473987579346} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "sigmoid", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [61, 0, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0, 0.0], "final_train_loss": 1.6196753151429808, "best_val_loss": 1.5504440069198608} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "tanh", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 31, "best_epoch": 19, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.60984196167826, "best_val_loss": 1.5914384126663208} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "tanh", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 21, "best_epoch": 9, "pred_counts": [0, 0, 61, 0, 0], "classes_predicted": 1, "recall": [0.0, 0.0, 1.0, 0.0, 0.0], "final_train_loss": 1.6190990326834507, "best_val_loss": 1.579171895980835} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "tanh", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.5901639344262295, "f1_macro": 0.5929791912732021, "f1_weighted": 0.5731392264342787, "epochs_completed": 78, "best_epoch": 66, "pred_counts": [14, 15, 17, 7, 8], "classes_predicted": 5, "recall": [0.5333333333333333, 0.8333333333333334, 0.6111111111111112, 1.0, 0.18181818181818182], "final_train_loss": 0.6872465102398981, "best_val_loss": 0.999639093875885} diff --git a/diagnostics/2026-10-30-other-datasets/results/grid_GBM_shard0.log b/diagnostics/2026-10-30-other-datasets/results/grid_GBM_shard0.log new file mode 100644 index 0000000000000000000000000000000000000000..d30aebedb4785aa941ddcba31daa8b3d0c92600a --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/grid_GBM_shard0.log @@ -0,0 +1,55 @@ +[0] 0 {'projection_dim': 64, 'act1': 'relu', 'act2': 'relu', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.243 classes=3 +[0] 2 {'projection_dim': 64, 'act1': 'relu', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 4 {'projection_dim': 64, 'act1': 'relu', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.422 classes=5 +[0] 6 {'projection_dim': 64, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 8 {'projection_dim': 64, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 10 {'projection_dim': 64, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[0] 12 {'projection_dim': 64, 'act1': 'relu', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.355 classes=3 +[0] 14 {'projection_dim': 64, 'act1': 'relu', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 16 {'projection_dim': 64, 'act1': 'relu', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.494 classes=5 +[0] 18 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 20 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 22 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[0] 24 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 26 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 28 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 30 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[0] 32 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 34 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 36 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'relu', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.429 classes=4 +[0] 38 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 40 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.611 classes=5 +[0] 42 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.066 classes=1 +[0] 44 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 46 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[0] 48 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.456 classes=5 +[0] 50 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 52 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.572 classes=5 +[0] 54 {'projection_dim': 128, 'act1': 'relu', 'act2': 'relu', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 56 {'projection_dim': 128, 'act1': 'relu', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 58 {'projection_dim': 128, 'act1': 'relu', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.426 classes=4 +[0] 60 {'projection_dim': 128, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[0] 62 {'projection_dim': 128, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 64 {'projection_dim': 128, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[0] 66 {'projection_dim': 128, 'act1': 'relu', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 68 {'projection_dim': 128, 'act1': 'relu', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 70 {'projection_dim': 128, 'act1': 'relu', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.244 classes=2 +[0] 72 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 74 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[0] 76 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 78 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 80 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 82 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 84 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[0] 86 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 88 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 90 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'relu', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 92 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 94 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.400 classes=4 +[0] 96 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[0] 98 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 100 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[0] 102 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 104 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[0] 106 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.593 classes=5 +SHARD 0 DONE diff --git a/diagnostics/2026-10-30-other-datasets/results/grid_GBM_shard1.log b/diagnostics/2026-10-30-other-datasets/results/grid_GBM_shard1.log new file mode 100644 index 0000000000000000000000000000000000000000..97ad116f93f1f2bc8259579ddb06ba9a57d8e876 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/grid_GBM_shard1.log @@ -0,0 +1,55 @@ +[1] 1 {'projection_dim': 64, 'act1': 'relu', 'act2': 'relu', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.255 classes=3 +[1] 3 {'projection_dim': 64, 'act1': 'relu', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 5 {'projection_dim': 64, 'act1': 'relu', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.277 classes=4 +[1] 7 {'projection_dim': 64, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 9 {'projection_dim': 64, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 11 {'projection_dim': 64, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[1] 13 {'projection_dim': 64, 'act1': 'relu', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.237 classes=3 +[1] 15 {'projection_dim': 64, 'act1': 'relu', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 17 {'projection_dim': 64, 'act1': 'relu', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.424 classes=4 +[1] 19 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[1] 21 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 23 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 25 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.066 classes=1 +[1] 27 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 29 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.066 classes=1 +[1] 31 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.066 classes=1 +[1] 33 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[1] 35 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[1] 37 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'relu', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.603 classes=5 +[1] 39 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 41 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.595 classes=5 +[1] 43 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 45 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 47 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[1] 49 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.505 classes=5 +[1] 51 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 53 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.561 classes=5 +[1] 55 {'projection_dim': 128, 'act1': 'relu', 'act2': 'relu', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[1] 57 {'projection_dim': 128, 'act1': 'relu', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 59 {'projection_dim': 128, 'act1': 'relu', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.570 classes=5 +[1] 61 {'projection_dim': 128, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 63 {'projection_dim': 128, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[1] 65 {'projection_dim': 128, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[1] 67 {'projection_dim': 128, 'act1': 'relu', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[1] 69 {'projection_dim': 128, 'act1': 'relu', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 71 {'projection_dim': 128, 'act1': 'relu', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.477 classes=5 +[1] 73 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[1] 75 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 77 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[1] 79 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.066 classes=1 +[1] 81 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.061 classes=1 +[1] 83 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.066 classes=1 +[1] 85 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.066 classes=1 +[1] 87 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.066 classes=1 +[1] 89 {'projection_dim': 128, 'act1': 'sigmoid', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 91 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'relu', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[1] 93 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 95 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.418 classes=4 +[1] 97 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 99 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[1] 101 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.091 classes=1 +[1] 103 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.464 classes=4 +[1] 105 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.079 classes=1 +[1] 107 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.364 classes=4 +SHARD 1 DONE diff --git a/diagnostics/2026-10-30-other-datasets/results/grid_GBM_summary.txt b/diagnostics/2026-10-30-other-datasets/results/grid_GBM_summary.txt new file mode 100644 index 0000000000000000000000000000000000000000..94526aa72737360db656f4251e688d352d769562 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/grid_GBM_summary.txt @@ -0,0 +1,62 @@ +configs: 108 collapsed (1 class): 83 all 5 classes predicted: 12 macro F1 >= 0.70: 0 + +TOP 20 by macro F1 +optimizer batch_size projection_dim acts acc f1_macro f1_w min_recall best epochs pred + Adam 16 64 tanh/relu/tanh 0.590 0.611 0.586 0.444 24 36 16/13/14/6/12 + Adam 32 64 tanh/relu/relu 0.557 0.603 0.555 0.400 43 55 15/14/16/5/11 + Adam 32 64 tanh/relu/tanh 0.541 0.595 0.533 0.389 32 44 14/15/15/6/11 + Adam 16 128 tanh/tanh/tanh 0.590 0.593 0.573 0.182 66 78 14/15/17/7/8 + Adam 16 64 tanh/tanh/tanh 0.557 0.572 0.522 0.091 32 44 11/14/30/5/1 + Adam 32 128 relu/relu/tanh 0.525 0.570 0.528 0.400 55 67 16/10/13/7/15 + Adam 32 64 tanh/tanh/tanh 0.541 0.561 0.523 0.091 60 72 15/12/26/4/4 + Adam 32 64 tanh/tanh/relu 0.443 0.505 0.427 0.133 42 54 9/16/16/5/15 + Adam 16 64 relu/tanh/tanh 0.475 0.494 0.449 0.091 56 68 20/14/19/6/2 + Adam 32 128 relu/tanh/tanh 0.426 0.477 0.418 0.267 51 63 9/17/16/6/13 + Adam 32 128 tanh/tanh/relu 0.459 0.464 0.418 0.000 44 56 33/13/11/4/0 + Adam 16 64 tanh/tanh/relu 0.459 0.456 0.401 0.000 22 34 9/18/26/6/2 + Adam 16 64 tanh/relu/relu 0.410 0.429 0.360 0.000 22 34 10/21/25/5/0 + Adam 16 128 relu/relu/tanh 0.459 0.426 0.381 0.000 32 44 5/19/31/6/0 + Adam 32 64 relu/tanh/tanh 0.443 0.424 0.372 0.000 34 46 7/11/37/6/0 + Adam 16 64 relu/relu/tanh 0.410 0.422 0.377 0.091 26 38 10/17/23/7/4 + Adam 32 128 tanh/relu/tanh 0.426 0.418 0.367 0.000 26 38 5/12/38/6/0 + Adam 16 128 tanh/relu/tanh 0.393 0.400 0.327 0.000 34 46 6/18/32/5/0 + Adam 32 128 tanh/tanh/tanh 0.361 0.364 0.300 0.000 25 37 8/18/29/6/0 + Adam 16 64 relu/tanh/relu 0.361 0.355 0.264 0.000 26 38 42/13/0/6/0 + +BY optimizer + n f1_mean f1_max collapsed working +optimizer +Adam 108 0.166 0.611 83 0 + +BY batch_size + n f1_mean f1_max collapsed working +batch_size +16 54 0.165 0.611 42 0 +32 54 0.168 0.603 41 0 + +BY projection_dim + n f1_mean f1_max collapsed working +projection_dim +64 54 0.189 0.611 38 0 +128 54 0.144 0.593 45 0 + +BY act1 + n f1_mean f1_max collapsed working +act1 +relu 36 0.180 0.570 24 0 +sigmoid 36 0.082 0.091 36 0 +tanh 36 0.237 0.611 23 0 + +BY act2 + n f1_mean f1_max collapsed working +act2 +relu 36 0.204 0.611 24 0 +sigmoid 36 0.082 0.091 36 0 +tanh 36 0.213 0.593 23 0 + +BY act3 + n f1_mean f1_max collapsed working +act3 +relu 36 0.161 0.603 27 0 +sigmoid 36 0.085 0.091 36 0 +tanh 36 0.254 0.611 20 0 diff --git a/diagnostics/2026-10-30-other-datasets/results/grid_scTrioseq2_reduced_shard0.log b/diagnostics/2026-10-30-other-datasets/results/grid_scTrioseq2_reduced_shard0.log new file mode 100644 index 0000000000000000000000000000000000000000..ecee240861dce9a95834095494980e837a479dd8 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/grid_scTrioseq2_reduced_shard0.log @@ -0,0 +1,5 @@ +[0] 8 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'relu', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.758 classes=4 +[0] 12 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.767 classes=4 +[0] 16 {'projection_dim': 128, 'act1': 'relu', 'act2': 'relu', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.579 classes=3 +[0] 20 {'projection_dim': 128, 'act1': 'relu', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.698 classes=4 +[0] 24 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'relu', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.540 classes=3 diff --git a/diagnostics/2026-10-30-other-datasets/results/grid_scTrioseq2_reduced_shard1.log b/diagnostics/2026-10-30-other-datasets/results/grid_scTrioseq2_reduced_shard1.log new file mode 100644 index 0000000000000000000000000000000000000000..44c8d393d3b5d5393a060f32620036cf8d1632d1 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/grid_scTrioseq2_reduced_shard1.log @@ -0,0 +1,7 @@ +[1] 9 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'relu', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.785 classes=4 +[1] 13 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.752 classes=4 +[1] 17 {'projection_dim': 128, 'act1': 'relu', 'act2': 'relu', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.681 classes=4 +[1] 21 {'projection_dim': 128, 'act1': 'relu', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.714 classes=4 +[1] 25 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'relu', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.790 classes=4 +[1] 29 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.793 classes=4 +SHARD 1 DONE diff --git a/diagnostics/2026-10-30-other-datasets/results/grid_scTrioseq2_reduced_shard2.log b/diagnostics/2026-10-30-other-datasets/results/grid_scTrioseq2_reduced_shard2.log new file mode 100644 index 0000000000000000000000000000000000000000..5591fe1f8e8543621dd824253f0a542b7fa72598 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/grid_scTrioseq2_reduced_shard2.log @@ -0,0 +1,5 @@ +[2] 6 {'projection_dim': 64, 'act1': 'relu', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.766 classes=4 +[2] 10 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.782 classes=4 +[2] 14 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.626 classes=4 +[2] 18 {'projection_dim': 128, 'act1': 'relu', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.773 classes=4 +[2] 22 {'projection_dim': 128, 'act1': 'relu', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.770 classes=4 diff --git a/diagnostics/2026-10-30-other-datasets/results/grid_scTrioseq2_reduced_shard3.log b/diagnostics/2026-10-30-other-datasets/results/grid_scTrioseq2_reduced_shard3.log new file mode 100644 index 0000000000000000000000000000000000000000..c256776340e99eed410a813fb247be2d0f8cf3d4 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/grid_scTrioseq2_reduced_shard3.log @@ -0,0 +1,7 @@ +[3] 11 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.839 classes=4 +[3] 15 {'projection_dim': 64, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.667 classes=4 +[3] 19 {'projection_dim': 128, 'act1': 'relu', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.744 classes=4 +[3] 23 {'projection_dim': 128, 'act1': 'relu', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.803 classes=4 +[3] 27 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.793 classes=4 +[3] 31 {'projection_dim': 128, 'act1': 'tanh', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.737 classes=4 +SHARD 3 DONE diff --git a/diagnostics/2026-10-30-other-datasets/results/grid_scTrioseq2_seed42.jsonl b/diagnostics/2026-10-30-other-datasets/results/grid_scTrioseq2_seed42.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..96487cfb8928c2928777f97c13001b2edca787f6 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/grid_scTrioseq2_seed42.jsonl @@ -0,0 +1,43 @@ +{"key": {"projection_dim": 64, "act1": "relu", "act2": "relu", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 30, "best_epoch": 18, "pred_counts": [0, 0, 0, 153], "classes_predicted": 1, "recall": [0.0, 0.0, 0.0, 1.0], "final_train_loss": 1.3873135102422613, "best_val_loss": 1.3700900077819824} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "relu", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 25, "best_epoch": 13, "pred_counts": [0, 0, 0, 153], "classes_predicted": 1, "recall": [0.0, 0.0, 0.0, 1.0], "final_train_loss": 1.3934418590445268, "best_val_loss": 1.3605917692184448} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "relu", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.7058823529411765, "f1_macro": 0.721741277442332, "f1_weighted": 0.7041269293147558, "epochs_completed": 44, "best_epoch": 32, "pred_counts": [33, 44, 20, 56], "classes_predicted": 4, "recall": [0.5789473684210527, 0.896551724137931, 0.85, 0.6515151515151515], "final_train_loss": 0.6618062205481947, "best_val_loss": 0.7239845395088196} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "relu", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.5098039215686274, "f1_macro": 0.5525734992198408, "f1_weighted": 0.4410209486329365, "epochs_completed": 47, "best_epoch": 35, "pred_counts": [70, 43, 24, 16], "classes_predicted": 4, "recall": [0.7631578947368421, 0.7931034482758621, 0.95, 0.10606060606060606], "final_train_loss": 0.6456542877774489, "best_val_loss": 0.7737857103347778} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "sigmoid", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [0, 0, 0, 153], "classes_predicted": 1, "recall": [0.0, 0.0, 0.0, 1.0], "final_train_loss": 1.3922991648054959, "best_val_loss": 1.3530312776565552} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "relu", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 19, "best_epoch": 7, "pred_counts": [0, 0, 0, 153], "classes_predicted": 1, "recall": [0.0, 0.0, 0.0, 1.0], "final_train_loss": 1.3946980781722487, "best_val_loss": 1.3498953580856323} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 19, "best_epoch": 7, "pred_counts": [0, 0, 0, 153], "classes_predicted": 1, "recall": [0.0, 0.0, 0.0, 1.0], "final_train_loss": 1.3929637231324847, "best_val_loss": 1.359687089920044} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "sigmoid", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 24, "best_epoch": 12, "pred_counts": [0, 0, 0, 153], "classes_predicted": 1, "recall": [0.0, 0.0, 0.0, 1.0], "final_train_loss": 1.391294981303968, "best_val_loss": 1.345849871635437} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "relu", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.5947712418300654, "f1_macro": 0.549165120593692, "f1_weighted": 0.5109437714479731, "epochs_completed": 32, "best_epoch": 20, "pred_counts": [0, 48, 24, 81], "classes_predicted": 3, "recall": [0.0, 0.896551724137931, 1.0, 0.6818181818181818], "final_train_loss": 0.7018749588414243, "best_val_loss": 0.7821705937385559} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "tanh", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 30, "best_epoch": 18, "pred_counts": [0, 0, 0, 153], "classes_predicted": 1, "recall": [0.0, 0.0, 0.0, 1.0], "final_train_loss": 1.3873152941988225, "best_val_loss": 1.370130181312561} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "tanh", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 25, "best_epoch": 13, "pred_counts": [0, 0, 0, 153], "classes_predicted": 1, "recall": [0.0, 0.0, 0.0, 1.0], "final_train_loss": 1.3934426830525983, "best_val_loss": 1.3605964183807373} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "sigmoid", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.24836601307189543, "f1_macro": 0.09947643979057591, "f1_weighted": 0.09882626698148718, "epochs_completed": 23, "best_epoch": 11, "pred_counts": [153, 0, 0, 0], "classes_predicted": 1, "recall": [1.0, 0.0, 0.0, 0.0], "final_train_loss": 1.392561299759045, "best_val_loss": 1.347267746925354} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 28, "best_epoch": 16, "pred_counts": [0, 0, 0, 153], "classes_predicted": 1, "recall": [0.0, 0.0, 0.0, 1.0], "final_train_loss": 1.3941203732239573, "best_val_loss": 1.3516309261322021} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "tanh", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.7320261437908496, "f1_macro": 0.7494775624212588, "f1_weighted": 0.7316650214051035, "epochs_completed": 58, "best_epoch": 46, "pred_counts": [38, 41, 23, 51], "classes_predicted": 4, "recall": [0.7105263157894737, 0.8275862068965517, 0.95, 0.6363636363636364], "final_train_loss": 0.5055314668437891, "best_val_loss": 0.593417227268219} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "relu", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 14, "best_epoch": 2, "pred_counts": [0, 0, 0, 153], "classes_predicted": 1, "recall": [0.0, 0.0, 0.0, 1.0], "final_train_loss": 1.3876600495555944, "best_val_loss": 1.3639287948608398} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "sigmoid", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [0, 0, 0, 153], "classes_predicted": 1, "recall": [0.0, 0.0, 0.0, 1.0], "final_train_loss": 1.3996041218439739, "best_val_loss": 1.3484610319137573} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "relu", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 25, "best_epoch": 13, "pred_counts": [0, 0, 0, 153], "classes_predicted": 1, "recall": [0.0, 0.0, 0.0, 1.0], "final_train_loss": 1.3939158916473389, "best_val_loss": 1.3590943813323975} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "relu", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [0, 0, 0, 153], "classes_predicted": 1, "recall": [0.0, 0.0, 0.0, 1.0], "final_train_loss": 1.3873379209585357, "best_val_loss": 1.3682854175567627} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "sigmoid", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 13, "best_epoch": 1, "pred_counts": [0, 0, 0, 153], "classes_predicted": 1, "recall": [0.0, 0.0, 0.0, 1.0], "final_train_loss": 1.393721109942386, "best_val_loss": 1.3561733961105347} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "tanh", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.6993464052287581, "f1_macro": 0.7149999077501911, "f1_weighted": 0.6956752523035177, "epochs_completed": 32, "best_epoch": 20, "pred_counts": [33, 38, 21, 61], "classes_predicted": 4, "recall": [0.5526315789473685, 0.8620689655172413, 0.85, 0.6666666666666666], "final_train_loss": 0.679388257495144, "best_val_loss": 0.7309417724609375} +{"key": {"projection_dim": 64, "act1": "sigmoid", "act2": "sigmoid", "act3": "sigmoid", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "sigmoid", "projection_activation": "sigmoid", "fusion_activation": "sigmoid", "sgd_momentum": 0.0}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 19, "best_epoch": 7, "pred_counts": [0, 0, 0, 153], "classes_predicted": 1, "recall": [0.0, 0.0, 0.0, 1.0], "final_train_loss": 1.3931162650125068, "best_val_loss": 1.3597816228866577} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "relu", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.7647058823529411, "f1_macro": 0.7852310106974286, "f1_weighted": 0.762519747023698, "epochs_completed": 79, "best_epoch": 67, "pred_counts": [42, 38, 22, 51], "classes_predicted": 4, "recall": [0.7631578947368421, 0.896551724137931, 0.95, 0.6515151515151515], "final_train_loss": 0.45690638506621645, "best_val_loss": 0.5384410619735718} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "relu", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.8300653594771242, "f1_macro": 0.8389094286870471, "f1_weighted": 0.8280242914220465, "epochs_completed": 87, "best_epoch": 75, "pred_counts": [30, 36, 21, 66], "classes_predicted": 4, "recall": [0.6842105263157895, 0.9310344827586207, 0.95, 0.8333333333333334], "final_train_loss": 0.4291579744272065, "best_val_loss": 0.48823636770248413} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "tanh", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.7450980392156863, "f1_macro": 0.7523300374306969, "f1_weighted": 0.7454765748454624, "epochs_completed": 34, "best_epoch": 22, "pred_counts": [29, 45, 16, 63], "classes_predicted": 4, "recall": [0.6052631578947368, 0.9310344827586207, 0.75, 0.7424242424242424], "final_train_loss": 0.5726166189762584, "best_val_loss": 0.6654586791992188} +{"key": {"projection_dim": 64, "act1": "relu", "act2": "tanh", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.7450980392156863, "f1_macro": 0.765710795935254, "f1_weighted": 0.7464899215976736, "epochs_completed": 62, "best_epoch": 50, "pred_counts": [47, 43, 18, 45], "classes_predicted": 4, "recall": [0.8421052631578947, 0.8620689655172413, 0.85, 0.6060606060606061], "final_train_loss": 0.5049633603346976, "best_val_loss": 0.6015363931655884} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "tanh", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.6143790849673203, "f1_macro": 0.6668107880818652, "f1_weighted": 0.589617804009155, "epochs_completed": 53, "best_epoch": 41, "pred_counts": [66, 38, 21, 28], "classes_predicted": 4, "recall": [0.7631578947368421, 0.9310344827586207, 0.9, 0.30303030303030304], "final_train_loss": 0.57662471919729, "best_val_loss": 0.697319746017456} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "relu", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.7516339869281046, "f1_macro": 0.7584784877811492, "f1_weighted": 0.751614431101028, "epochs_completed": 78, "best_epoch": 66, "pred_counts": [34, 38, 23, 58], "classes_predicted": 4, "recall": [0.6842105263157895, 0.8275862068965517, 0.9, 0.7121212121212122], "final_train_loss": 0.49395094419780533, "best_val_loss": 0.5397946238517761} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "relu", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.6535947712418301, "f1_macro": 0.6810335457084683, "f1_weighted": 0.647391953704101, "epochs_completed": 37, "best_epoch": 25, "pred_counts": [36, 46, 18, 53], "classes_predicted": 4, "recall": [0.5, 0.9655172413793104, 0.8, 0.5606060606060606], "final_train_loss": 0.7002025215249312, "best_val_loss": 0.7461352348327637} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "relu", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.7320261437908496, "f1_macro": 0.7436367558239863, "f1_weighted": 0.7365263045108754, "epochs_completed": 49, "best_epoch": 37, "pred_counts": [38, 43, 16, 56], "classes_predicted": 4, "recall": [0.7368421052631579, 0.8275862068965517, 0.75, 0.6818181818181818], "final_train_loss": 0.4792041349829289, "best_val_loss": 0.5984147191047668} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "tanh", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.7581699346405228, "f1_macro": 0.7671443260981062, "f1_weighted": 0.7502624803508725, "epochs_completed": 31, "best_epoch": 19, "pred_counts": [22, 44, 22, 65], "classes_predicted": 4, "recall": [0.5, 0.9655172413793104, 0.95, 0.7575757575757576], "final_train_loss": 0.6614926405120314, "best_val_loss": 0.6519563794136047} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "tanh", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.6993464052287581, "f1_macro": 0.7140984015984017, "f1_weighted": 0.6961626608685433, "epochs_completed": 58, "best_epoch": 46, "pred_counts": [32, 37, 20, 64], "classes_predicted": 4, "recall": [0.5263157894736842, 0.8275862068965517, 0.85, 0.696969696969697], "final_train_loss": 0.5954567716832746, "best_val_loss": 0.6591925621032715} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "relu", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.7843137254901961, "f1_macro": 0.7824440052700923, "f1_weighted": 0.7863976508477787, "epochs_completed": 58, "best_epoch": 46, "pred_counts": [31, 40, 16, 66], "classes_predicted": 4, "recall": [0.6842105263157895, 0.8620689655172413, 0.75, 0.8181818181818182], "final_train_loss": 0.48087429320603087, "best_val_loss": 0.5209798216819763} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "tanh", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.7973856209150327, "f1_macro": 0.8029562542720436, "f1_weighted": 0.8037893307243152, "epochs_completed": 62, "best_epoch": 50, "pred_counts": [38, 48, 18, 49], "classes_predicted": 4, "recall": [0.868421052631579, 0.9310344827586207, 0.8, 0.696969696969697], "final_train_loss": 0.4686367522206223, "best_val_loss": 0.5512884259223938} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "relu", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.7777777777777778, "f1_macro": 0.7896159238488785, "f1_weighted": 0.7760403403785758, "epochs_completed": 62, "best_epoch": 50, "pred_counts": [39, 35, 24, 55], "classes_predicted": 4, "recall": [0.7631578947368421, 0.8620689655172413, 0.95, 0.696969696969697], "final_train_loss": 0.5012885635359245, "best_val_loss": 0.5802960991859436} +{"key": {"projection_dim": 64, "act1": "tanh", "act2": "tanh", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.5751633986928104, "f1_macro": 0.6264729173265758, "f1_weighted": 0.5401471157568719, "epochs_completed": 38, "best_epoch": 26, "pred_counts": [74, 36, 21, 22], "classes_predicted": 4, "recall": [0.7894736842105263, 0.896551724137931, 0.85, 0.22727272727272727], "final_train_loss": 0.6228602784767485, "best_val_loss": 0.7110384106636047} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "relu", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.6535947712418301, "f1_macro": 0.5785984020062404, "f1_weighted": 0.5638941158555005, "epochs_completed": 53, "best_epoch": 41, "pred_counts": [0, 42, 16, 95], "classes_predicted": 3, "recall": [0.0, 0.9310344827586207, 0.75, 0.8787878787878788], "final_train_loss": 0.7438108549829114, "best_val_loss": 0.7206449508666992} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "relu", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.7908496732026143, "f1_macro": 0.7933473389355743, "f1_weighted": 0.7922172790684902, "epochs_completed": 80, "best_epoch": 68, "pred_counts": [32, 39, 22, 60], "classes_predicted": 4, "recall": [0.7105263157894737, 0.8620689655172413, 0.9, 0.7727272727272727], "final_train_loss": 0.41584639015950653, "best_val_loss": 0.4509934186935425} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "tanh", "act3": "relu", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.7843137254901961, "f1_macro": 0.7928065044344114, "f1_weighted": 0.7787988606958055, "epochs_completed": 64, "best_epoch": 52, "pred_counts": [27, 34, 23, 69], "classes_predicted": 4, "recall": [0.5789473684210527, 0.896551724137931, 0.95, 0.803030303030303], "final_train_loss": 0.6363272583275511, "best_val_loss": 0.6751461029052734} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "tanh", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.6862745098039216, "f1_macro": 0.6983873894895128, "f1_weighted": 0.6823450785860826, "epochs_completed": 36, "best_epoch": 24, "pred_counts": [31, 40, 19, 63], "classes_predicted": 4, "recall": [0.5, 0.8620689655172413, 0.8, 0.6818181818181818], "final_train_loss": 0.6203715529358178, "best_val_loss": 0.671847403049469} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "tanh", "act3": "tanh", "batch_size": 32, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.7189542483660131, "f1_macro": 0.7369299516908212, "f1_weighted": 0.7149142117394461, "epochs_completed": 64, "best_epoch": 52, "pred_counts": [34, 40, 20, 59], "classes_predicted": 4, "recall": [0.5789473684210527, 0.9310344827586207, 0.85, 0.6666666666666666], "final_train_loss": 0.5900494874569407, "best_val_loss": 0.675346314907074} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "relu", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.7647058823529411, "f1_macro": 0.7727172614946489, "f1_weighted": 0.7689027887730965, "epochs_completed": 56, "best_epoch": 44, "pred_counts": [34, 46, 21, 52], "classes_predicted": 4, "recall": [0.7631578947368421, 0.896551724137931, 0.85, 0.6818181818181818], "final_train_loss": 0.5068238717422151, "best_val_loss": 0.6093516945838928} +{"key": {"projection_dim": 128, "act1": "tanh", "act2": "relu", "act3": "relu", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.0}, "accuracy": 0.5882352941176471, "f1_macro": 0.540374677002584, "f1_weighted": 0.5054972893550186, "epochs_completed": 31, "best_epoch": 19, "pred_counts": [0, 52, 23, 78], "classes_predicted": 3, "recall": [0.0, 0.9310344827586207, 0.95, 0.6666666666666666], "final_train_loss": 0.716982879136738, "best_val_loss": 0.8123916387557983} +{"key": {"projection_dim": 128, "act1": "relu", "act2": "tanh", "act3": "tanh", "batch_size": 16, "optimizer": "Adam", "seed": 42}, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.0}, "accuracy": 0.7647058823529411, "f1_macro": 0.7698453537137181, "f1_weighted": 0.7665044761972772, "epochs_completed": 28, "best_epoch": 16, "pred_counts": [42, 39, 21, 51], "classes_predicted": 4, "recall": [0.8157894736842105, 0.8275862068965517, 0.85, 0.6818181818181818], "final_train_loss": 0.5418980226182101, "best_val_loss": 0.6495389342308044} diff --git a/diagnostics/2026-10-30-other-datasets/results/grid_scTrioseq2_shard0.log b/diagnostics/2026-10-30-other-datasets/results/grid_scTrioseq2_shard0.log new file mode 100644 index 0000000000000000000000000000000000000000..37568e61d3233d62143d25df05b2381046fd714c --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/grid_scTrioseq2_shard0.log @@ -0,0 +1,7 @@ +[0] 0 {'projection_dim': 64, 'act1': 'relu', 'act2': 'relu', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.722 classes=4 +[0] 2 {'projection_dim': 64, 'act1': 'relu', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.151 classes=1 +[0] 4 {'projection_dim': 64, 'act1': 'relu', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.549 classes=3 +[0] 6 {'projection_dim': 64, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.099 classes=1 +[0] 8 {'projection_dim': 64, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.151 classes=1 +[0] 10 {'projection_dim': 64, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.151 classes=1 +[0] 12 {'projection_dim': 64, 'act1': 'relu', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 16, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.715 classes=4 diff --git a/diagnostics/2026-10-30-other-datasets/results/grid_scTrioseq2_shard1.log b/diagnostics/2026-10-30-other-datasets/results/grid_scTrioseq2_shard1.log new file mode 100644 index 0000000000000000000000000000000000000000..3451586f479951108bed2df3028684b23cc03f9a --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/grid_scTrioseq2_shard1.log @@ -0,0 +1,14 @@ +[1] 1 {'projection_dim': 64, 'act1': 'relu', 'act2': 'relu', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.151 classes=1 +[1] 3 {'projection_dim': 64, 'act1': 'relu', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.151 classes=1 +[1] 5 {'projection_dim': 64, 'act1': 'relu', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.553 classes=4 +[1] 7 {'projection_dim': 64, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.151 classes=1 +[1] 9 {'projection_dim': 64, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.151 classes=1 +[1] 11 {'projection_dim': 64, 'act1': 'relu', 'act2': 'sigmoid', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.151 classes=1 +[1] 13 {'projection_dim': 64, 'act1': 'relu', 'act2': 'tanh', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.151 classes=1 +[1] 15 {'projection_dim': 64, 'act1': 'relu', 'act2': 'tanh', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.151 classes=1 +[1] 17 {'projection_dim': 64, 'act1': 'relu', 'act2': 'tanh', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.749 classes=4 +[1] 19 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.151 classes=1 +[1] 21 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.151 classes=1 +[1] 23 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'relu', 'act3': 'tanh', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.151 classes=1 +[1] 25 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'relu', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.151 classes=1 +[1] 27 {'projection_dim': 64, 'act1': 'sigmoid', 'act2': 'sigmoid', 'act3': 'sigmoid', 'batch_size': 32, 'optimizer': 'Adam', 'seed': 42} -> f1m=0.151 classes=1 diff --git a/diagnostics/2026-10-30-other-datasets/results/sanity_baselines.txt b/diagnostics/2026-10-30-other-datasets/results/sanity_baselines.txt new file mode 100644 index 0000000000000000000000000000000000000000..f4d1e61db1652932a433cf93e5e5dc642058f08d --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/sanity_baselines.txt @@ -0,0 +1,15 @@ + print(f" {name:8s} acc={np.nanmean(r[:,0]):.3f} macroF1={np.nanmean(r[:,1]):.3f} weightedF1={np.nanmean(r[:,2]):.3f} (3 seeds)") + print(f" {name:8s} acc={np.nanmean(r[:,0]):.3f} macroF1={np.nanmean(r[:,1]):.3f} weightedF1={np.nanmean(r[:,2]):.3f} (3 seeds)") + print(f" {name:8s} acc={np.nanmean(r[:,0]):.3f} macroF1={np.nanmean(r[:,1]):.3f} weightedF1={np.nanmean(r[:,2]):.3f} (3 seeds)") +COAD.pkl + majority acc=0.672 macroF1=nan weightedF1=nan (3 seeds) + logreg acc=0.505 macroF1=0.219 weightedF1=0.481 (3 seeds) + forest acc=0.630 macroF1=0.222 weightedF1=0.536 (3 seeds) +GBM.pkl + majority acc=0.301 macroF1=nan weightedF1=nan (3 seeds) + logreg acc=0.781 macroF1=0.808 weightedF1=0.780 (3 seeds) + forest acc=0.738 macroF1=0.763 weightedF1=0.736 (3 seeds) +BRCA_singlecell.pkl + majority acc=0.498 macroF1=nan weightedF1=nan (3 seeds) + logreg acc=0.746 macroF1=0.690 weightedF1=0.745 (3 seeds) + forest acc=0.763 macroF1=0.734 weightedF1=0.768 (3 seeds) diff --git a/diagnostics/2026-10-30-other-datasets/results/seeds_GBM.jsonl b/diagnostics/2026-10-30-other-datasets/results/seeds_GBM.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..c8b22f94586476031b1f3b71b94780e0816f15e5 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/seeds_GBM.jsonl @@ -0,0 +1,48 @@ +{"config": "D default (lab grid, tanh/relu/tanh, b32)", "seed": 1, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.47540983606557374, "f1_macro": 0.43307602339181284, "f1_weighted": 0.3943897996357012, "epochs_completed": 42, "pred_counts": [5, 14, 38, 4, 0], "collapsed": false, "recall": [0.13333333333333333, 0.5454545454545454, 0.8947368421052632, 0.8, 0.0]} +{"config": "D default (lab grid, tanh/relu/tanh, b32)", "seed": 2, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.4262295081967213, "f1_macro": 0.46047619047619043, "f1_weighted": 0.34648711943793914, "epochs_completed": 43, "pred_counts": [6, 21, 0, 5, 29], "collapsed": false, "recall": [0.2, 0.9166666666666666, 0.0, 1.0, 0.6363636363636364]} +{"config": "D default (lab grid, tanh/relu/tanh, b32)", "seed": 0, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.5573770491803278, "f1_macro": 0.6098390404162899, "f1_weighted": 0.5592879969261165, "epochs_completed": 62, "pred_counts": [16, 7, 22, 6, 10], "collapsed": false, "recall": [0.4, 0.5, 0.6111111111111112, 1.0, 0.5454545454545454]} +{"config": "D default (lab grid, tanh/relu/tanh, b32)", "seed": 42, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.5409836065573771, "f1_macro": 0.5952165331475676, "f1_weighted": 0.5327840302402201, "epochs_completed": 44, "pred_counts": [14, 15, 15, 6, 11], "collapsed": false, "recall": [0.4, 0.6666666666666666, 0.3888888888888889, 1.0, 0.6363636363636364]} +{"config": "D default (lab grid, tanh/relu/tanh, b32)", "seed": 123, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 123, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.6721311475409836, "f1_macro": 0.688671613156739, "f1_weighted": 0.6753076778210889, "epochs_completed": 55, "pred_counts": [23, 11, 15, 4, 8], "collapsed": false, "recall": [0.8, 0.5833333333333334, 0.7222222222222222, 0.8, 0.45454545454545453]} +{"config": "D default (lab grid, tanh/relu/tanh, b32)", "seed": 3, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.7704918032786885, "f1_macro": 0.8044513457556937, "f1_weighted": 0.7581712656552285, "epochs_completed": 102, "pred_counts": [8, 12, 24, 5, 12], "collapsed": false, "recall": [0.4, 0.9166666666666666, 0.8333333333333334, 1.0, 0.9090909090909091]} +{"config": "D default (lab grid, tanh/relu/tanh, b32)", "seed": 7, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 7, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.7049180327868853, "f1_macro": 0.7004285714285714, "f1_weighted": 0.6357962529274006, "epochs_completed": 64, "pred_counts": [1, 10, 31, 5, 14], "collapsed": false, "recall": [0.06666666666666667, 0.8181818181818182, 0.9473684210526315, 1.0, 0.9090909090909091]} +{"config": "PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03)", "seed": 1, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.3114754098360656, "f1_macro": 0.095, "f1_weighted": 0.14795081967213114, "epochs_completed": 28, "pred_counts": [0, 0, 61, 0, 0], "collapsed": true, "recall": [0.0, 0.0, 1.0, 0.0, 0.0]} +{"config": "PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03)", "seed": 2, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 24, "pred_counts": [61, 0, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0, 0.0]} +{"config": "PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03)", "seed": 0, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 18, "pred_counts": [61, 0, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0, 0.0]} +{"config": "D default (lab grid, tanh/relu/tanh, b32)", "seed": 2024, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2024, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.6557377049180327, "f1_macro": 0.6852649572649573, "f1_weighted": 0.6498080425949277, "epochs_completed": 52, "pred_counts": [11, 14, 18, 4, 14], "collapsed": false, "recall": [0.4, 0.8333333333333334, 0.6666666666666666, 0.8, 0.7272727272727273]} +{"config": "PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03)", "seed": 42, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 21, "pred_counts": [61, 0, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0, 0.0]} +{"config": "PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03)", "seed": 7, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 7, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.3114754098360656, "f1_macro": 0.095, "f1_weighted": 0.14795081967213114, "epochs_completed": 13, "pred_counts": [0, 0, 61, 0, 0], "collapsed": true, "recall": [0.0, 0.0, 1.0, 0.0, 0.0]} +{"config": "PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03)", "seed": 3, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 21, "pred_counts": [0, 0, 61, 0, 0], "collapsed": true, "recall": [0.0, 0.0, 1.0, 0.0, 0.0]} +{"config": "PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03)", "seed": 123, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 123, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 37, "pred_counts": [61, 0, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0, 0.0]} +{"config": "PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03)", "seed": 2024, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 2024, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 42, "pred_counts": [0, 0, 61, 0, 0], "collapsed": true, "recall": [0.0, 0.0, 1.0, 0.0, 0.0]} +{"config": "BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep)", "seed": 1, "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.6065573770491803, "f1_macro": 0.6519371873560633, "f1_weighted": 0.597909189420411, "epochs_completed": 100, "pred_counts": [12, 12, 22, 6, 9], "collapsed": false, "recall": [0.4, 0.9090909090909091, 0.5789473684210527, 1.0, 0.45454545454545453]} +{"config": "BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep)", "seed": 0, "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.639344262295082, "f1_macro": 0.6848047538200339, "f1_weighted": 0.6377021347657882, "epochs_completed": 98, "pred_counts": [16, 12, 20, 5, 8], "collapsed": false, "recall": [0.5333333333333333, 0.75, 0.6666666666666666, 1.0, 0.45454545454545453]} +{"config": "BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep)", "seed": 2, "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.6885245901639344, "f1_macro": 0.7357125632987702, "f1_weighted": 0.6752593553045786, "epochs_completed": 100, "pred_counts": [18, 14, 11, 5, 13], "collapsed": false, "recall": [0.6666666666666666, 1.0, 0.3888888888888889, 1.0, 0.7272727272727273]} +{"config": "BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep)", "seed": 3, "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.6229508196721312, "f1_macro": 0.6648750317715836, "f1_weighted": 0.6146956247578069, "epochs_completed": 100, "pred_counts": [10, 17, 19, 6, 9], "collapsed": false, "recall": [0.3333333333333333, 0.75, 0.6111111111111112, 1.0, 0.7272727272727273]} +{"config": "BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep)", "seed": 7, "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 7, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.5737704918032787, "f1_macro": 0.6060140897293288, "f1_weighted": 0.5766546922202569, "epochs_completed": 97, "pred_counts": [13, 13, 12, 5, 18], "collapsed": false, "recall": [0.4666666666666667, 0.7272727272727273, 0.47368421052631576, 0.8, 0.6363636363636364]} +{"config": "BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep)", "seed": 42, "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.5573770491803278, "f1_macro": 0.5982382728833033, "f1_weighted": 0.5501787143126551, "epochs_completed": 100, "pred_counts": [19, 14, 11, 6, 11], "collapsed": false, "recall": [0.6, 0.6666666666666666, 0.3888888888888889, 1.0, 0.45454545454545453]} +{"config": "BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep)", "seed": 123, "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 123, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.639344262295082, "f1_macro": 0.6691968648490387, "f1_weighted": 0.6386531999504202, "epochs_completed": 100, "pred_counts": [12, 11, 19, 6, 13], "collapsed": false, "recall": [0.5333333333333333, 0.6666666666666666, 0.6666666666666666, 1.0, 0.5454545454545454]} +{"config": "G1 grid b16 p64 tanh/relu/tanh", "seed": 0, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.639344262295082, "f1_macro": 0.6753141167775314, "f1_weighted": 0.6359577381168745, "epochs_completed": 39, "pred_counts": [13, 9, 23, 6, 10], "collapsed": false, "recall": [0.4666666666666667, 0.5833333333333334, 0.7222222222222222, 1.0, 0.6363636363636364]} +{"config": "BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep)", "seed": 2024, "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2024, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.5409836065573771, "f1_macro": 0.5938261638411564, "f1_weighted": 0.5299037973848111, "epochs_completed": 92, "pred_counts": [14, 14, 15, 6, 12], "collapsed": false, "recall": [0.3333333333333333, 0.8333333333333334, 0.3888888888888889, 1.0, 0.5454545454545454]} +{"config": "G1 grid b16 p64 tanh/relu/tanh", "seed": 2, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.5081967213114754, "f1_macro": 0.5089147733825394, "f1_weighted": 0.44799227120361634, "epochs_completed": 41, "pred_counts": [11, 17, 28, 5, 0], "collapsed": false, "recall": [0.26666666666666666, 0.9166666666666666, 0.6111111111111112, 1.0, 0.0]} +{"config": "G1 grid b16 p64 tanh/relu/tanh", "seed": 1, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.4426229508196721, "f1_macro": 0.48704381274040714, "f1_weighted": 0.3704265924739964, "epochs_completed": 67, "pred_counts": [4, 11, 0, 6, 40], "collapsed": false, "recall": [0.2, 0.8181818181818182, 0.0, 1.0, 0.9090909090909091]} +{"config": "G1 grid b16 p64 tanh/relu/tanh", "seed": 7, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 7, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.4426229508196721, "f1_macro": 0.5011815718157181, "f1_weighted": 0.38266826602692255, "epochs_completed": 62, "pred_counts": [12, 14, 0, 5, 30], "collapsed": false, "recall": [0.4, 0.8181818181818182, 0.0, 1.0, 0.6363636363636364]} +{"config": "G1 grid b16 p64 tanh/relu/tanh", "seed": 42, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.5901639344262295, "f1_macro": 0.6113227081473926, "f1_weighted": 0.5856015283546151, "epochs_completed": 36, "pred_counts": [16, 13, 14, 6, 12], "collapsed": false, "recall": [0.6, 0.6666666666666666, 0.4444444444444444, 0.8, 0.6363636363636364]} +{"config": "G2 grid b32 p64 tanh/relu/relu", "seed": 0, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.03}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 23, "pred_counts": [0, 0, 61, 0, 0], "collapsed": true, "recall": [0.0, 0.0, 1.0, 0.0, 0.0]} +{"config": "G2 grid b32 p64 tanh/relu/relu", "seed": 1, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.03}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 14, "pred_counts": [61, 0, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0, 0.0]} +{"config": "G1 grid b16 p64 tanh/relu/tanh", "seed": 123, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 123, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.32786885245901637, "f1_macro": 0.36420168067226893, "f1_weighted": 0.2766221242595399, "epochs_completed": 55, "pred_counts": [13, 20, 0, 5, 23], "collapsed": false, "recall": [0.26666666666666666, 0.6666666666666666, 0.0, 0.8, 0.36363636363636365]} +{"config": "G1 grid b16 p64 tanh/relu/tanh", "seed": 3, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.7377049180327869, "f1_macro": 0.7528246820929747, "f1_weighted": 0.717632605077627, "epochs_completed": 70, "pred_counts": [5, 14, 23, 4, 15], "collapsed": false, "recall": [0.3333333333333333, 1.0, 0.7777777777777778, 0.8, 0.9090909090909091]} +{"config": "G2 grid b32 p64 tanh/relu/relu", "seed": 7, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 7, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.03}, "accuracy": 0.2459016393442623, "f1_macro": 0.07894736842105263, "f1_weighted": 0.09706643658326145, "epochs_completed": 32, "pred_counts": [61, 0, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0, 0.0]} +{"config": "G2 grid b32 p64 tanh/relu/relu", "seed": 42, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.03}, "accuracy": 0.5573770491803278, "f1_macro": 0.6026491155902921, "f1_weighted": 0.5550033380313033, "epochs_completed": 55, "pred_counts": [15, 14, 16, 5, 11], "collapsed": false, "recall": [0.4, 0.6666666666666666, 0.4444444444444444, 0.8, 0.7272727272727273]} +{"config": "G2 grid b32 p64 tanh/relu/relu", "seed": 2, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.03}, "accuracy": 0.5737704918032787, "f1_macro": 0.5452063492063491, "f1_weighted": 0.5140775435857402, "epochs_completed": 63, "pred_counts": [10, 16, 31, 4, 0], "collapsed": false, "recall": [0.4, 0.9166666666666666, 0.7777777777777778, 0.8, 0.0]} +{"config": "G3 grid b16 p128 tanh/tanh/tanh", "seed": 0, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.6229508196721312, "f1_macro": 0.6716450216450217, "f1_weighted": 0.6048186785891704, "epochs_completed": 44, "pred_counts": [6, 10, 30, 5, 10], "collapsed": false, "recall": [0.2, 0.6666666666666666, 0.7777777777777778, 1.0, 0.7272727272727273]} +{"config": "G2 grid b32 p64 tanh/relu/relu", "seed": 123, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 123, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.03}, "accuracy": 0.3770491803278688, "f1_macro": 0.42608695652173917, "f1_weighted": 0.368496079828938, "epochs_completed": 65, "pred_counts": [5, 11, 28, 5, 12], "collapsed": false, "recall": [0.13333333333333333, 0.5833333333333334, 0.4444444444444444, 0.8, 0.18181818181818182]} +{"config": "G1 grid b16 p64 tanh/relu/tanh", "seed": 2024, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2024, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.7213114754098361, "f1_macro": 0.7524021076962254, "f1_weighted": 0.7176338287235107, "epochs_completed": 74, "pred_counts": [11, 9, 16, 6, 19], "collapsed": false, "recall": [0.5333333333333333, 0.75, 0.6111111111111112, 1.0, 1.0]} +{"config": "G2 grid b32 p64 tanh/relu/relu", "seed": 3, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.03}, "accuracy": 0.29508196721311475, "f1_macro": 0.09113924050632911, "f1_weighted": 0.13446773189458391, "epochs_completed": 18, "pred_counts": [0, 0, 61, 0, 0], "collapsed": true, "recall": [0.0, 0.0, 1.0, 0.0, 0.0]} +{"config": "G3 grid b16 p128 tanh/tanh/tanh", "seed": 7, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 7, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.6065573770491803, "f1_macro": 0.6433397129186603, "f1_weighted": 0.5959173268491645, "epochs_completed": 39, "pred_counts": [10, 13, 25, 5, 8], "collapsed": false, "recall": [0.4666666666666667, 0.8181818181818182, 0.6842105263157895, 1.0, 0.2727272727272727]} +{"config": "G2 grid b32 p64 tanh/relu/relu", "seed": 2024, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2024, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "relu", "sgd_momentum": 0.03}, "accuracy": 0.45901639344262296, "f1_macro": 0.44644403215831785, "f1_weighted": 0.3911767389519146, "epochs_completed": 58, "pred_counts": [9, 15, 31, 6, 0], "collapsed": false, "recall": [0.13333333333333333, 0.75, 0.6666666666666666, 1.0, 0.0]} +{"config": "G3 grid b16 p128 tanh/tanh/tanh", "seed": 2, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.5573770491803278, "f1_macro": 0.5318747010999523, "f1_weighted": 0.5044374407099905, "epochs_completed": 58, "pred_counts": [19, 12, 23, 7, 0], "collapsed": false, "recall": [0.6666666666666666, 0.75, 0.5555555555555556, 1.0, 0.0]} +{"config": "G3 grid b16 p128 tanh/tanh/tanh", "seed": 1, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.7213114754098361, "f1_macro": 0.7603030303030303, "f1_weighted": 0.7188169753743524, "epochs_completed": 98, "pred_counts": [13, 11, 21, 6, 10], "collapsed": false, "recall": [0.5333333333333333, 0.9090909090909091, 0.6842105263157895, 1.0, 0.7272727272727273]} +{"config": "G3 grid b16 p128 tanh/tanh/tanh", "seed": 42, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.5901639344262295, "f1_macro": 0.5929791912732021, "f1_weighted": 0.5731392264342787, "epochs_completed": 78, "pred_counts": [14, 15, 17, 7, 8], "collapsed": false, "recall": [0.5333333333333333, 0.8333333333333334, 0.6111111111111112, 1.0, 0.18181818181818182]} +{"config": "G3 grid b16 p128 tanh/tanh/tanh", "seed": 3, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.7868852459016393, "f1_macro": 0.7962876254180602, "f1_weighted": 0.7849936948297604, "epochs_completed": 92, "pred_counts": [11, 11, 22, 5, 12], "collapsed": false, "recall": [0.6, 0.8333333333333334, 0.8333333333333334, 0.8, 0.9090909090909091]} +{"config": "G3 grid b16 p128 tanh/tanh/tanh", "seed": 123, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 123, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.7213114754098361, "f1_macro": 0.7444357298474946, "f1_weighted": 0.7206428801028608, "epochs_completed": 84, "pred_counts": [12, 15, 16, 4, 14], "collapsed": false, "recall": [0.6, 0.8333333333333334, 0.6666666666666666, 0.8, 0.8181818181818182]} +{"config": "G3 grid b16 p128 tanh/tanh/tanh", "seed": 2024, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2024, "biological_activation": "tanh", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.6557377049180327, "f1_macro": 0.704962439521263, "f1_weighted": 0.6595193218867279, "epochs_completed": 55, "pred_counts": [19, 9, 14, 6, 13], "collapsed": false, "recall": [0.6, 0.75, 0.5, 1.0, 0.7272727272727273]} diff --git a/diagnostics/2026-10-30-other-datasets/results/seeds_GBM_shard0.log b/diagnostics/2026-10-30-other-datasets/results/seeds_GBM_shard0.log new file mode 100644 index 0000000000000000000000000000000000000000..22600a239649497398f4d48cdcdd0088ca791013 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/seeds_GBM_shard0.log @@ -0,0 +1,13 @@ +[0] D default (lab grid, tanh/relu/tanh, b32) seed=0 f1m=0.610 +[0] D default (lab grid, tanh/relu/tanh, b32) seed=7 f1m=0.700 +[0] PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03) seed=0 f1m=0.079 +[0] PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03) seed=7 f1m=0.095 +[0] BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep) seed=0 f1m=0.685 +[0] BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep) seed=7 f1m=0.606 +[0] G1 grid b16 p64 tanh/relu/tanh seed=0 f1m=0.675 +[0] G1 grid b16 p64 tanh/relu/tanh seed=7 f1m=0.501 +[0] G2 grid b32 p64 tanh/relu/relu seed=0 f1m=0.091 +[0] G2 grid b32 p64 tanh/relu/relu seed=7 f1m=0.079 +[0] G3 grid b16 p128 tanh/tanh/tanh seed=0 f1m=0.672 +[0] G3 grid b16 p128 tanh/tanh/tanh seed=7 f1m=0.643 +SHARD 0 DONE diff --git a/diagnostics/2026-10-30-other-datasets/results/seeds_GBM_shard1.log b/diagnostics/2026-10-30-other-datasets/results/seeds_GBM_shard1.log new file mode 100644 index 0000000000000000000000000000000000000000..82aebd44674f0e1a8b9556a4cdd7538e22465474 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/seeds_GBM_shard1.log @@ -0,0 +1,13 @@ +[1] D default (lab grid, tanh/relu/tanh, b32) seed=1 f1m=0.433 +[1] D default (lab grid, tanh/relu/tanh, b32) seed=42 f1m=0.595 +[1] PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03) seed=1 f1m=0.095 +[1] PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03) seed=42 f1m=0.079 +[1] BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep) seed=1 f1m=0.652 +[1] BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep) seed=42 f1m=0.598 +[1] G1 grid b16 p64 tanh/relu/tanh seed=1 f1m=0.487 +[1] G1 grid b16 p64 tanh/relu/tanh seed=42 f1m=0.611 +[1] G2 grid b32 p64 tanh/relu/relu seed=1 f1m=0.079 +[1] G2 grid b32 p64 tanh/relu/relu seed=42 f1m=0.603 +[1] G3 grid b16 p128 tanh/tanh/tanh seed=1 f1m=0.760 +[1] G3 grid b16 p128 tanh/tanh/tanh seed=42 f1m=0.593 +SHARD 1 DONE diff --git a/diagnostics/2026-10-30-other-datasets/results/seeds_GBM_shard2.log b/diagnostics/2026-10-30-other-datasets/results/seeds_GBM_shard2.log new file mode 100644 index 0000000000000000000000000000000000000000..1a728253cf6c3aa74f64c558347f910e2d3fe078 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/seeds_GBM_shard2.log @@ -0,0 +1,13 @@ +[2] D default (lab grid, tanh/relu/tanh, b32) seed=2 f1m=0.460 +[2] D default (lab grid, tanh/relu/tanh, b32) seed=123 f1m=0.689 +[2] PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03) seed=2 f1m=0.079 +[2] PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03) seed=123 f1m=0.079 +[2] BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep) seed=2 f1m=0.736 +[2] BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep) seed=123 f1m=0.669 +[2] G1 grid b16 p64 tanh/relu/tanh seed=2 f1m=0.509 +[2] G1 grid b16 p64 tanh/relu/tanh seed=123 f1m=0.364 +[2] G2 grid b32 p64 tanh/relu/relu seed=2 f1m=0.545 +[2] G2 grid b32 p64 tanh/relu/relu seed=123 f1m=0.426 +[2] G3 grid b16 p128 tanh/tanh/tanh seed=2 f1m=0.532 +[2] G3 grid b16 p128 tanh/tanh/tanh seed=123 f1m=0.744 +SHARD 2 DONE diff --git a/diagnostics/2026-10-30-other-datasets/results/seeds_GBM_shard3.log b/diagnostics/2026-10-30-other-datasets/results/seeds_GBM_shard3.log new file mode 100644 index 0000000000000000000000000000000000000000..ed8054f9b35320da6c02d1543a8d12b048d9e870 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/seeds_GBM_shard3.log @@ -0,0 +1,13 @@ +[3] D default (lab grid, tanh/relu/tanh, b32) seed=3 f1m=0.804 +[3] D default (lab grid, tanh/relu/tanh, b32) seed=2024 f1m=0.685 +[3] PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03) seed=3 f1m=0.091 +[3] PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03) seed=2024 f1m=0.091 +[3] BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep) seed=3 f1m=0.665 +[3] BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep) seed=2024 f1m=0.594 +[3] G1 grid b16 p64 tanh/relu/tanh seed=3 f1m=0.753 +[3] G1 grid b16 p64 tanh/relu/tanh seed=2024 f1m=0.752 +[3] G2 grid b32 p64 tanh/relu/relu seed=3 f1m=0.091 +[3] G2 grid b32 p64 tanh/relu/relu seed=2024 f1m=0.446 +[3] G3 grid b16 p128 tanh/tanh/tanh seed=3 f1m=0.796 +[3] G3 grid b16 p128 tanh/tanh/tanh seed=2024 f1m=0.705 +SHARD 3 DONE diff --git a/diagnostics/2026-10-30-other-datasets/results/seeds_GBM_summary.txt b/diagnostics/2026-10-30-other-datasets/results/seeds_GBM_summary.txt new file mode 100644 index 0000000000000000000000000000000000000000..9480b6e084f3b86b6ccb9da507eb2b79132db89a --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/seeds_GBM_summary.txt @@ -0,0 +1,8 @@ + accuracy macro F1 weighted F1 macro F1 min/max collapsed worst-class recall +config +G3 grid b16 p128 tanh/tanh/tanh 0.658 ± 0.079 0.681 ± 0.089 0.645 ± 0.093 0.532 / 0.796 0 0.36 +BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep) 0.609 ± 0.049 0.651 ± 0.049 0.603 ± 0.049 0.594 / 0.736 0 0.41 +D default (lab grid, tanh/relu/tanh, b32) 0.600 ± 0.119 0.622 ± 0.126 0.569 ± 0.141 0.433 / 0.804 0 0.26 +G1 grid b16 p64 tanh/relu/tanh 0.551 ± 0.146 0.582 ± 0.139 0.517 ± 0.170 0.364 / 0.753 0 0.22 +G2 grid b32 p64 tanh/relu/relu 0.381 ± 0.134 0.295 ± 0.231 0.286 ± 0.193 0.079 / 0.603 4 0.07 +PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03) 0.275 ± 0.031 0.086 ± 0.008 0.119 ± 0.024 0.079 / 0.095 8 0.00 diff --git a/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2.jsonl b/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..6d297af9a84f471103c7303c3c8b971a2a043b95 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2.jsonl @@ -0,0 +1,19 @@ +{"config": "D default (lab grid, tanh/relu/tanh, b32)", "seed": 3, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.6928104575163399, "f1_macro": 0.7207951274123041, "f1_weighted": 0.6888702127615093, "epochs_completed": 42, "pred_counts": [45, 41, 20, 47], "collapsed": false, "recall": [0.631578947368421, 0.9655172413793104, 0.85, 0.5606060606060606]} +{"config": "D default (lab grid, tanh/relu/tanh, b32)", "seed": 2, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.8300653594771242, "f1_macro": 0.8439550045793448, "f1_weighted": 0.8307953025557261, "epochs_completed": 48, "pred_counts": [33, 40, 18, 62], "collapsed": false, "recall": [0.7368421052631579, 0.9655172413793104, 0.9, 0.803030303030303]} +{"config": "D default (lab grid, tanh/relu/tanh, b32)", "seed": 0, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.8169934640522876, "f1_macro": 0.8298007798103784, "f1_weighted": 0.8170906328057523, "epochs_completed": 66, "pred_counts": [30, 42, 21, 60], "collapsed": false, "recall": [0.7105263157894737, 0.9655172413793104, 0.95, 0.7727272727272727]} +{"config": "D default (lab grid, tanh/relu/tanh, b32)", "seed": 2024, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2024, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.5686274509803921, "f1_macro": 0.6142857142857143, "f1_weighted": 0.5486150015561779, "epochs_completed": 32, "pred_counts": [46, 43, 25, 39], "collapsed": false, "recall": [0.5, 0.896551724137931, 0.95, 0.3484848484848485]} +{"config": "D default (lab grid, tanh/relu/tanh, b32)", "seed": 123, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 123, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.7843137254901961, "f1_macro": 0.7930667179662562, "f1_weighted": 0.7778169882484077, "epochs_completed": 32, "pred_counts": [25, 35, 23, 70], "collapsed": false, "recall": [0.5526315789473685, 0.8620689655172413, 1.0, 0.8181818181818182]} +{"config": "D default (lab grid, tanh/relu/tanh, b32)", "seed": 1, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.8235294117647058, "f1_macro": 0.8218085785164465, "f1_weighted": 0.8214747925563567, "epochs_completed": 84, "pred_counts": [30, 38, 23, 62], "collapsed": false, "recall": [0.631578947368421, 0.896551724137931, 1.0, 0.8484848484848485]} +{"config": "PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03)", "seed": 3, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 13, "pred_counts": [0, 0, 0, 153], "collapsed": true, "recall": [0.0, 0.0, 0.0, 1.0]} +{"config": "PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03)", "seed": 2, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 23, "pred_counts": [0, 0, 0, 153], "collapsed": true, "recall": [0.0, 0.0, 0.0, 1.0]} +{"config": "PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03)", "seed": 2024, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 2024, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 26, "pred_counts": [0, 0, 0, 153], "collapsed": true, "recall": [0.0, 0.0, 0.0, 1.0]} +{"config": "D default (lab grid, tanh/relu/tanh, b32)", "seed": 7, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 7, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.8562091503267973, "f1_macro": 0.8613644046523303, "f1_weighted": 0.8590313081247338, "epochs_completed": 52, "pred_counts": [38, 39, 17, 59], "collapsed": false, "recall": [0.868421052631579, 0.9310344827586207, 0.85, 0.8181818181818182]} +{"config": "PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03)", "seed": 123, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 123, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 27, "pred_counts": [0, 0, 0, 153], "collapsed": true, "recall": [0.0, 0.0, 0.0, 1.0]} +{"config": "PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03)", "seed": 0, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 15, "pred_counts": [0, 0, 0, 153], "collapsed": true, "recall": [0.0, 0.0, 0.0, 1.0]} +{"config": "PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03)", "seed": 7, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 7, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 15, "pred_counts": [0, 0, 0, 153], "collapsed": true, "recall": [0.0, 0.0, 0.0, 1.0]} +{"config": "D default (lab grid, tanh/relu/tanh, b32)", "seed": 42, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.8300653594771242, "f1_macro": 0.8389094286870471, "f1_weighted": 0.8280242914220465, "epochs_completed": 87, "pred_counts": [30, 36, 21, 66], "collapsed": false, "recall": [0.6842105263157895, 0.9310344827586207, 0.95, 0.8333333333333334]} +{"config": "PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03)", "seed": 1, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 22, "pred_counts": [0, 0, 0, 153], "collapsed": true, "recall": [0.0, 0.0, 0.0, 1.0]} +{"config": "PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03)", "seed": 42, "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.43137254901960786, "f1_macro": 0.1506849315068493, "f1_weighted": 0.26000537201181845, "epochs_completed": 28, "pred_counts": [0, 0, 0, 153], "collapsed": true, "recall": [0.0, 0.0, 0.0, 1.0]} +{"config": "BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep)", "seed": 2, "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.8300653594771242, "f1_macro": 0.8342136222910217, "f1_weighted": 0.8319748274954977, "epochs_completed": 67, "pred_counts": [37, 39, 18, 59], "collapsed": false, "recall": [0.7894736842105263, 0.9310344827586207, 0.85, 0.803030303030303]} +{"config": "BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep)", "seed": 0, "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.803921568627451, "f1_macro": 0.809327868852459, "f1_weighted": 0.8050590378227794, "epochs_completed": 68, "pred_counts": [42, 32, 20, 59], "collapsed": false, "recall": [0.8421052631578947, 0.7586206896551724, 0.9, 0.7727272727272727]} +{"config": "BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep)", "seed": 3, "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.8562091503267973, "f1_macro": 0.85758547008547, "f1_weighted": 0.8586168370482097, "epochs_completed": 100, "pred_counts": [34, 36, 19, 64], "collapsed": false, "recall": [0.8157894736842105, 0.8620689655172413, 0.9, 0.8636363636363636]} diff --git a/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2_common_shard0.log b/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2_common_shard0.log new file mode 100644 index 0000000000000000000000000000000000000000..55005ddf6f6726babfb5267c973429f05e07be77 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2_common_shard0.log @@ -0,0 +1,5 @@ +[0] D default (lab grid, tanh/relu/tanh, b32) seed=0 f1m=0.830 +[0] D default (lab grid, tanh/relu/tanh, b32) seed=7 f1m=0.861 +[0] PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03) seed=0 f1m=0.151 +[0] PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03) seed=7 f1m=0.151 +[0] BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep) seed=0 f1m=0.809 diff --git a/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2_common_shard1.log b/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2_common_shard1.log new file mode 100644 index 0000000000000000000000000000000000000000..7419136d06a9c043bab271fdbaf559c9b98a9ffe --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2_common_shard1.log @@ -0,0 +1,4 @@ +[1] D default (lab grid, tanh/relu/tanh, b32) seed=1 f1m=0.822 +[1] D default (lab grid, tanh/relu/tanh, b32) seed=42 f1m=0.839 +[1] PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03) seed=1 f1m=0.151 +[1] PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03) seed=42 f1m=0.151 diff --git a/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2_common_shard2.log b/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2_common_shard2.log new file mode 100644 index 0000000000000000000000000000000000000000..758fd90997537badcc3ccf797642d8e60aee2c01 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2_common_shard2.log @@ -0,0 +1,5 @@ +[2] D default (lab grid, tanh/relu/tanh, b32) seed=2 f1m=0.844 +[2] D default (lab grid, tanh/relu/tanh, b32) seed=123 f1m=0.793 +[2] PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03) seed=2 f1m=0.151 +[2] PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03) seed=123 f1m=0.151 +[2] BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep) seed=2 f1m=0.834 diff --git a/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2_common_shard3.log b/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2_common_shard3.log new file mode 100644 index 0000000000000000000000000000000000000000..19a3496c7e888c69ba9b0de58bdbbd6ed57dde2a --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2_common_shard3.log @@ -0,0 +1,5 @@ +[3] D default (lab grid, tanh/relu/tanh, b32) seed=3 f1m=0.721 +[3] D default (lab grid, tanh/relu/tanh, b32) seed=2024 f1m=0.614 +[3] PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03) seed=3 f1m=0.151 +[3] PAPER notebook (SGD lr0.01 L2 0.01 b16 relu/sigmoid/tanh, mom 0.03) seed=2024 f1m=0.151 +[3] BASE (Adam lr1e-3 wd1e-4 b16 relu/sigmoid/tanh, 100 ep) seed=3 f1m=0.858 diff --git a/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2_top.jsonl b/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2_top.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..2763a388e75962cb5b5ed4bdeedec527bbe2afd0 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2_top.jsonl @@ -0,0 +1,2 @@ +{"config": "S1 grid b32 p128 relu/tanh/tanh", "seed": 1, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.7189542483660131, "f1_macro": 0.7299061528679224, "f1_weighted": 0.7073511931624367, "epochs_completed": 34, "pred_counts": [23, 45, 20, 65], "collapsed": false, "recall": [0.39473684210526316, 0.9310344827586207, 0.95, 0.7424242424242424]} +{"config": "S1 grid b32 p128 relu/tanh/tanh", "seed": 0, "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 128, "fusion_dim": 12, "validation_fraction": 0.25, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "relu", "projection_activation": "tanh", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "accuracy": 0.7254901960784313, "f1_macro": 0.7439041014698909, "f1_weighted": 0.7286905326379011, "epochs_completed": 52, "pred_counts": [36, 51, 18, 48], "collapsed": false, "recall": [0.7368421052631579, 0.9310344827586207, 0.8, 0.6060606060606061]} diff --git a/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2_top_shard0.log b/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2_top_shard0.log new file mode 100644 index 0000000000000000000000000000000000000000..8541bc967727557d41f9ddd30cd17a87a3dc1c4f --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2_top_shard0.log @@ -0,0 +1 @@ +[0] S1 grid b32 p128 relu/tanh/tanh seed=0 f1m=0.744 diff --git a/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2_top_shard1.log b/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2_top_shard1.log new file mode 100644 index 0000000000000000000000000000000000000000..3bdce11ee4e49b669ec1fffba6d2e64ea9f51b14 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/results/seeds_scTrioseq2_top_shard1.log @@ -0,0 +1 @@ +[1] S1 grid b32 p128 relu/tanh/tanh seed=1 f1m=0.730 diff --git a/diagnostics/2026-10-30-other-datasets/scripts/build_workspace.py b/diagnostics/2026-10-30-other-datasets/scripts/build_workspace.py new file mode 100644 index 0000000000000000000000000000000000000000..4d2e12eb051a9a223e4ef808794dd23e78c50252 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/scripts/build_workspace.py @@ -0,0 +1,26 @@ +"""Build and cache one example dataset's workspace with the app's own prepare step (auto GenePT). +Usage: build_workspace.py """ +import pickle +import sys +from pathlib import Path + +import numpy as np + +import app + +DATASET, CACHE = sys.argv[1], Path(sys.argv[2]) +CACHE.mkdir(parents=True, exist_ok=True) +result = app.prepare_workspace( + "BioLM-NET examples", DATASET, "", None, None, None, None, None, None, None, + True, list(app.GENEPT_OPTIONS)[0], progress=lambda *a, **k: None, +) +workspace = result[0] +if workspace is None: + sys.exit(f"{DATASET}: {result[1]}") +(CACHE / f"workspace_{DATASET}.pkl").write_bytes(pickle.dumps(workspace)) +counts = np.bincount(workspace.labels) +print(f"{DATASET}: samples={len(workspace.labels)} classes={workspace.label_names} counts={counts.tolist()} " + f"majority share={counts.max() / counts.sum():.3f} warnings={workspace.warnings}") +for name, branch in (("gene", workspace.gene_branch), ("dna", workspace.dna_branch)): + m = branch.biological_mask + print(f" {name}: inputs={m.shape[0]} hidden={m.shape[1]} edges={(m > 0).sum()} pathways={len(branch.pathways)}") diff --git a/diagnostics/2026-10-30-other-datasets/scripts/grid.py b/diagnostics/2026-10-30-other-datasets/scripts/grid.py new file mode 100644 index 0000000000000000000000000000000000000000..8dd693f02c0c548731c44e886517c3a7eebdf5a1 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/scripts/grid.py @@ -0,0 +1,71 @@ +"""Jubair's hyperparameter grid (email of 2026-10-17) on BRCA, using the Workbench's train() unchanged. + +Usage: grid.py [seed] +Resumable: configs already in are skipped. + +Grid as sent, mapped onto Workbench Hyperparameters: + layer_one -> projection_dim; activation_function_one/two/three -> biological/projection/fusion activation; + l2_reg -> weight_decay; epoch -> epochs (early stopping, patience 12, still applies); + SGD gets momentum 0.03 as instructed. num_ex_neuron is not supported by the Workbench (fixed at 0). +""" +import itertools +import json +import sys +from dataclasses import asdict, replace +from pathlib import Path + +import numpy as np +import pickle +import torch +from sklearn.metrics import recall_score + +import biolmnet.training as training + +WORKSPACE, OUT, SHARD, SHARDS = Path(sys.argv[1]), Path(sys.argv[2]), int(sys.argv[3]), int(sys.argv[4]) +SEED = int(sys.argv[5]) if len(sys.argv) > 5 else 42 +torch.set_num_threads(int(__import__("os").environ.get("THREADS", "2"))) + +ACT = __import__("os").environ.get("ACTS", "relu,sigmoid,tanh").split(",") +OPTIMIZERS = ["Adam"] if __import__("os").environ.get("ADAM_ONLY") else ["SGD", "Adam"] +GRID = list(itertools.product([64, 128], ACT, ACT, ACT, [16, 32], OPTIMIZERS)) + +workspace = pickle.loads(WORKSPACE.read_bytes()) +names = workspace.label_names +done = set() +if OUT.exists(): + done = {json.dumps(json.loads(line)["key"]) for line in OUT.read_text().splitlines() if line.strip()} + +for index, (proj, a1, a2, a3, batch, opt) in enumerate(GRID): + if index % SHARDS != SHARD: + continue + key = {"projection_dim": proj, "act1": a1, "act2": a2, "act3": a3, "batch_size": batch, "optimizer": opt, "seed": SEED} + if json.dumps(key) in done: + continue + hp = replace( + training.Hyperparameters(), + epochs=200, batch_size=batch, learning_rate=0.01, weight_decay=0.01, dropout=0.3, + projection_dim=proj, optimizer=opt, sgd_momentum=0.03 if opt == "SGD" else 0.0, + biological_activation=a1, projection_activation=a2, fusion_activation=a3, seed=SEED, + ) + try: + result = training.train(workspace, hp) + except Exception as exc: # record and keep going + row = {"key": key, "error": repr(exc)} + else: + m = result.bundle.metrics + vp = result.validation_predictions + counts = vp["predicted"].value_counts().reindex(names, fill_value=0).astype(int).tolist() + history = result.bundle.history + best_epoch = int(min(history, key=lambda h: h["validation_loss"])["epoch"]) + row = { + "key": key, "hyperparameters": asdict(hp), + "accuracy": m["accuracy"], "f1_macro": m["f1_macro"], "f1_weighted": m["f1_weighted"], + "epochs_completed": m["epochs_completed"], "best_epoch": best_epoch, + "pred_counts": counts, "classes_predicted": int((np.array(counts) > 0).sum()), + "recall": recall_score(vp["observed"], vp["predicted"], labels=names, average=None, zero_division=0).tolist(), + "final_train_loss": history[-1]["training_loss"], "best_val_loss": m["validation_loss"], + } + with OUT.open("a") as handle: + handle.write(json.dumps(row) + "\n") + print(f"[{SHARD}] {index} {key} -> " + (row.get("error") or f"f1m={row['f1_macro']:.3f} classes={row['classes_predicted']}"), flush=True) +print(f"SHARD {SHARD} DONE", flush=True) diff --git a/diagnostics/2026-10-30-other-datasets/scripts/grid_summary.py b/diagnostics/2026-10-30-other-datasets/scripts/grid_summary.py new file mode 100644 index 0000000000000000000000000000000000000000..0518050654744c7dac8cb6196a26a2faefce48bc --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/scripts/grid_summary.py @@ -0,0 +1,31 @@ +"""Summarise grid.py output. Usage: grid_summary.py """ +import json +import sys + +import pandas as pd + +rows = [] +for line in open(sys.argv[1]): + r = json.loads(line) + if "error" in r: + print("ERROR", r["key"], r["error"]) + continue + rows.append({**r["key"], "acc": r["accuracy"], "f1_macro": r["f1_macro"], "f1_w": r["f1_weighted"], + "epochs": r["epochs_completed"], "best": r["best_epoch"], "classes": r["classes_predicted"], + "collapsed": r["classes_predicted"] == 1, "pred": "/".join(map(str, r["pred_counts"])), + "min_recall": min(r["recall"])}) +df = pd.DataFrame(rows) +df["acts"] = df.act1 + "/" + df.act2 + "/" + df.act3 +pd.set_option("display.width", 220) +pd.set_option("display.max_rows", 300) + +print(f"configs: {len(df)} collapsed (1 class): {df.collapsed.sum()} " + f"all 5 classes predicted: {(df.classes == 5).sum()} macro F1 >= 0.70: {(df.f1_macro >= 0.70).sum()}") +print("\nTOP 20 by macro F1") +print(df.sort_values("f1_macro", ascending=False).head(20)[ + ["optimizer", "batch_size", "projection_dim", "acts", "acc", "f1_macro", "f1_w", "min_recall", "best", "epochs", "pred"] +].round(3).to_string(index=False)) +for factor in ["optimizer", "batch_size", "projection_dim", "act1", "act2", "act3"]: + g = df.groupby(factor).agg(n=("f1_macro", "size"), f1_mean=("f1_macro", "mean"), f1_max=("f1_macro", "max"), + collapsed=("collapsed", "sum"), working=("f1_macro", lambda s: int((s >= 0.70).sum()))) + print(f"\nBY {factor}\n" + g.round(3).to_string()) diff --git a/diagnostics/2026-10-30-other-datasets/scripts/sanity_baselines.py b/diagnostics/2026-10-30-other-datasets/scripts/sanity_baselines.py new file mode 100644 index 0000000000000000000000000000000000000000..d9100780a3e7eb2c75febccdd7fa381282d29e7e --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/scripts/sanity_baselines.py @@ -0,0 +1,32 @@ +"""Label-alignment sanity check: classical models on the Workbench's own split and scaling. +If simple models beat the majority-class rate clearly, labels carry signal for these rows. +Usage: sanity_baselines.py [ ...]""" +import pickle +import sys + +import numpy as np +from sklearn.ensemble import RandomForestClassifier +from sklearn.linear_model import LogisticRegression +from sklearn.metrics import accuracy_score, f1_score + +from biolmnet.training import split_train_validation + +for path in sys.argv[1:]: + ws = pickle.load(open(path, "rb")) + X = np.hstack([ws.gene_expression, ws.dna_methylation]).astype(np.float64) + y = ws.labels + rows = [] + for seed in (0, 1, 42): + tr, va = split_train_validation(y, 0.25, seed) + mean, std = X[tr].mean(0), X[tr].std(0) + 1e-8 + Xtr, Xva = (X[tr] - mean) / std, (X[va] - mean) / std + majority = np.bincount(y[tr]).argmax() + for name, model in (("logreg", LogisticRegression(max_iter=3000, class_weight="balanced", C=0.1)), + ("forest", RandomForestClassifier(500, class_weight="balanced", random_state=seed, n_jobs=2))): + p = model.fit(Xtr, y[tr]).predict(Xva) + rows.append((name, accuracy_score(y[va], p), f1_score(y[va], p, average="macro"), f1_score(y[va], p, average="weighted"))) + rows.append(("majority", accuracy_score(y[va], np.full(len(va), majority)), np.nan, np.nan)) + print(path.split("workspace_")[-1]) + for name in ("majority", "logreg", "forest"): + r = np.array([x[1:] for x in rows if x[0] == name], dtype=float) + print(f" {name:8s} acc={np.nanmean(r[:,0]):.3f} macroF1={np.nanmean(r[:,1]):.3f} weightedF1={np.nanmean(r[:,2]):.3f} (3 seeds)") diff --git a/diagnostics/2026-10-30-other-datasets/scripts/seed_check.py b/diagnostics/2026-10-30-other-datasets/scripts/seed_check.py new file mode 100644 index 0000000000000000000000000000000000000000..f44bacac6b52930d1de325208b96a4fbe2646811 --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/scripts/seed_check.py @@ -0,0 +1,51 @@ +"""Run named configurations across 8 seeds on one workspace. +Usage: seed_check.py +configs.json maps a name to Hyperparameters overrides. Resumable: (config, seed) pairs already in out.jsonl are skipped.""" +import json +import os +import pickle +import sys +from dataclasses import asdict, replace +from pathlib import Path + +import numpy as np +import torch +from sklearn.metrics import recall_score + +import biolmnet.training as training + +WORKSPACE, CONFIGS, OUT = Path(sys.argv[1]), json.loads(Path(sys.argv[2]).read_text()), Path(sys.argv[3]) +SHARD, SHARDS = int(sys.argv[4]), int(sys.argv[5]) +torch.set_num_threads(int(os.environ.get("THREADS", "2"))) +SEEDS = [0, 1, 2, 3, 7, 42, 123, 2024] + +workspace = pickle.loads(WORKSPACE.read_bytes()) +names = workspace.label_names +done = set() +if OUT.exists(): + done = {(r["config"], r["seed"]) for r in map(json.loads, OUT.read_text().splitlines()) if r} + +jobs = [(c, s) for c in CONFIGS for s in SEEDS] +for index, (config, seed) in enumerate(jobs): + if index % SHARDS != SHARD or (config, seed) in done: + continue + hp = replace(training.Hyperparameters(), seed=seed, **CONFIGS[config]) + try: + result = training.train(workspace, hp) + except Exception as exc: + row = {"config": config, "seed": seed, "error": repr(exc)} + else: + m = result.bundle.metrics + vp = result.validation_predictions + counts = vp["predicted"].value_counts().reindex(names, fill_value=0).astype(int).tolist() + row = { + "config": config, "seed": seed, "hyperparameters": asdict(hp), + "accuracy": m["accuracy"], "f1_macro": m["f1_macro"], "f1_weighted": m["f1_weighted"], + "epochs_completed": m["epochs_completed"], + "pred_counts": counts, "collapsed": int((np.array(counts) > 0).sum()) == 1, + "recall": recall_score(vp["observed"], vp["predicted"], labels=names, average=None, zero_division=0).tolist(), + } + with OUT.open("a") as handle: + handle.write(json.dumps(row) + "\n") + print(f"[{SHARD}] {config} seed={seed} " + (row.get("error") or f"f1m={row['f1_macro']:.3f}"), flush=True) +print(f"SHARD {SHARD} DONE", flush=True) diff --git a/diagnostics/2026-10-30-other-datasets/scripts/seed_summary.py b/diagnostics/2026-10-30-other-datasets/scripts/seed_summary.py new file mode 100644 index 0000000000000000000000000000000000000000..e2cb8014823502e99e0c9f9e57ecd9a961572bbe --- /dev/null +++ b/diagnostics/2026-10-30-other-datasets/scripts/seed_summary.py @@ -0,0 +1,23 @@ +"""Summarise seed_check.py output. Usage: seed_summary.py """ +import json +import sys + +import pandas as pd + +rows = [json.loads(l) for l in open(sys.argv[1])] +errors = [r for r in rows if "error" in r] +df = pd.DataFrame([r for r in rows if "error" not in r]) +df["min_recall"] = df.recall.map(min) +g = df.groupby("config") +s = pd.DataFrame({ + "accuracy": g.accuracy.mean().map("{:.3f}".format) + " ± " + g.accuracy.std().map("{:.3f}".format), + "macro F1": g.f1_macro.mean().map("{:.3f}".format) + " ± " + g.f1_macro.std().map("{:.3f}".format), + "weighted F1": g.f1_weighted.mean().map("{:.3f}".format) + " ± " + g.f1_weighted.std().map("{:.3f}".format), + "macro F1 min/max": g.f1_macro.min().map("{:.3f}".format) + " / " + g.f1_macro.max().map("{:.3f}".format), + "collapsed": g.collapsed.sum(), "worst-class recall": g.min_recall.mean().round(2), + "_sort": g.f1_macro.mean(), +}).sort_values("_sort", ascending=False).drop(columns="_sort") +pd.set_option("display.width", 250); pd.set_option("display.max_colwidth", 70) +print(s.to_string()) +for r in errors: + print("ERROR", r["config"], r["seed"], r["error"]) diff --git a/diagnostics/2026-11-06-test-split-reproduction/README.md b/diagnostics/2026-11-06-test-split-reproduction/README.md new file mode 100644 index 0000000000000000000000000000000000000000..ce98b42ca9157fb043720b1d03596c33b4439bb4 --- /dev/null +++ b/diagnostics/2026-11-06-test-split-reproduction/README.md @@ -0,0 +1,36 @@ +# Paper reproduction on the 80/20 test split (2026-11-06) + +All scores are on the untouched 20% test set, averaged over 10 seeds (the paper reports 10 runs). Early stopping uses 1/8 of the training part. The variants marked "notebook first layer" swap in, for these runs only, the upstream notebook's first layer: ReLU(xW + b) followed by a fixed multiplication by the biological mask. In the Workbench, the mask is applied to the trainable weights (W ⊙ M) instead. `biolmnet/model.py` was not changed. + +Configurations: + +- **D:** the current Workbench default, from the lab grid. Adam, lr 0.01, L2 0.01, batch 32, tanh/relu/tanh activations. +- **BASE:** Adam, lr 0.001, weight decay 1e-4, batch 16, published relu/sigmoid/tanh activations, 100 epochs. +- **Notebook replica:** the notebook's first layer with the paper notebook's SGD settings (lr 0.01, L2 0.01, batch 16, relu/sigmoid/tanh, 200 epochs, momentum 0.03). + +| Dataset | Paper | D | BASE | BASE + notebook first layer | +|---|---|---|---|---| +| scTrioseq2 | acc 0.95, F1 0.95 | macro F1 0.756 ± 0.031 | 0.826 ± 0.027 | **acc 0.935 ± 0.021, macro F1 0.937 ± 0.020** | +| BRCA | F1 0.82 | 0.720 ± 0.036 | 0.726 ± 0.032 | **macro F1 0.762 ± 0.027 (weighted 0.796)** | +| GBM | not stated in text | 0.481 ± 0.138 | 0.608 ± 0.081 | **0.785 ± 0.067** | +| COAD | acc 0.87 | about 0.21 | about 0.20 | about 0.21 | + +Other runs: + +- On scTrioseq2, the notebook replica reaches macro F1 0.910 ± 0.025. The same SGD settings collapse with the Workbench's first layer. +- D with the notebook first layer scores 0.359 on scTrioseq2, so the lab-grid lr 0.01 and L2 0.01 don't suit that layer. + +## Conclusions + +- The first-layer formulation, not the hyperparameters, explains most of the gap to the paper. The BASE settings with the notebook first layer come within about 0.015 of the paper on scTrioseq2. +- The current default D is the weakest option on every dataset. +- COAD can't be evaluated. Its `label.csv` has 260 rows against 256 omics rows, and simple classifiers score below the majority-class rate, so the labels don't match the samples (see `diagnostics/2026-10-30-other-datasets`). + +## Files + +- `scripts/run_seeds.py [seeds]`: runs the seed experiments. +- `scripts/summary.py ...`: prints the summary tables. +- `results/configs_*.json`: the configurations used. +- `results/.jsonl`: per-run results. +- `results/_summary.txt`: summary tables. +- Workspaces are built with `diagnostics/2026-10-30-other-datasets/scripts/build_workspace.py `. diff --git a/diagnostics/2026-11-06-test-split-reproduction/results/BRCA.jsonl b/diagnostics/2026-11-06-test-split-reproduction/results/BRCA.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..002819986a1906a59c418d5d5f75482abd2d67e0 --- /dev/null +++ b/diagnostics/2026-11-06-test-split-reproduction/results/BRCA.jsonl @@ -0,0 +1,30 @@ +{"config": "D default", "seed": 0, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.7657142857142857, "f1_macro": 0.7439761687651749, "f1_weighted": 0.7744612874011866, "precision_macro": 0.7690821256038647, "recall_macro": 0.7526857084278374, "epochs_completed": 32, "pred_counts": [38, 23, 6, 72, 36], "collapsed": false, "recall": [0.8260869565217391, 0.8461538461538461, 0.5555555555555556, 0.735632183908046, 0.8]} +{"config": "D default", "seed": 1, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.7657142857142857, "f1_macro": 0.6759025837512732, "f1_weighted": 0.7657823862954246, "precision_macro": 0.7876636034530773, "recall_macro": 0.6892635733415344, "epochs_completed": 29, "pred_counts": [42, 27, 2, 76, 28], "collapsed": false, "recall": [0.9130434782608695, 0.8846153846153846, 0.2222222222222222, 0.7931034482758621, 0.6333333333333333]} +{"config": "D default", "seed": 2, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.8114285714285714, "f1_macro": 0.7931423407972349, "f1_weighted": 0.8242114013510994, "precision_macro": 0.7855668855668856, "recall_macro": 0.8410087264060279, "epochs_completed": 48, "pred_counts": [39, 21, 14, 74, 27], "collapsed": false, "recall": [0.8260869565217391, 0.8076923076923077, 1.0, 0.8045977011494253, 0.7666666666666667]} +{"config": "D default", "seed": 3, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.7028571428571428, "f1_macro": 0.6854015689284652, "f1_weighted": 0.7129273789673773, "precision_macro": 0.7492819339318195, "recall_macro": 0.6869713861018208, "epochs_completed": 57, "pred_counts": [38, 25, 4, 69, 39], "collapsed": false, "recall": [0.7391304347826086, 0.8846153846153846, 0.4444444444444444, 0.6666666666666666, 0.7]} +{"config": "D default", "seed": 7, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 7, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.7485714285714286, "f1_macro": 0.7070707070707071, "f1_weighted": 0.7620531849103278, "precision_macro": 0.7081360974031091, "recall_macro": 0.7380568690013967, "epochs_completed": 56, "pred_counts": [43, 23, 9, 67, 33], "collapsed": false, "recall": [0.8695652173913043, 0.8076923076923077, 0.5555555555555556, 0.7241379310344828, 0.7333333333333333]} +{"config": "D default", "seed": 42, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.7657142857142857, "f1_macro": 0.7492861748180897, "f1_weighted": 0.774858721825287, "precision_macro": 0.7457303457303457, "recall_macro": 0.7788264841938005, "epochs_completed": 38, "pred_counts": [37, 21, 12, 78, 27], "collapsed": false, "recall": [0.782608695652174, 0.8076923076923077, 0.8888888888888888, 0.7816091954022989, 0.6333333333333333]} +{"config": "D default", "seed": 123, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 123, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.7542857142857143, "f1_macro": 0.7201195060134211, "f1_weighted": 0.7634685266063347, "precision_macro": 0.7472849614153961, "recall_macro": 0.7293035533515292, "epochs_completed": 33, "pred_counts": [44, 24, 7, 77, 23], "collapsed": false, "recall": [0.8695652173913043, 0.8846153846153846, 0.5555555555555556, 0.7701149425287356, 0.5666666666666667]} +{"config": "D default", "seed": 2024, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2024, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.7485714285714286, "f1_macro": 0.7016661557387966, "f1_weighted": 0.7539214248225028, "precision_macro": 0.7418721069329589, "recall_macro": 0.6974505055164726, "epochs_completed": 28, "pred_counts": [39, 24, 8, 87, 17], "collapsed": false, "recall": [0.6956521739130435, 0.8076923076923077, 0.6666666666666666, 0.8505747126436781, 0.4666666666666667]} +{"config": "D default", "seed": 5, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 5, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.6971428571428572, "f1_macro": 0.689902352745196, "f1_weighted": 0.7148133916235159, "precision_macro": 0.6850793650793651, "recall_macro": 0.7404756596060944, "epochs_completed": 45, "pred_counts": [45, 22, 14, 66, 28], "collapsed": false, "recall": [0.7391304347826086, 0.8076923076923077, 0.8888888888888888, 0.6666666666666666, 0.6]} +{"config": "D default", "seed": 11, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 11, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.76, "f1_macro": 0.7352592131058812, "f1_weighted": 0.7704836446663403, "precision_macro": 0.7287831537831538, "recall_macro": 0.7692845884749933, "epochs_completed": 33, "pred_counts": [39, 24, 9, 66, 37], "collapsed": false, "recall": [0.8260869565217391, 0.8076923076923077, 0.6666666666666666, 0.7126436781609196, 0.8333333333333334]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 0, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.7657142857142857, "f1_macro": 0.723169147176943, "f1_weighted": 0.7732641490331073, "precision_macro": 0.7304852320675106, "recall_macro": 0.7281018465126412, "epochs_completed": 83, "pred_counts": [33, 22, 8, 79, 33], "collapsed": false, "recall": [0.6956521739130435, 0.8076923076923077, 0.5555555555555556, 0.7816091954022989, 0.8]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 1, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.7657142857142857, "f1_macro": 0.7295344971397809, "f1_weighted": 0.7765742951286274, "precision_macro": 0.7356186486621269, "recall_macro": 0.7433475569907354, "epochs_completed": 59, "pred_counts": [39, 23, 9, 77, 27], "collapsed": false, "recall": [0.782608695652174, 0.8076923076923077, 0.6666666666666666, 0.7931034482758621, 0.6666666666666666]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 2, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.7885714285714286, "f1_macro": 0.7661212712908659, "f1_weighted": 0.7994210189745566, "precision_macro": 0.75359754896597, "recall_macro": 0.8085700739373903, "epochs_completed": 84, "pred_counts": [38, 25, 13, 74, 25], "collapsed": false, "recall": [0.782608695652174, 0.9230769230769231, 0.8888888888888888, 0.7816091954022989, 0.6666666666666666]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 3, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.6914285714285714, "f1_macro": 0.6681744271988175, "f1_weighted": 0.6976819278282693, "precision_macro": 0.7298090798090798, "recall_macro": 0.6585056189853791, "epochs_completed": 82, "pred_counts": [32, 26, 4, 77, 36], "collapsed": false, "recall": [0.6956521739130435, 0.8846153846153846, 0.4444444444444444, 0.7011494252873564, 0.5666666666666667]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 7, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 7, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.76, "f1_macro": 0.7440209632239994, "f1_weighted": 0.7735428623062127, "precision_macro": 0.7939330065359478, "recall_macro": 0.752415843360371, "epochs_completed": 100, "pred_counts": [45, 25, 5, 68, 32], "collapsed": false, "recall": [0.8695652173913043, 0.8461538461538461, 0.5555555555555556, 0.7241379310344828, 0.7666666666666667]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 42, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.7714285714285715, "f1_macro": 0.7511083718541346, "f1_weighted": 0.7789654351595475, "precision_macro": 0.7439566755083996, "recall_macro": 0.7699598918489473, "epochs_completed": 67, "pred_counts": [34, 24, 10, 78, 29], "collapsed": false, "recall": [0.7391304347826086, 0.8846153846153846, 0.7777777777777778, 0.7816091954022989, 0.6666666666666666]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 123, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 123, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.7428571428571429, "f1_macro": 0.736214405095514, "f1_weighted": 0.7546953012010187, "precision_macro": 0.7446586879432624, "recall_macro": 0.7705262753238765, "epochs_completed": 63, "pred_counts": [47, 24, 8, 64, 32], "collapsed": false, "recall": [0.9565217391304348, 0.8846153846153846, 0.6666666666666666, 0.6781609195402298, 0.6666666666666666]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 2024, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2024, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.7028571428571428, "f1_macro": 0.6746308798047022, "f1_weighted": 0.7178134086700942, "precision_macro": 0.6763854425144747, "recall_macro": 0.690134163687387, "epochs_completed": 43, "pred_counts": [39, 24, 9, 72, 31], "collapsed": false, "recall": [0.6521739130434783, 0.8076923076923077, 0.6666666666666666, 0.7241379310344828, 0.6]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 5, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 5, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.7371428571428571, "f1_macro": 0.7239937823440212, "f1_weighted": 0.7518301531247148, "precision_macro": 0.7429420289855072, "recall_macro": 0.7392757467420136, "epochs_completed": 66, "pred_counts": [46, 24, 8, 72, 25], "collapsed": false, "recall": [0.8260869565217391, 0.8461538461538461, 0.6666666666666666, 0.7241379310344828, 0.6333333333333333]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 11, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 11, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.7542857142857143, "f1_macro": 0.7455712704303955, "f1_weighted": 0.7641840784622984, "precision_macro": 0.7350595238095238, "recall_macro": 0.7903645613090889, "epochs_completed": 50, "pred_counts": [40, 21, 12, 70, 32], "collapsed": false, "recall": [0.8695652173913043, 0.7692307692307693, 0.8888888888888888, 0.7241379310344828, 0.7]} +{"config": "BASE + notebook first layer", "seed": 0, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.8057142857142857, "f1_macro": 0.7486689105350364, "f1_weighted": 0.8098612601539289, "precision_macro": 0.7727109497199451, "recall_macro": 0.7459129409654147, "epochs_completed": 27, "pred_counts": [35, 23, 8, 87, 22], "collapsed": false, "recall": [0.8260869565217391, 0.8076923076923077, 0.5555555555555556, 0.8735632183908046, 0.6666666666666666]} +{"config": "BASE + notebook first layer", "seed": 1, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.8057142857142857, "f1_macro": 0.804470792148277, "f1_weighted": 0.8159936388348442, "precision_macro": 0.8251428571428571, "recall_macro": 0.8116790322787324, "epochs_completed": 20, "pred_counts": [40, 25, 7, 75, 28], "collapsed": false, "recall": [0.8695652173913043, 0.8846153846153846, 0.7777777777777778, 0.7931034482758621, 0.7333333333333333]} +{"config": "BASE + notebook first layer", "seed": 2, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.8114285714285714, "f1_macro": 0.802543539409899, "f1_weighted": 0.8229224626591317, "precision_macro": 0.7916146916146917, "recall_macro": 0.8474278245492638, "epochs_completed": 26, "pred_counts": [39, 23, 13, 74, 26], "collapsed": false, "recall": [0.8260869565217391, 0.8846153846153846, 1.0, 0.7931034482758621, 0.7333333333333333]} +{"config": "BASE + notebook first layer", "seed": 3, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.7657142857142857, "f1_macro": 0.7449397815671768, "f1_weighted": 0.7756279760656937, "precision_macro": 0.7904485538169749, "recall_macro": 0.7429467317623241, "epochs_completed": 29, "pred_counts": [38, 25, 5, 74, 33], "collapsed": false, "recall": [0.782608695652174, 0.8846153846153846, 0.5555555555555556, 0.7586206896551724, 0.7333333333333333]} +{"config": "BASE + notebook first layer", "seed": 7, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 7, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.7942857142857143, "f1_macro": 0.7324595729473778, "f1_weighted": 0.8045701012112162, "precision_macro": 0.7693969093969095, "recall_macro": 0.7294340009482438, "epochs_completed": 27, "pred_counts": [37, 26, 5, 77, 30], "collapsed": false, "recall": [0.6956521739130435, 0.8461538461538461, 0.4444444444444444, 0.8275862068965517, 0.8333333333333334]} +{"config": "BASE + notebook first layer", "seed": 42, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.7942857142857143, "f1_macro": 0.772506264805419, "f1_weighted": 0.8051051192818919, "precision_macro": 0.7638562091503268, "recall_macro": 0.8006845295301067, "epochs_completed": 20, "pred_counts": [36, 23, 10, 72, 34], "collapsed": false, "recall": [0.8260869565217391, 0.8846153846153846, 0.7777777777777778, 0.7816091954022989, 0.7333333333333333]} +{"config": "BASE + notebook first layer", "seed": 123, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 123, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.76, "f1_macro": 0.7653114989585579, "f1_weighted": 0.7733159971680982, "precision_macro": 0.772829131652661, "recall_macro": 0.7868914260818309, "epochs_completed": 24, "pred_counts": [42, 24, 8, 68, 33], "collapsed": false, "recall": [0.8260869565217391, 0.8846153846153846, 0.7777777777777778, 0.7126436781609196, 0.7333333333333333]} +{"config": "BASE + notebook first layer", "seed": 2024, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2024, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.7885714285714286, "f1_macro": 0.7301927208914126, "f1_weighted": 0.798480521444571, "precision_macro": 0.743921448921449, "recall_macro": 0.7438073271056779, "epochs_completed": 25, "pred_counts": [39, 24, 10, 81, 21], "collapsed": false, "recall": [0.782608695652174, 0.8076923076923077, 0.6666666666666666, 0.8620689655172413, 0.6]} +{"config": "BASE + notebook first layer", "seed": 5, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 5, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.7428571428571429, "f1_macro": 0.7404041515384897, "f1_weighted": 0.7535813867851026, "precision_macro": 0.7227756452843188, "recall_macro": 0.7958379784466741, "epochs_completed": 20, "pred_counts": [38, 23, 12, 62, 40], "collapsed": false, "recall": [0.782608695652174, 0.8076923076923077, 0.8888888888888888, 0.6666666666666666, 0.8333333333333334]} +{"config": "BASE + notebook first layer", "seed": 11, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 11, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 175, "accuracy": 0.7942857142857143, "f1_macro": 0.7783553678827129, "f1_weighted": 0.8033745148460598, "precision_macro": 0.7791058201058201, "recall_macro": 0.791233101398019, "epochs_completed": 37, "pred_counts": [35, 27, 8, 75, 30], "collapsed": false, "recall": [0.7391304347826086, 0.8461538461538461, 0.7777777777777778, 0.7931034482758621, 0.8]} diff --git a/diagnostics/2026-11-06-test-split-reproduction/results/BRCA_shard0.log b/diagnostics/2026-11-06-test-split-reproduction/results/BRCA_shard0.log new file mode 100644 index 0000000000000000000000000000000000000000..cd93fd77e41349ac5d851f3da75422421e2278d5 --- /dev/null +++ b/diagnostics/2026-11-06-test-split-reproduction/results/BRCA_shard0.log @@ -0,0 +1,31 @@ +[0] D default seed=0 acc=0.766 f1w=0.774 +[0] D default seed=1 acc=0.766 f1w=0.766 +[0] D default seed=2 acc=0.811 f1w=0.824 +[0] D default seed=3 acc=0.703 f1w=0.713 +[0] D default seed=7 acc=0.749 f1w=0.762 +[0] D default seed=42 acc=0.766 f1w=0.775 +[0] D default seed=123 acc=0.754 f1w=0.763 +[0] D default seed=2024 acc=0.749 f1w=0.754 +[0] D default seed=5 acc=0.697 f1w=0.715 +[0] D default seed=11 acc=0.760 f1w=0.770 +[0] BASE lr1e-3 wd1e-4 seed=0 acc=0.766 f1w=0.773 +[0] BASE lr1e-3 wd1e-4 seed=1 acc=0.766 f1w=0.777 +[0] BASE lr1e-3 wd1e-4 seed=2 acc=0.789 f1w=0.799 +[0] BASE lr1e-3 wd1e-4 seed=3 acc=0.691 f1w=0.698 +[0] BASE lr1e-3 wd1e-4 seed=7 acc=0.760 f1w=0.774 +[0] BASE lr1e-3 wd1e-4 seed=42 acc=0.771 f1w=0.779 +[0] BASE lr1e-3 wd1e-4 seed=123 acc=0.743 f1w=0.755 +[0] BASE lr1e-3 wd1e-4 seed=2024 acc=0.703 f1w=0.718 +[0] BASE lr1e-3 wd1e-4 seed=5 acc=0.737 f1w=0.752 +[0] BASE lr1e-3 wd1e-4 seed=11 acc=0.754 f1w=0.764 +[0] BASE + notebook first layer seed=0 acc=0.806 f1w=0.810 +[0] BASE + notebook first layer seed=1 acc=0.806 f1w=0.816 +[0] BASE + notebook first layer seed=2 acc=0.811 f1w=0.823 +[0] BASE + notebook first layer seed=3 acc=0.766 f1w=0.776 +[0] BASE + notebook first layer seed=7 acc=0.794 f1w=0.805 +[0] BASE + notebook first layer seed=42 acc=0.794 f1w=0.805 +[0] BASE + notebook first layer seed=123 acc=0.760 f1w=0.773 +[0] BASE + notebook first layer seed=2024 acc=0.789 f1w=0.798 +[0] BASE + notebook first layer seed=5 acc=0.743 f1w=0.754 +[0] BASE + notebook first layer seed=11 acc=0.794 f1w=0.803 +SHARD 0 DONE diff --git a/diagnostics/2026-11-06-test-split-reproduction/results/BRCA_summary.txt b/diagnostics/2026-11-06-test-split-reproduction/results/BRCA_summary.txt new file mode 100644 index 0000000000000000000000000000000000000000..52ecd283e00b02f6640f53d8c974e7b8b8b1e381 --- /dev/null +++ b/diagnostics/2026-11-06-test-split-reproduction/results/BRCA_summary.txt @@ -0,0 +1,7 @@ + +=== BRCA (paper: F1 0.82) + seeds accuracy macro F1 weighted F1 precision recall macro F1 min/max collapsed +config +BASE + notebook first layer 10 0.786 ± 0.023 0.762 ± 0.027 0.796 ± 0.022 0.773 ± 0.028 0.780 ± 0.038 0.730 / 0.804 0 +BASE lr1e-3 wd1e-4 10 0.748 ± 0.031 0.726 ± 0.032 0.759 ± 0.030 0.739 ± 0.029 0.745 ± 0.045 0.668 / 0.766 0 +D default 10 0.752 ± 0.033 0.720 ± 0.036 0.762 ± 0.032 0.745 ± 0.032 0.742 ± 0.047 0.676 / 0.793 0 diff --git a/diagnostics/2026-11-06-test-split-reproduction/results/COAD.jsonl b/diagnostics/2026-11-06-test-split-reproduction/results/COAD.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..19f815b3b17bcbda6051a469a20468ee44d07213 --- /dev/null +++ b/diagnostics/2026-11-06-test-split-reproduction/results/COAD.jsonl @@ -0,0 +1,30 @@ +{"config": "D default", "seed": 0, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.6730769230769231, "f1_macro": 0.2845528455284553, "f1_weighted": 0.599437148217636, "precision_macro": 0.3421985815602837, "recall_macro": 0.29126984126984123, "epochs_completed": 29, "pred_counts": [47, 3, 0, 2], "collapsed": false, "recall": [0.9428571428571428, 0.2222222222222222, 0.0, 0.0]} +{"config": "D default", "seed": 1, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.6730769230769231, "f1_macro": 0.20114942528735633, "f1_weighted": 0.541556145004421, "precision_macro": 0.16826923076923078, "recall_macro": 0.25, "epochs_completed": 31, "pred_counts": [52, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0]} +{"config": "D default", "seed": 2, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.6730769230769231, "f1_macro": 0.20114942528735633, "f1_weighted": 0.541556145004421, "precision_macro": 0.16826923076923078, "recall_macro": 0.25, "epochs_completed": 14, "pred_counts": [52, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0]} +{"config": "D default", "seed": 3, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.6730769230769231, "f1_macro": 0.20114942528735633, "f1_weighted": 0.541556145004421, "precision_macro": 0.16826923076923078, "recall_macro": 0.25, "epochs_completed": 53, "pred_counts": [52, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0]} +{"config": "D default", "seed": 7, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 7, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.6346153846153846, "f1_macro": 0.23123262279888787, "f1_weighted": 0.5456262921508519, "precision_macro": 0.22916666666666666, "recall_macro": 0.2563492063492063, "epochs_completed": 40, "pred_counts": [48, 4, 0, 0], "collapsed": false, "recall": [0.9142857142857143, 0.1111111111111111, 0.0, 0.0]} +{"config": "D default", "seed": 42, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.36538461538461536, "f1_macro": 0.2188825373409552, "f1_weighted": 0.40316174697513446, "precision_macro": 0.25554347826086954, "recall_macro": 0.22619047619047616, "epochs_completed": 38, "pred_counts": [23, 25, 4, 0], "collapsed": false, "recall": [0.42857142857142855, 0.3333333333333333, 0.14285714285714285, 0.0]} +{"config": "D default", "seed": 123, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 123, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.6730769230769231, "f1_macro": 0.20114942528735633, "f1_weighted": 0.541556145004421, "precision_macro": 0.16826923076923078, "recall_macro": 0.25, "epochs_completed": 13, "pred_counts": [52, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0]} +{"config": "D default", "seed": 2024, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2024, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.17307692307692307, "f1_macro": 0.08927738927738926, "f1_weighted": 0.10329926483772636, "precision_macro": 0.16145833333333334, "recall_macro": 0.2642857142857143, "epochs_completed": 30, "pred_counts": [4, 0, 48, 0], "collapsed": false, "recall": [0.05714285714285714, 0.0, 1.0, 0.0]} +{"config": "D default", "seed": 5, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 5, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.5769230769230769, "f1_macro": 0.24386724386724387, "f1_weighted": 0.5369075369075369, "precision_macro": 0.2976190476190476, "recall_macro": 0.24285714285714288, "epochs_completed": 45, "pred_counts": [42, 0, 2, 8], "collapsed": false, "recall": [0.8285714285714286, 0.0, 0.14285714285714285, 0.0]} +{"config": "D default", "seed": 11, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 11, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.6730769230769231, "f1_macro": 0.20114942528735633, "f1_weighted": 0.541556145004421, "precision_macro": 0.16826923076923078, "recall_macro": 0.25, "epochs_completed": 13, "pred_counts": [52, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 0, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.6730769230769231, "f1_macro": 0.20114942528735633, "f1_weighted": 0.541556145004421, "precision_macro": 0.16826923076923078, "recall_macro": 0.25, "epochs_completed": 33, "pred_counts": [52, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 1, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.6730769230769231, "f1_macro": 0.20114942528735633, "f1_weighted": 0.541556145004421, "precision_macro": 0.16826923076923078, "recall_macro": 0.25, "epochs_completed": 20, "pred_counts": [52, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 2, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.6730769230769231, "f1_macro": 0.20114942528735633, "f1_weighted": 0.541556145004421, "precision_macro": 0.16826923076923078, "recall_macro": 0.25, "epochs_completed": 14, "pred_counts": [52, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 3, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.6730769230769231, "f1_macro": 0.20114942528735633, "f1_weighted": 0.541556145004421, "precision_macro": 0.16826923076923078, "recall_macro": 0.25, "epochs_completed": 23, "pred_counts": [52, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 7, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 7, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.6730769230769231, "f1_macro": 0.20114942528735633, "f1_weighted": 0.541556145004421, "precision_macro": 0.16826923076923078, "recall_macro": 0.25, "epochs_completed": 40, "pred_counts": [52, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 42, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.6730769230769231, "f1_macro": 0.20114942528735633, "f1_weighted": 0.541556145004421, "precision_macro": 0.16826923076923078, "recall_macro": 0.25, "epochs_completed": 26, "pred_counts": [52, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 123, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 123, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.6730769230769231, "f1_macro": 0.20114942528735633, "f1_weighted": 0.541556145004421, "precision_macro": 0.16826923076923078, "recall_macro": 0.25, "epochs_completed": 15, "pred_counts": [52, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 2024, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2024, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.6730769230769231, "f1_macro": 0.20114942528735633, "f1_weighted": 0.541556145004421, "precision_macro": 0.16826923076923078, "recall_macro": 0.25, "epochs_completed": 27, "pred_counts": [52, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 5, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 5, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.6730769230769231, "f1_macro": 0.20114942528735633, "f1_weighted": 0.541556145004421, "precision_macro": 0.16826923076923078, "recall_macro": 0.25, "epochs_completed": 44, "pred_counts": [52, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 11, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 11, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.6730769230769231, "f1_macro": 0.20114942528735633, "f1_weighted": 0.541556145004421, "precision_macro": 0.16826923076923078, "recall_macro": 0.25, "epochs_completed": 36, "pred_counts": [52, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0]} +{"config": "BASE + notebook first layer", "seed": 0, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.5384615384615384, "f1_macro": 0.22095238095238096, "f1_weighted": 0.49963369963369964, "precision_macro": 0.20416666666666666, "recall_macro": 0.24126984126984127, "epochs_completed": 18, "pred_counts": [40, 12, 0, 0], "collapsed": false, "recall": [0.7428571428571429, 0.2222222222222222, 0.0, 0.0]} +{"config": "BASE + notebook first layer", "seed": 1, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.6538461538461539, "f1_macro": 0.244249726177437, "f1_weighted": 0.5666863257224704, "precision_macro": 0.296875, "recall_macro": 0.2634920634920635, "epochs_completed": 20, "pred_counts": [48, 2, 2, 0], "collapsed": false, "recall": [0.9428571428571428, 0.1111111111111111, 0.0, 0.0]} +{"config": "BASE + notebook first layer", "seed": 2, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.6730769230769231, "f1_macro": 0.20114942528735633, "f1_weighted": 0.541556145004421, "precision_macro": 0.16826923076923078, "recall_macro": 0.25, "epochs_completed": 14, "pred_counts": [52, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0]} +{"config": "BASE + notebook first layer", "seed": 3, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.6730769230769231, "f1_macro": 0.20114942528735633, "f1_weighted": 0.541556145004421, "precision_macro": 0.16826923076923078, "recall_macro": 0.25, "epochs_completed": 13, "pred_counts": [52, 0, 0, 0], "collapsed": true, "recall": [1.0, 0.0, 0.0, 0.0]} +{"config": "BASE + notebook first layer", "seed": 7, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 7, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.4807692307692308, "f1_macro": 0.18777777777777777, "f1_weighted": 0.45000000000000007, "precision_macro": 0.17777777777777776, "recall_macro": 0.19920634920634922, "epochs_completed": 28, "pred_counts": [40, 9, 3, 0], "collapsed": false, "recall": [0.6857142857142857, 0.1111111111111111, 0.0, 0.0]} +{"config": "BASE + notebook first layer", "seed": 42, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.4807692307692308, "f1_macro": 0.16447368421052633, "f1_weighted": 0.44281376518218624, "precision_macro": 0.1524390243902439, "recall_macro": 0.17857142857142858, "epochs_completed": 25, "pred_counts": [41, 4, 7, 0], "collapsed": false, "recall": [0.7142857142857143, 0.0, 0.0, 0.0]} +{"config": "BASE + notebook first layer", "seed": 123, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 123, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.6346153846153846, "f1_macro": 0.23123262279888787, "f1_weighted": 0.5456262921508519, "precision_macro": 0.22916666666666666, "recall_macro": 0.2563492063492063, "epochs_completed": 22, "pred_counts": [48, 4, 0, 0], "collapsed": false, "recall": [0.9142857142857143, 0.1111111111111111, 0.0, 0.0]} +{"config": "BASE + notebook first layer", "seed": 2024, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2024, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.5576923076923077, "f1_macro": 0.20889894419306185, "f1_weighted": 0.503596704954171, "precision_macro": 0.19404069767441862, "recall_macro": 0.2277777777777778, "epochs_completed": 23, "pred_counts": [43, 8, 1, 0], "collapsed": false, "recall": [0.8, 0.1111111111111111, 0.0, 0.0]} +{"config": "BASE + notebook first layer", "seed": 5, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 5, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.5, "f1_macro": 0.17333333333333334, "f1_weighted": 0.4666666666666667, "precision_macro": 0.1625, "recall_macro": 0.18571428571428572, "epochs_completed": 24, "pred_counts": [40, 7, 5, 0], "collapsed": false, "recall": [0.7428571428571429, 0.0, 0.0, 0.0]} +{"config": "BASE + notebook first layer", "seed": 11, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 11, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 52, "accuracy": 0.5769230769230769, "f1_macro": 0.23444976076555024, "f1_weighted": 0.5259477364740524, "precision_macro": 0.21666666666666667, "recall_macro": 0.25555555555555554, "epochs_completed": 26, "pred_counts": [42, 10, 0, 0], "collapsed": false, "recall": [0.8, 0.2222222222222222, 0.0, 0.0]} diff --git a/diagnostics/2026-11-06-test-split-reproduction/results/COAD_shard0.log b/diagnostics/2026-11-06-test-split-reproduction/results/COAD_shard0.log new file mode 100644 index 0000000000000000000000000000000000000000..1cebecbf6d4c2aa6b53bfae6c43f88d321b0a153 --- /dev/null +++ b/diagnostics/2026-11-06-test-split-reproduction/results/COAD_shard0.log @@ -0,0 +1,31 @@ +[0] D default seed=0 acc=0.673 f1w=0.599 +[0] D default seed=1 acc=0.673 f1w=0.542 +[0] D default seed=2 acc=0.673 f1w=0.542 +[0] D default seed=3 acc=0.673 f1w=0.542 +[0] D default seed=7 acc=0.635 f1w=0.546 +[0] D default seed=42 acc=0.365 f1w=0.403 +[0] D default seed=123 acc=0.673 f1w=0.542 +[0] D default seed=2024 acc=0.173 f1w=0.103 +[0] D default seed=5 acc=0.577 f1w=0.537 +[0] D default seed=11 acc=0.673 f1w=0.542 +[0] BASE lr1e-3 wd1e-4 seed=0 acc=0.673 f1w=0.542 +[0] BASE lr1e-3 wd1e-4 seed=1 acc=0.673 f1w=0.542 +[0] BASE lr1e-3 wd1e-4 seed=2 acc=0.673 f1w=0.542 +[0] BASE lr1e-3 wd1e-4 seed=3 acc=0.673 f1w=0.542 +[0] BASE lr1e-3 wd1e-4 seed=7 acc=0.673 f1w=0.542 +[0] BASE lr1e-3 wd1e-4 seed=42 acc=0.673 f1w=0.542 +[0] BASE lr1e-3 wd1e-4 seed=123 acc=0.673 f1w=0.542 +[0] BASE lr1e-3 wd1e-4 seed=2024 acc=0.673 f1w=0.542 +[0] BASE lr1e-3 wd1e-4 seed=5 acc=0.673 f1w=0.542 +[0] BASE lr1e-3 wd1e-4 seed=11 acc=0.673 f1w=0.542 +[0] BASE + notebook first layer seed=0 acc=0.538 f1w=0.500 +[0] BASE + notebook first layer seed=1 acc=0.654 f1w=0.567 +[0] BASE + notebook first layer seed=2 acc=0.673 f1w=0.542 +[0] BASE + notebook first layer seed=3 acc=0.673 f1w=0.542 +[0] BASE + notebook first layer seed=7 acc=0.481 f1w=0.450 +[0] BASE + notebook first layer seed=42 acc=0.481 f1w=0.443 +[0] BASE + notebook first layer seed=123 acc=0.635 f1w=0.546 +[0] BASE + notebook first layer seed=2024 acc=0.558 f1w=0.504 +[0] BASE + notebook first layer seed=5 acc=0.500 f1w=0.467 +[0] BASE + notebook first layer seed=11 acc=0.577 f1w=0.526 +SHARD 0 DONE diff --git a/diagnostics/2026-11-06-test-split-reproduction/results/COAD_summary.txt b/diagnostics/2026-11-06-test-split-reproduction/results/COAD_summary.txt new file mode 100644 index 0000000000000000000000000000000000000000..b70d68564fab573335184f439022e4416e1987d8 --- /dev/null +++ b/diagnostics/2026-11-06-test-split-reproduction/results/COAD_summary.txt @@ -0,0 +1,7 @@ + +=== COAD (paper: accuracy 0.87, recall 0.80) + seeds accuracy macro F1 weighted F1 precision recall macro F1 min/max collapsed +config +D default 10 0.579 ± 0.172 0.207 ± 0.050 0.490 ± 0.144 0.213 ± 0.065 0.253 ± 0.017 0.089 / 0.285 5 +BASE + notebook first layer 10 0.577 ± 0.077 0.207 ± 0.026 0.508 ± 0.043 0.197 ± 0.043 0.231 ± 0.032 0.164 / 0.244 2 +BASE lr1e-3 wd1e-4 10 0.673 ± 0.000 0.201 ± 0.000 0.542 ± 0.000 0.168 ± 0.000 0.250 ± 0.000 0.201 / 0.201 10 diff --git a/diagnostics/2026-11-06-test-split-reproduction/results/GBM.jsonl b/diagnostics/2026-11-06-test-split-reproduction/results/GBM.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..6c46e31a1a801a4a46747080f1241dcb0aa6518a --- /dev/null +++ b/diagnostics/2026-11-06-test-split-reproduction/results/GBM.jsonl @@ -0,0 +1,30 @@ +{"config": "D default", "seed": 0, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.4897959183673469, "f1_macro": 0.48192717086834735, "f1_weighted": 0.47811581775567374, "precision_macro": 0.5482051282051282, "recall_macro": 0.45999999999999996, "epochs_completed": 83, "pred_counts": [13, 6, 20, 2, 8], "collapsed": false, "recall": [0.3333333333333333, 0.2222222222222222, 0.8, 0.5, 0.4444444444444444]} +{"config": "D default", "seed": 1, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.3469387755102041, "f1_macro": 0.3978494623655914, "f1_weighted": 0.28308097432521395, "precision_macro": 0.3592982456140351, "recall_macro": 0.4611111111111111, "epochs_completed": 61, "pred_counts": [19, 15, 2, 4, 9], "collapsed": false, "recall": [0.4166666666666667, 0.8888888888888888, 0.0, 1.0, 0.0]} +{"config": "D default", "seed": 2, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.673469387755102, "f1_macro": 0.7275295524908528, "f1_weighted": 0.6543101690436941, "precision_macro": 0.7300581771170007, "recall_macro": 0.7477777777777778, "epochs_completed": 110, "pred_counts": [7, 8, 17, 4, 13], "collapsed": false, "recall": [0.25, 0.8888888888888888, 0.6, 1.0, 1.0]} +{"config": "D default", "seed": 3, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.5714285714285714, "f1_macro": 0.5458326635001827, "f1_weighted": 0.5129229828577393, "precision_macro": 0.5310005783689994, "recall_macro": 0.5866666666666667, "epochs_completed": 86, "pred_counts": [13, 14, 19, 3, 0], "collapsed": false, "recall": [0.5833333333333334, 1.0, 0.6, 0.75, 0.0]} +{"config": "D default", "seed": 7, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 7, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.46938775510204084, "f1_macro": 0.43533936651583716, "f1_weighted": 0.39019761750854187, "precision_macro": 0.3780922895473979, "recall_macro": 0.5333333333333334, "epochs_completed": 29, "pred_counts": [0, 17, 19, 6, 7], "collapsed": false, "recall": [0.0, 0.7777777777777778, 0.6666666666666666, 1.0, 0.2222222222222222]} +{"config": "D default", "seed": 42, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.40816326530612246, "f1_macro": 0.4068878833584716, "f1_weighted": 0.3476108392074778, "precision_macro": 0.38666666666666666, "recall_macro": 0.47333333333333333, "epochs_completed": 32, "pred_counts": [5, 15, 24, 5, 0], "collapsed": false, "recall": [0.16666666666666666, 0.6666666666666666, 0.5333333333333333, 1.0, 0.0]} +{"config": "D default", "seed": 123, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 123, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.42857142857142855, "f1_macro": 0.40385281385281385, "f1_weighted": 0.372806785051683, "precision_macro": 0.41053639846743295, "recall_macro": 0.4322222222222222, "epochs_completed": 48, "pred_counts": [9, 6, 29, 4, 1], "collapsed": false, "recall": [0.16666666666666666, 0.4444444444444444, 0.8, 0.75, 0.0]} +{"config": "D default", "seed": 2024, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2024, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.5918367346938775, "f1_macro": 0.6480392156862744, "f1_weighted": 0.5843670801653995, "precision_macro": 0.635, "recall_macro": 0.6744444444444444, "epochs_completed": 142, "pred_counts": [12, 8, 12, 5, 12], "collapsed": false, "recall": [0.4166666666666667, 0.7777777777777778, 0.4, 1.0, 0.7777777777777778]} +{"config": "D default", "seed": 5, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 5, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.3673469387755102, "f1_macro": 0.24370015948963317, "f1_weighted": 0.20974514207596917, "precision_macro": 0.18095238095238092, "recall_macro": 0.38666666666666666, "epochs_completed": 44, "pred_counts": [0, 0, 42, 7, 0], "collapsed": false, "recall": [0.0, 0.0, 0.9333333333333333, 1.0, 0.0]} +{"config": "D default", "seed": 11, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 11, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.4897959183673469, "f1_macro": 0.5145318493366031, "f1_weighted": 0.4643598242919126, "precision_macro": 0.5568253968253968, "recall_macro": 0.5566666666666666, "epochs_completed": 60, "pred_counts": [7, 18, 16, 5, 3], "collapsed": false, "recall": [0.25, 0.7777777777777778, 0.5333333333333333, 1.0, 0.2222222222222222]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 0, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.5714285714285714, "f1_macro": 0.5842610162344508, "f1_weighted": 0.5696472137241994, "precision_macro": 0.586111111111111, "recall_macro": 0.5833333333333334, "epochs_completed": 100, "pred_counts": [12, 9, 16, 4, 8], "collapsed": false, "recall": [0.5, 0.5555555555555556, 0.6666666666666666, 0.75, 0.4444444444444444]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 1, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.6530612244897959, "f1_macro": 0.6954822954822955, "f1_weighted": 0.6474545862300963, "precision_macro": 0.7388888888888889, "recall_macro": 0.6888888888888889, "epochs_completed": 100, "pred_counts": [12, 9, 20, 4, 4], "collapsed": false, "recall": [0.6666666666666666, 0.7777777777777778, 0.6666666666666666, 1.0, 0.3333333333333333]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 2, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.6326530612244898, "f1_macro": 0.6542457542457543, "f1_weighted": 0.6135089400395524, "precision_macro": 0.6610314685314684, "recall_macro": 0.6788888888888889, "epochs_completed": 100, "pred_counts": [16, 13, 11, 4, 5], "collapsed": false, "recall": [0.75, 0.8888888888888888, 0.5333333333333333, 1.0, 0.2222222222222222]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 3, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.5714285714285714, "f1_macro": 0.6111838817433388, "f1_weighted": 0.5574804792376458, "precision_macro": 0.6012698412698413, "recall_macro": 0.6455555555555555, "epochs_completed": 100, "pred_counts": [7, 14, 14, 5, 9], "collapsed": false, "recall": [0.25, 0.6666666666666666, 0.5333333333333333, 1.0, 0.7777777777777778]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 7, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 7, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.4897959183673469, "f1_macro": 0.4295061260578502, "f1_weighted": 0.4443550819342516, "precision_macro": 0.38412698412698415, "recall_macro": 0.5222222222222223, "epochs_completed": 44, "pred_counts": [14, 12, 14, 9, 0], "collapsed": false, "recall": [0.5, 0.4444444444444444, 0.6666666666666666, 1.0, 0.0]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 42, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.6326530612244898, "f1_macro": 0.6864206552186092, "f1_weighted": 0.6364525106108696, "precision_macro": 0.7193833067517278, "recall_macro": 0.6866666666666668, "epochs_completed": 100, "pred_counts": [11, 9, 19, 5, 5], "collapsed": false, "recall": [0.5, 0.7777777777777778, 0.6, 1.0, 0.5555555555555556]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 123, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 123, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.6326530612244898, "f1_macro": 0.6517024872496672, "f1_weighted": 0.6295384667978556, "precision_macro": 0.6407575757575759, "recall_macro": 0.6688888888888889, "epochs_completed": 100, "pred_counts": [11, 8, 15, 5, 10], "collapsed": false, "recall": [0.5, 0.5555555555555556, 0.7333333333333333, 1.0, 0.5555555555555556]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 2024, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2024, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.5306122448979592, "f1_macro": 0.5896825396825397, "f1_weighted": 0.5324748947197927, "precision_macro": 0.5891208791208791, "recall_macro": 0.6044444444444445, "epochs_completed": 100, "pred_counts": [12, 7, 13, 5, 12], "collapsed": false, "recall": [0.3333333333333333, 0.6666666666666666, 0.4666666666666667, 1.0, 0.5555555555555556]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 5, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 5, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.6326530612244898, "f1_macro": 0.650937019969278, "f1_weighted": 0.6301514154048716, "precision_macro": 0.6849999999999999, "recall_macro": 0.6666666666666666, "epochs_completed": 100, "pred_counts": [16, 6, 16, 6, 5], "collapsed": false, "recall": [0.6666666666666666, 0.5555555555555556, 0.6666666666666666, 1.0, 0.4444444444444444]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 11, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 11, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.4489795918367347, "f1_macro": 0.5281470588235294, "f1_weighted": 0.43916566626650655, "precision_macro": 0.5407142857142857, "recall_macro": 0.5344444444444444, "epochs_completed": 100, "pred_counts": [20, 8, 10, 4, 7], "collapsed": false, "recall": [0.5833333333333334, 0.5555555555555556, 0.2, 1.0, 0.3333333333333333]} +{"config": "BASE + notebook first layer", "seed": 0, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.6938775510204082, "f1_macro": 0.6944444444444444, "f1_weighted": 0.6700680272108844, "precision_macro": 0.7404761904761904, "recall_macro": 0.7011111111111111, "epochs_completed": 43, "pred_counts": [12, 3, 21, 4, 9], "collapsed": false, "recall": [0.75, 0.2222222222222222, 0.8666666666666667, 1.0, 0.6666666666666666]} +{"config": "BASE + notebook first layer", "seed": 1, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.7755102040816326, "f1_macro": 0.7887445887445887, "f1_weighted": 0.7633183143387224, "precision_macro": 0.8144444444444445, "recall_macro": 0.7955555555555556, "epochs_completed": 36, "pred_counts": [10, 5, 18, 4, 12], "collapsed": false, "recall": [0.6666666666666666, 0.4444444444444444, 0.8666666666666667, 1.0, 1.0]} +{"config": "BASE + notebook first layer", "seed": 2, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.8367346938775511, "f1_macro": 0.8549344375431331, "f1_weighted": 0.8359741412092787, "precision_macro": 0.8538073038073039, "recall_macro": 0.8655555555555555, "epochs_completed": 38, "pred_counts": [11, 12, 13, 4, 9], "collapsed": false, "recall": [0.75, 1.0, 0.8, 1.0, 0.7777777777777778]} +{"config": "BASE + notebook first layer", "seed": 3, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.8775510204081632, "f1_macro": 0.8919621666899344, "f1_weighted": 0.878695293557881, "precision_macro": 0.9028571428571428, "recall_macro": 0.89, "epochs_completed": 36, "pred_counts": [14, 10, 14, 4, 7], "collapsed": false, "recall": [0.9166666666666666, 0.8888888888888888, 0.8666666666666667, 1.0, 0.7777777777777778]} +{"config": "BASE + notebook first layer", "seed": 7, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 7, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.7346938775510204, "f1_macro": 0.7529765886287626, "f1_weighted": 0.7326394102791619, "precision_macro": 0.7642357642357643, "recall_macro": 0.7666666666666666, "epochs_completed": 62, "pred_counts": [13, 14, 11, 4, 7], "collapsed": false, "recall": [0.8333333333333334, 0.8888888888888888, 0.6666666666666666, 1.0, 0.4444444444444444]} +{"config": "BASE + notebook first layer", "seed": 42, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.7551020408163265, "f1_macro": 0.7448458310527275, "f1_weighted": 0.739558549199647, "precision_macro": 0.806078431372549, "recall_macro": 0.7744444444444445, "epochs_completed": 56, "pred_counts": [17, 12, 12, 5, 3], "collapsed": false, "recall": [0.9166666666666666, 0.8888888888888888, 0.7333333333333333, 1.0, 0.3333333333333333]} +{"config": "BASE + notebook first layer", "seed": 123, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 123, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.8367346938775511, "f1_macro": 0.8583333333333334, "f1_weighted": 0.838265306122449, "precision_macro": 0.8687878787878788, "recall_macro": 0.8566666666666667, "epochs_completed": 44, "pred_counts": [12, 7, 15, 4, 11], "collapsed": false, "recall": [0.75, 0.7777777777777778, 0.8666666666666667, 1.0, 0.8888888888888888]} +{"config": "BASE + notebook first layer", "seed": 2024, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2024, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.7755102040816326, "f1_macro": 0.8109433002796846, "f1_weighted": 0.7750898986596927, "precision_macro": 0.8098989898989899, "recall_macro": 0.8133333333333332, "epochs_completed": 39, "pred_counts": [11, 9, 15, 4, 10], "collapsed": false, "recall": [0.6666666666666666, 0.8888888888888888, 0.7333333333333333, 1.0, 0.7777777777777778]} +{"config": "BASE + notebook first layer", "seed": 5, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 5, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.7346938775510204, "f1_macro": 0.7329006906406287, "f1_weighted": 0.7105870688356314, "precision_macro": 0.746842105263158, "recall_macro": 0.7455555555555555, "epochs_completed": 27, "pred_counts": [12, 10, 19, 4, 4], "collapsed": false, "recall": [0.75, 0.8888888888888888, 0.8666666666666667, 1.0, 0.2222222222222222]} +{"config": "BASE + notebook first layer", "seed": 11, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 11, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 49, "accuracy": 0.6938775510204082, "f1_macro": 0.7203842940685046, "f1_weighted": 0.6930847526980717, "precision_macro": 0.7838095238095238, "recall_macro": 0.7122222222222222, "epochs_completed": 39, "pred_counts": [7, 12, 21, 4, 5], "collapsed": false, "recall": [0.5833333333333334, 0.6666666666666666, 0.8666666666666667, 1.0, 0.4444444444444444]} diff --git a/diagnostics/2026-11-06-test-split-reproduction/results/GBM_shard0.log b/diagnostics/2026-11-06-test-split-reproduction/results/GBM_shard0.log new file mode 100644 index 0000000000000000000000000000000000000000..47ea8bb94677bf23787d48d06ca217a80bb137e0 --- /dev/null +++ b/diagnostics/2026-11-06-test-split-reproduction/results/GBM_shard0.log @@ -0,0 +1,31 @@ +[0] D default seed=0 acc=0.490 f1w=0.478 +[0] D default seed=1 acc=0.347 f1w=0.283 +[0] D default seed=2 acc=0.673 f1w=0.654 +[0] D default seed=3 acc=0.571 f1w=0.513 +[0] D default seed=7 acc=0.469 f1w=0.390 +[0] D default seed=42 acc=0.408 f1w=0.348 +[0] D default seed=123 acc=0.429 f1w=0.373 +[0] D default seed=2024 acc=0.592 f1w=0.584 +[0] D default seed=5 acc=0.367 f1w=0.210 +[0] D default seed=11 acc=0.490 f1w=0.464 +[0] BASE lr1e-3 wd1e-4 seed=0 acc=0.571 f1w=0.570 +[0] BASE lr1e-3 wd1e-4 seed=1 acc=0.653 f1w=0.647 +[0] BASE lr1e-3 wd1e-4 seed=2 acc=0.633 f1w=0.614 +[0] BASE lr1e-3 wd1e-4 seed=3 acc=0.571 f1w=0.557 +[0] BASE lr1e-3 wd1e-4 seed=7 acc=0.490 f1w=0.444 +[0] BASE lr1e-3 wd1e-4 seed=42 acc=0.633 f1w=0.636 +[0] BASE lr1e-3 wd1e-4 seed=123 acc=0.633 f1w=0.630 +[0] BASE lr1e-3 wd1e-4 seed=2024 acc=0.531 f1w=0.532 +[0] BASE lr1e-3 wd1e-4 seed=5 acc=0.633 f1w=0.630 +[0] BASE lr1e-3 wd1e-4 seed=11 acc=0.449 f1w=0.439 +[0] BASE + notebook first layer seed=0 acc=0.694 f1w=0.670 +[0] BASE + notebook first layer seed=1 acc=0.776 f1w=0.763 +[0] BASE + notebook first layer seed=2 acc=0.837 f1w=0.836 +[0] BASE + notebook first layer seed=3 acc=0.878 f1w=0.879 +[0] BASE + notebook first layer seed=7 acc=0.735 f1w=0.733 +[0] BASE + notebook first layer seed=42 acc=0.755 f1w=0.740 +[0] BASE + notebook first layer seed=123 acc=0.837 f1w=0.838 +[0] BASE + notebook first layer seed=2024 acc=0.776 f1w=0.775 +[0] BASE + notebook first layer seed=5 acc=0.735 f1w=0.711 +[0] BASE + notebook first layer seed=11 acc=0.694 f1w=0.693 +SHARD 0 DONE diff --git a/diagnostics/2026-11-06-test-split-reproduction/results/GBM_summary.txt b/diagnostics/2026-11-06-test-split-reproduction/results/GBM_summary.txt new file mode 100644 index 0000000000000000000000000000000000000000..4b11f4719ce5ba221b70561d3ab5bf0eaede9ead --- /dev/null +++ b/diagnostics/2026-11-06-test-split-reproduction/results/GBM_summary.txt @@ -0,0 +1,7 @@ + +=== GBM (paper: deep models underperformed classical ones (no number in text)) + seeds accuracy macro F1 weighted F1 precision recall macro F1 min/max collapsed +config +BASE + notebook first layer 10 0.771 ± 0.062 0.785 ± 0.067 0.764 ± 0.069 0.809 ± 0.053 0.792 ± 0.064 0.694 / 0.892 0 +BASE lr1e-3 wd1e-4 10 0.580 ± 0.070 0.608 ± 0.081 0.570 ± 0.077 0.615 ± 0.102 0.628 ± 0.063 0.430 / 0.695 0 +D default 10 0.484 ± 0.104 0.481 ± 0.138 0.430 ± 0.136 0.472 ± 0.159 0.531 ± 0.113 0.244 / 0.728 0 diff --git a/diagnostics/2026-11-06-test-split-reproduction/results/configs_other.json b/diagnostics/2026-11-06-test-split-reproduction/results/configs_other.json new file mode 100644 index 0000000000000000000000000000000000000000..eb6d37cb095c29dfe05b4a62f4f9faae726d6471 --- /dev/null +++ b/diagnostics/2026-11-06-test-split-reproduction/results/configs_other.json @@ -0,0 +1,26 @@ +{ + "D default": {}, + "BASE lr1e-3 wd1e-4": { + "optimizer": "Adam", + "learning_rate": 0.001, + "weight_decay": 0.0001, + "batch_size": 16, + "projection_dim": 64, + "epochs": 100, + "biological_activation": "relu", + "projection_activation": "sigmoid", + "fusion_activation": "tanh" + }, + "BASE + notebook first layer": { + "optimizer": "Adam", + "learning_rate": 0.001, + "weight_decay": 0.0001, + "batch_size": 16, + "projection_dim": 64, + "epochs": 100, + "biological_activation": "relu", + "projection_activation": "sigmoid", + "fusion_activation": "tanh", + "first_layer": "dense_then_mask" + } +} \ No newline at end of file diff --git a/diagnostics/2026-11-06-test-split-reproduction/results/configs_scTrioseq2.json b/diagnostics/2026-11-06-test-split-reproduction/results/configs_scTrioseq2.json new file mode 100644 index 0000000000000000000000000000000000000000..7bdc0d71732c02d2c4a4404d0afb2402645cc3df --- /dev/null +++ b/diagnostics/2026-11-06-test-split-reproduction/results/configs_scTrioseq2.json @@ -0,0 +1,42 @@ +{ + "D default": {}, + "BASE lr1e-3 wd1e-4": { + "optimizer": "Adam", + "learning_rate": 0.001, + "weight_decay": 0.0001, + "batch_size": 16, + "projection_dim": 64, + "epochs": 100, + "biological_activation": "relu", + "projection_activation": "sigmoid", + "fusion_activation": "tanh" + }, + "D + notebook first layer": { + "first_layer": "dense_then_mask" + }, + "BASE + notebook first layer": { + "optimizer": "Adam", + "learning_rate": 0.001, + "weight_decay": 0.0001, + "batch_size": 16, + "projection_dim": 64, + "epochs": 100, + "biological_activation": "relu", + "projection_activation": "sigmoid", + "fusion_activation": "tanh", + "first_layer": "dense_then_mask" + }, + "Notebook replica (SGD + notebook first layer)": { + "optimizer": "SGD", + "learning_rate": 0.01, + "weight_decay": 0.01, + "batch_size": 16, + "projection_dim": 64, + "epochs": 200, + "sgd_momentum": 0.03, + "biological_activation": "relu", + "projection_activation": "sigmoid", + "fusion_activation": "tanh", + "first_layer": "dense_then_mask" + } +} \ No newline at end of file diff --git a/diagnostics/2026-11-06-test-split-reproduction/results/scTrioseq2.jsonl b/diagnostics/2026-11-06-test-split-reproduction/results/scTrioseq2.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..c7095ff009e75e34587849fafa56659994fabecd --- /dev/null +++ b/diagnostics/2026-11-06-test-split-reproduction/results/scTrioseq2.jsonl @@ -0,0 +1,50 @@ +{"config": "D default", "seed": 2, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.7295081967213115, "f1_macro": 0.751659043262817, "f1_weighted": 0.7209132775325167, "precision_macro": 0.7542335115864527, "recall_macro": 0.7680202351654362, "epochs_completed": 27, "pred_counts": [20, 30, 17, 55], "collapsed": false, "recall": [0.4666666666666667, 0.8695652173913043, 1.0, 0.7358490566037735]} +{"config": "D default", "seed": 3, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.7213114754098361, "f1_macro": 0.7484949406119036, "f1_weighted": 0.7177566794922314, "precision_macro": 0.7414701526543632, "recall_macro": 0.7910949890620728, "epochs_completed": 47, "pred_counts": [42, 28, 19, 33], "collapsed": false, "recall": [0.7666666666666667, 0.9130434782608695, 0.9375, 0.5471698113207547]} +{"config": "D default", "seed": 1, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.8114754098360656, "f1_macro": 0.7971620640960264, "f1_weighted": 0.8087276449401412, "precision_macro": 0.7991596638655463, "recall_macro": 0.8249000205086137, "epochs_completed": 61, "pred_counts": [20, 30, 21, 51], "collapsed": false, "recall": [0.6, 0.9130434782608695, 0.9375, 0.8490566037735849]} +{"config": "D default", "seed": 2024, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2024, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.7213114754098361, "f1_macro": 0.7317720932107724, "f1_weighted": 0.7134650413338938, "precision_macro": 0.7311580119515299, "recall_macro": 0.7476398003828274, "epochs_completed": 52, "pred_counts": [22, 31, 16, 53], "collapsed": false, "recall": [0.4666666666666667, 0.9130434782608695, 0.875, 0.7358490566037735]} +{"config": "D default", "seed": 7, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 7, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.7950819672131147, "f1_macro": 0.778657616892911, "f1_weighted": 0.7983210938852212, "precision_macro": 0.7820706993442054, "recall_macro": 0.805638159693738, "epochs_completed": 81, "pred_counts": [21, 29, 23, 49], "collapsed": false, "recall": [0.6666666666666666, 0.8695652173913043, 0.875, 0.8113207547169812]} +{"config": "D default", "seed": 5, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 5, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.7868852459016393, "f1_macro": 0.7766153569551627, "f1_weighted": 0.7869880421686882, "precision_macro": 0.7764099398055255, "recall_macro": 0.7973048263604047, "epochs_completed": 39, "pred_counts": [22, 31, 19, 50], "collapsed": false, "recall": [0.6333333333333333, 0.8695652173913043, 0.875, 0.8113207547169812]} +{"config": "D default", "seed": 0, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.7786885245901639, "f1_macro": 0.7876486408232388, "f1_weighted": 0.77160751735807, "precision_macro": 0.8224789915966386, "recall_macro": 0.8013356234618539, "epochs_completed": 87, "pred_counts": [16, 34, 16, 56], "collapsed": false, "recall": [0.5, 0.9565217391304348, 0.9375, 0.8113207547169812]} +{"config": "D default", "seed": 123, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 123, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.680327868852459, "f1_macro": 0.7098941376210925, "f1_weighted": 0.6814416995610624, "precision_macro": 0.7086953494848232, "recall_macro": 0.7225398208914411, "epochs_completed": 32, "pred_counts": [38, 27, 15, 42], "collapsed": false, "recall": [0.6666666666666666, 0.8260869565217391, 0.8125, 0.5849056603773585]} +{"config": "D default", "seed": 11, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 11, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.6557377049180327, "f1_macro": 0.7104243275746434, "f1_weighted": 0.6485566214320025, "precision_macro": 0.7165666710143037, "recall_macro": 0.7402908805031446, "epochs_completed": 31, "pred_counts": [41, 33, 14, 34], "collapsed": false, "recall": [0.6333333333333333, 1.0, 0.875, 0.4528301886792453]} +{"config": "D default", "seed": 42, "first_layer": "masked_weights", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.7704918032786885, "f1_macro": 0.766777219196574, "f1_weighted": 0.7756670733931442, "precision_macro": 0.7975836120401337, "recall_macro": 0.7746846800656275, "epochs_completed": 64, "pred_counts": [25, 39, 12, 46], "collapsed": false, "recall": [0.7, 0.9565217391304348, 0.6875, 0.7547169811320755]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 2, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.8360655737704918, "f1_macro": 0.8465899138034427, "f1_weighted": 0.8374058836416072, "precision_macro": 0.8387656810035842, "recall_macro": 0.8643654293136451, "epochs_completed": 76, "pred_counts": [31, 30, 16, 45], "collapsed": false, "recall": [0.8333333333333334, 0.9130434782608695, 0.9375, 0.7735849056603774]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 7, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 7, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.8032786885245902, "f1_macro": 0.7887728465829038, "f1_weighted": 0.8100378631478334, "precision_macro": 0.7981004901960785, "recall_macro": 0.7993881596937381, "epochs_completed": 61, "pred_counts": [24, 34, 16, 48], "collapsed": false, "recall": [0.7666666666666667, 0.8695652173913043, 0.75, 0.8113207547169812]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 1, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.819672131147541, "f1_macro": 0.8128122989815874, "f1_weighted": 0.8209569310943867, "precision_macro": 0.8216402116402116, "recall_macro": 0.8111002871205907, "epochs_completed": 78, "pred_counts": [25, 28, 15, 54], "collapsed": false, "recall": [0.7, 0.782608695652174, 0.875, 0.8867924528301887]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 3, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.8114754098360656, "f1_macro": 0.8254936813834444, "f1_weighted": 0.8134202540779337, "precision_macro": 0.8300575037707391, "recall_macro": 0.8270012988788624, "epochs_completed": 84, "pred_counts": [34, 26, 14, 48], "collapsed": false, "recall": [0.8333333333333334, 0.8260869565217391, 0.875, 0.7735849056603774]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 0, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.8278688524590164, "f1_macro": 0.8254499034809141, "f1_weighted": 0.8318520129154888, "precision_macro": 0.8311011904761906, "recall_macro": 0.8291435944763468, "epochs_completed": 86, "pred_counts": [30, 30, 14, 48], "collapsed": false, "recall": [0.8666666666666667, 0.8260869565217391, 0.8125, 0.8113207547169812]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 2024, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2024, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.8524590163934426, "f1_macro": 0.8502710561981078, "f1_weighted": 0.8535018471985935, "precision_macro": 0.841421568627451, "recall_macro": 0.8661906959256221, "epochs_completed": 51, "pred_counts": [36, 24, 17, 45], "collapsed": false, "recall": [0.9333333333333333, 0.782608695652174, 0.9375, 0.8113207547169812]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 11, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 11, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.8524590163934426, "f1_macro": 0.8528409432326406, "f1_weighted": 0.8543641825801677, "precision_macro": 0.864367816091954, "recall_macro": 0.851983353841947, "epochs_completed": 59, "pred_counts": [30, 29, 13, 50], "collapsed": false, "recall": [0.8333333333333334, 0.9130434782608695, 0.8125, 0.8490566037735849]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 5, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 5, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.8360655737704918, "f1_macro": 0.8323365068077934, "f1_weighted": 0.8402245058252327, "precision_macro": 0.8449978434087762, "recall_macro": 0.8302442234071643, "epochs_completed": 68, "pred_counts": [31, 29, 13, 49], "collapsed": false, "recall": [0.8333333333333334, 0.8260869565217391, 0.8125, 0.8490566037735849]} +{"config": "D + notebook first layer", "seed": 2, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.5, "f1_macro": 0.3043684710351377, "f1_weighted": 0.3892547444460013, "precision_macro": 0.24757543103448276, "recall_macro": 0.4292452830188679, "epochs_completed": 56, "pred_counts": [0, 58, 0, 64], "collapsed": false, "recall": [0.0, 1.0, 0.0, 0.7169811320754716]} +{"config": "D + notebook first layer", "seed": 7, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 7, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.4098360655737705, "f1_macro": 0.28710575139146566, "f1_weighted": 0.24812190151768604, "precision_macro": 0.20574618736383443, "recall_macro": 0.475, "epochs_completed": 46, "pred_counts": [68, 54, 0, 0], "collapsed": false, "recall": [0.9, 1.0, 0.0, 0.0]} +{"config": "D + notebook first layer", "seed": 2024, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2024, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.6065573770491803, "f1_macro": 0.5001879699248121, "f1_weighted": 0.5824109453962776, "precision_macro": 0.49989824989824994, "recall_macro": 0.6031446540880503, "epochs_completed": 17, "pred_counts": [26, 0, 54, 42], "collapsed": false, "recall": [0.7333333333333333, 0.0, 1.0, 0.6792452830188679]} +{"config": "D + notebook first layer", "seed": 3, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.36885245901639346, "f1_macro": 0.331726146351795, "f1_weighted": 0.3302520906283256, "precision_macro": 0.38831168831168833, "recall_macro": 0.38683688815969375, "epochs_completed": 21, "pred_counts": [70, 12, 22, 18], "collapsed": false, "recall": [0.8666666666666667, 0.2608695652173913, 0.25, 0.16981132075471697]} +{"config": "D + notebook first layer", "seed": 11, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 11, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.45901639344262296, "f1_macro": 0.2798764552150154, "f1_weighted": 0.3456917066862967, "precision_macro": 0.22039923434509162, "recall_macro": 0.4386792452830189, "epochs_completed": 18, "pred_counts": [0, 0, 53, 69], "collapsed": false, "recall": [0.0, 0.0, 1.0, 0.7547169811320755]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 123, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 123, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.860655737704918, "f1_macro": 0.8501400560224089, "f1_weighted": 0.8628828580612573, "precision_macro": 0.8678107606679036, "recall_macro": 0.8468724364232978, "epochs_completed": 100, "pred_counts": [33, 28, 12, 49], "collapsed": false, "recall": [0.9, 0.8695652173913043, 0.75, 0.8679245283018868]} +{"config": "D + notebook first layer", "seed": 5, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 5, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.3360655737704918, "f1_macro": 0.24116798846431145, "f1_weighted": 0.1869112484782583, "precision_macro": 0.16632928475033737, "recall_macro": 0.45833333333333337, "epochs_completed": 42, "pred_counts": [65, 0, 57, 0], "collapsed": false, "recall": [0.8333333333333334, 0.0, 1.0, 0.0]} +{"config": "D + notebook first layer", "seed": 1, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.6311475409836066, "f1_macro": 0.48859265734265733, "f1_weighted": 0.5962355267683136, "precision_macro": 0.534000442282176, "recall_macro": 0.5517295597484277, "epochs_completed": 27, "pred_counts": [14, 57, 0, 51], "collapsed": false, "recall": [0.43333333333333335, 1.0, 0.0, 0.7735849056603774]} +{"config": "BASE lr1e-3 wd1e-4", "seed": 42, "first_layer": "masked_weights", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.7950819672131147, "f1_macro": 0.7746201417268064, "f1_weighted": 0.7959649089672888, "precision_macro": 0.7945558176100629, "recall_macro": 0.7638492958709324, "epochs_completed": 75, "pred_counts": [32, 25, 12, 53], "collapsed": false, "recall": [0.8666666666666667, 0.6521739130434783, 0.6875, 0.8490566037735849]} +{"config": "BASE + notebook first layer", "seed": 2, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.9262295081967213, "f1_macro": 0.934510863857946, "f1_weighted": 0.9259998610184894, "precision_macro": 0.9304341491841492, "recall_macro": 0.9430227987421385, "epochs_completed": 40, "pred_counts": [33, 26, 15, 48], "collapsed": false, "recall": [0.9666666666666667, 1.0, 0.9375, 0.8679245283018868]} +{"config": "D + notebook first layer", "seed": 123, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 123, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.5901639344262295, "f1_macro": 0.36156427166493116, "f1_weighted": 0.45946402186551044, "precision_macro": 0.29186602870813394, "recall_macro": 0.4811320754716981, "epochs_completed": 32, "pred_counts": [0, 44, 2, 76], "collapsed": false, "recall": [0.0, 1.0, 0.0, 0.9245283018867925]} +{"config": "D + notebook first layer", "seed": 0, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.39344262295081966, "f1_macro": 0.3152363184079602, "f1_weighted": 0.3023917298752141, "precision_macro": 0.3602495543672014, "recall_macro": 0.4877358490566038, "epochs_completed": 23, "pred_counts": [60, 0, 51, 11], "collapsed": false, "recall": [0.8, 0.0, 1.0, 0.1509433962264151]} +{"config": "BASE + notebook first layer", "seed": 3, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.9016393442622951, "f1_macro": 0.9024180230925848, "f1_weighted": 0.9006265707258292, "precision_macro": 0.9156583988707401, "recall_macro": 0.8981678971834837, "epochs_completed": 26, "pred_counts": [29, 18, 18, 57], "collapsed": false, "recall": [0.8666666666666667, 0.782608695652174, 1.0, 0.9433962264150944]} +{"config": "BASE + notebook first layer", "seed": 7, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 7, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.9508196721311475, "f1_macro": 0.9549373433583961, "f1_weighted": 0.9512888779325361, "precision_macro": 0.9533475783475784, "recall_macro": 0.9608490566037735, "epochs_completed": 47, "pred_counts": [27, 27, 16, 52], "collapsed": false, "recall": [0.9, 1.0, 1.0, 0.9433962264150944]} +{"config": "D + notebook first layer", "seed": 42, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 200, "batch_size": 32, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "tanh", "projection_activation": "relu", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.6065573770491803, "f1_macro": 0.47953947368421057, "f1_weighted": 0.5700251653724475, "precision_macro": 0.46565874908898164, "recall_macro": 0.5520440251572327, "epochs_completed": 26, "pred_counts": [27, 52, 0, 43], "collapsed": false, "recall": [0.5666666666666667, 1.0, 0.0, 0.6415094339622641]} +{"config": "BASE + notebook first layer", "seed": 1, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.9262295081967213, "f1_macro": 0.9309260862039122, "f1_weighted": 0.9245415256089013, "precision_macro": 0.9406415343915344, "recall_macro": 0.9322327044025157, "epochs_completed": 25, "pred_counts": [23, 27, 16, 56], "collapsed": false, "recall": [0.7666666666666667, 1.0, 1.0, 0.9622641509433962]} +{"config": "BASE + notebook first layer", "seed": 5, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 5, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.9426229508196722, "f1_macro": 0.9456872069216036, "f1_weighted": 0.943280702381223, "precision_macro": 0.9398717948717948, "recall_macro": 0.9535958435876402, "epochs_completed": 32, "pred_counts": [30, 26, 16, 50], "collapsed": false, "recall": [0.9333333333333333, 0.9565217391304348, 1.0, 0.9245283018867925]} +{"config": "BASE + notebook first layer", "seed": 11, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 11, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.9508196721311475, "f1_macro": 0.9530465949820788, "f1_weighted": 0.9508696163111816, "precision_macro": 0.9621212121212122, "recall_macro": 0.944868573967733, "epochs_completed": 31, "pred_counts": [30, 22, 15, 55], "collapsed": false, "recall": [0.9666666666666667, 0.9130434782608695, 0.9375, 0.9622641509433962]} +{"config": "BASE + notebook first layer", "seed": 0, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.9262295081967213, "f1_macro": 0.9216397849462366, "f1_weighted": 0.9262427287149656, "precision_macro": 0.9222435897435899, "recall_macro": 0.9223458435876402, "epochs_completed": 31, "pred_counts": [30, 25, 15, 52], "collapsed": false, "recall": [0.9333333333333333, 0.9565217391304348, 0.875, 0.9245283018867925]} +{"config": "BASE + notebook first layer", "seed": 2024, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 2024, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.9180327868852459, "f1_macro": 0.9238131313131313, "f1_weighted": 0.9182389468455042, "precision_macro": 0.9120961808321008, "recall_macro": 0.9394449001914138, "epochs_completed": 51, "pred_counts": [33, 25, 17, 47], "collapsed": false, "recall": [0.9333333333333333, 0.9565217391304348, 1.0, 0.8679245283018868]} +{"config": "BASE + notebook first layer", "seed": 123, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 123, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.9344262295081968, "f1_macro": 0.9315744828426548, "f1_weighted": 0.9342755694542849, "precision_macro": 0.9316973289620168, "recall_macro": 0.9329573420836751, "epochs_completed": 32, "pred_counts": [31, 21, 17, 53], "collapsed": false, "recall": [0.9, 0.8695652173913043, 1.0, 0.9622641509433962]} +{"config": "BASE + notebook first layer", "seed": 42, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 100, "batch_size": 16, "learning_rate": 0.001, "weight_decay": 0.0001, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "Adam", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.9754098360655737, "f1_macro": 0.9745179915455202, "f1_weighted": 0.975203798262116, "precision_macro": 0.975854866375256, "recall_macro": 0.973543888433142, "epochs_completed": 29, "pred_counts": [31, 22, 16, 53], "collapsed": false, "recall": [1.0, 0.9130434782608695, 1.0, 0.9811320754716981]} +{"config": "Notebook replica (SGD + notebook first layer)", "seed": 7, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 7, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.9016393442622951, "f1_macro": 0.8995659981130353, "f1_weighted": 0.9020568713805527, "precision_macro": 0.8914347961406784, "recall_macro": 0.9118702146568226, "epochs_completed": 99, "pred_counts": [28, 27, 17, 50], "collapsed": false, "recall": [0.8666666666666667, 0.9565217391304348, 0.9375, 0.8867924528301887]} +{"config": "Notebook replica (SGD + notebook first layer)", "seed": 1, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 1, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.9016393442622951, "f1_macro": 0.9066064754138148, "f1_weighted": 0.9007679875479544, "precision_macro": 0.9147206959706959, "recall_macro": 0.9057765928356576, "epochs_completed": 85, "pred_counts": [24, 26, 16, 56], "collapsed": false, "recall": [0.7666666666666667, 0.9130434782608695, 1.0, 0.9433962264150944]} +{"config": "Notebook replica (SGD + notebook first layer)", "seed": 0, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 0, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.8688524590163934, "f1_macro": 0.8722173865738222, "f1_weighted": 0.8687099461162113, "precision_macro": 0.8601173856391642, "recall_macro": 0.8966776045939295, "epochs_completed": 86, "pred_counts": [26, 29, 19, 48], "collapsed": false, "recall": [0.8, 0.9565217391304348, 1.0, 0.8301886792452831]} +{"config": "Notebook replica (SGD + notebook first layer)", "seed": 3, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 3, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.8770491803278688, "f1_macro": 0.8735930735930736, "f1_weighted": 0.8761833794620679, "precision_macro": 0.8674402352033931, "recall_macro": 0.8840169537872573, "epochs_completed": 156, "pred_counts": [30, 21, 19, 52], "collapsed": false, "recall": [0.8666666666666667, 0.782608695652174, 1.0, 0.8867924528301887]} +{"config": "Notebook replica (SGD + notebook first layer)", "seed": 2, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 2, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.8934426229508197, "f1_macro": 0.9055378017907891, "f1_weighted": 0.8929186467805367, "precision_macro": 0.8948795035255481, "recall_macro": 0.9242138364779875, "epochs_completed": 142, "pred_counts": [28, 29, 17, 48], "collapsed": false, "recall": [0.8666666666666667, 1.0, 1.0, 0.8301886792452831]} +{"config": "Notebook replica (SGD + notebook first layer)", "seed": 123, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 123, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.9508196721311475, "f1_macro": 0.9431289058097159, "f1_weighted": 0.9506411384274704, "precision_macro": 0.9393807357517034, "recall_macro": 0.9496240087503418, "epochs_completed": 108, "pred_counts": [31, 21, 18, 52], "collapsed": false, "recall": [0.9666666666666667, 0.8695652173913043, 1.0, 0.9622641509433962]} +{"config": "Notebook replica (SGD + notebook first layer)", "seed": 11, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 11, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.9180327868852459, "f1_macro": 0.9179947578647704, "f1_weighted": 0.9197812377898781, "precision_macro": 0.9228418953788412, "recall_macro": 0.9237814465408806, "epochs_completed": 141, "pred_counts": [29, 30, 14, 49], "collapsed": false, "recall": [0.9333333333333333, 1.0, 0.875, 0.8867924528301887]} +{"config": "Notebook replica (SGD + notebook first layer)", "seed": 2024, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 2024, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.9262295081967213, "f1_macro": 0.9265564640480441, "f1_weighted": 0.9264566908182353, "precision_macro": 0.9179360812425328, "recall_macro": 0.9394064465408806, "epochs_completed": 101, "pred_counts": [31, 27, 16, 48], "collapsed": false, "recall": [0.9333333333333333, 1.0, 0.9375, 0.8867924528301887]} +{"config": "Notebook replica (SGD + notebook first layer)", "seed": 42, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 42, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.9508196721311475, "f1_macro": 0.9465219993710118, "f1_weighted": 0.9522703232470696, "precision_macro": 0.9389978213507626, "recall_macro": 0.9583128247197157, "epochs_completed": 134, "pred_counts": [28, 27, 17, 50], "collapsed": false, "recall": [0.9333333333333333, 0.9565217391304348, 1.0, 0.9433962264150944]} +{"config": "Notebook replica (SGD + notebook first layer)", "seed": 5, "first_layer": "dense_then_mask", "hyperparameters": {"epochs": 200, "batch_size": 16, "learning_rate": 0.01, "weight_decay": 0.01, "dropout": 0.3, "projection_dim": 64, "fusion_dim": 12, "test_fraction": 0.2, "early_stopping_fraction": 0.125, "optimizer": "SGD", "class_weighting": true, "early_stopping_patience": 12, "seed": 5, "biological_activation": "relu", "projection_activation": "sigmoid", "fusion_activation": "tanh", "sgd_momentum": 0.03}, "split": "test", "test_samples": 122, "accuracy": 0.8934426229508197, "f1_macro": 0.9034802043422733, "f1_weighted": 0.8947103406402446, "precision_macro": 0.8980598588096873, "recall_macro": 0.9205974842767296, "epochs_completed": 130, "pred_counts": [28, 31, 16, 47], "collapsed": false, "recall": [0.8333333333333334, 1.0, 1.0, 0.8490566037735849]} diff --git a/diagnostics/2026-11-06-test-split-reproduction/results/scTrioseq2_shard0.log b/diagnostics/2026-11-06-test-split-reproduction/results/scTrioseq2_shard0.log new file mode 100644 index 0000000000000000000000000000000000000000..be52a0ccf012902ce98a59ddb26bc2e372f6818d --- /dev/null +++ b/diagnostics/2026-11-06-test-split-reproduction/results/scTrioseq2_shard0.log @@ -0,0 +1,11 @@ +[0] D default seed=0 acc=0.779 f1w=0.772 +[0] D default seed=42 acc=0.770 f1w=0.776 +[0] BASE lr1e-3 wd1e-4 seed=0 acc=0.828 f1w=0.832 +[0] BASE lr1e-3 wd1e-4 seed=42 acc=0.795 f1w=0.796 +[0] D + notebook first layer seed=0 acc=0.393 f1w=0.302 +[0] D + notebook first layer seed=42 acc=0.607 f1w=0.570 +[0] BASE + notebook first layer seed=0 acc=0.926 f1w=0.926 +[0] BASE + notebook first layer seed=42 acc=0.975 f1w=0.975 +[0] Notebook replica (SGD + notebook first layer) seed=0 acc=0.869 f1w=0.869 +[0] Notebook replica (SGD + notebook first layer) seed=42 acc=0.951 f1w=0.952 +SHARD 0 DONE diff --git a/diagnostics/2026-11-06-test-split-reproduction/results/scTrioseq2_shard1.log b/diagnostics/2026-11-06-test-split-reproduction/results/scTrioseq2_shard1.log new file mode 100644 index 0000000000000000000000000000000000000000..888a95b06ffd1b19114a5af9988d4a3de2705461 --- /dev/null +++ b/diagnostics/2026-11-06-test-split-reproduction/results/scTrioseq2_shard1.log @@ -0,0 +1,11 @@ +[1] D default seed=1 acc=0.811 f1w=0.809 +[1] D default seed=123 acc=0.680 f1w=0.681 +[1] BASE lr1e-3 wd1e-4 seed=1 acc=0.820 f1w=0.821 +[1] BASE lr1e-3 wd1e-4 seed=123 acc=0.861 f1w=0.863 +[1] D + notebook first layer seed=1 acc=0.631 f1w=0.596 +[1] D + notebook first layer seed=123 acc=0.590 f1w=0.459 +[1] BASE + notebook first layer seed=1 acc=0.926 f1w=0.925 +[1] BASE + notebook first layer seed=123 acc=0.934 f1w=0.934 +[1] Notebook replica (SGD + notebook first layer) seed=1 acc=0.902 f1w=0.901 +[1] Notebook replica (SGD + notebook first layer) seed=123 acc=0.951 f1w=0.951 +SHARD 1 DONE diff --git a/diagnostics/2026-11-06-test-split-reproduction/results/scTrioseq2_shard2.log b/diagnostics/2026-11-06-test-split-reproduction/results/scTrioseq2_shard2.log new file mode 100644 index 0000000000000000000000000000000000000000..f26b6817d3e40b68218d90066d9b330648eb5020 --- /dev/null +++ b/diagnostics/2026-11-06-test-split-reproduction/results/scTrioseq2_shard2.log @@ -0,0 +1,11 @@ +[2] D default seed=2 acc=0.730 f1w=0.721 +[2] D default seed=2024 acc=0.721 f1w=0.713 +[2] BASE lr1e-3 wd1e-4 seed=2 acc=0.836 f1w=0.837 +[2] BASE lr1e-3 wd1e-4 seed=2024 acc=0.852 f1w=0.854 +[2] D + notebook first layer seed=2 acc=0.500 f1w=0.389 +[2] D + notebook first layer seed=2024 acc=0.607 f1w=0.582 +[2] BASE + notebook first layer seed=2 acc=0.926 f1w=0.926 +[2] BASE + notebook first layer seed=2024 acc=0.918 f1w=0.918 +[2] Notebook replica (SGD + notebook first layer) seed=2 acc=0.893 f1w=0.893 +[2] Notebook replica (SGD + notebook first layer) seed=2024 acc=0.926 f1w=0.926 +SHARD 2 DONE diff --git a/diagnostics/2026-11-06-test-split-reproduction/results/scTrioseq2_shard3.log b/diagnostics/2026-11-06-test-split-reproduction/results/scTrioseq2_shard3.log new file mode 100644 index 0000000000000000000000000000000000000000..b261044e66ae1b2461ae602f771c10359d64fc7c --- /dev/null +++ b/diagnostics/2026-11-06-test-split-reproduction/results/scTrioseq2_shard3.log @@ -0,0 +1,11 @@ +[3] D default seed=3 acc=0.721 f1w=0.718 +[3] D default seed=5 acc=0.787 f1w=0.787 +[3] BASE lr1e-3 wd1e-4 seed=3 acc=0.811 f1w=0.813 +[3] BASE lr1e-3 wd1e-4 seed=5 acc=0.836 f1w=0.840 +[3] D + notebook first layer seed=3 acc=0.369 f1w=0.330 +[3] D + notebook first layer seed=5 acc=0.336 f1w=0.187 +[3] BASE + notebook first layer seed=3 acc=0.902 f1w=0.901 +[3] BASE + notebook first layer seed=5 acc=0.943 f1w=0.943 +[3] Notebook replica (SGD + notebook first layer) seed=3 acc=0.877 f1w=0.876 +[3] Notebook replica (SGD + notebook first layer) seed=5 acc=0.893 f1w=0.895 +SHARD 3 DONE diff --git a/diagnostics/2026-11-06-test-split-reproduction/results/scTrioseq2_shard4.log b/diagnostics/2026-11-06-test-split-reproduction/results/scTrioseq2_shard4.log new file mode 100644 index 0000000000000000000000000000000000000000..7e067d8b307a8a444a672570015d15198915ad23 --- /dev/null +++ b/diagnostics/2026-11-06-test-split-reproduction/results/scTrioseq2_shard4.log @@ -0,0 +1,11 @@ +[4] D default seed=7 acc=0.795 f1w=0.798 +[4] D default seed=11 acc=0.656 f1w=0.649 +[4] BASE lr1e-3 wd1e-4 seed=7 acc=0.803 f1w=0.810 +[4] BASE lr1e-3 wd1e-4 seed=11 acc=0.852 f1w=0.854 +[4] D + notebook first layer seed=7 acc=0.410 f1w=0.248 +[4] D + notebook first layer seed=11 acc=0.459 f1w=0.346 +[4] BASE + notebook first layer seed=7 acc=0.951 f1w=0.951 +[4] BASE + notebook first layer seed=11 acc=0.951 f1w=0.951 +[4] Notebook replica (SGD + notebook first layer) seed=7 acc=0.902 f1w=0.902 +[4] Notebook replica (SGD + notebook first layer) seed=11 acc=0.918 f1w=0.920 +SHARD 4 DONE diff --git a/diagnostics/2026-11-06-test-split-reproduction/results/scTrioseq2_summary.txt b/diagnostics/2026-11-06-test-split-reproduction/results/scTrioseq2_summary.txt new file mode 100644 index 0000000000000000000000000000000000000000..cecee9e16b0ff6096644082ddd02e19e6b24057e --- /dev/null +++ b/diagnostics/2026-11-06-test-split-reproduction/results/scTrioseq2_summary.txt @@ -0,0 +1,9 @@ + +=== scTrioseq2 (paper: accuracy 0.95, F1 0.95, precision 0.95, recall 0.96) + seeds accuracy macro F1 weighted F1 precision recall macro F1 min/max collapsed +config +BASE + notebook first layer 10 0.935 ± 0.021 0.937 ± 0.020 0.935 ± 0.021 0.938 ± 0.020 0.940 ± 0.021 0.902 / 0.975 0 +Notebook replica (SGD + notebook first layer) 10 0.908 ± 0.028 0.910 ± 0.025 0.908 ± 0.028 0.905 ± 0.027 0.921 ± 0.023 0.872 / 0.947 0 +BASE lr1e-3 wd1e-4 10 0.830 ± 0.022 0.826 ± 0.027 0.832 ± 0.022 0.833 ± 0.024 0.829 ± 0.031 0.775 / 0.853 0 +D default 10 0.745 ± 0.052 0.756 ± 0.031 0.742 ± 0.054 0.763 ± 0.038 0.777 ± 0.033 0.710 / 0.797 0 +D + notebook first layer 10 0.490 ± 0.112 0.359 ± 0.096 0.401 ± 0.145 0.338 ± 0.131 0.486 ± 0.065 0.241 / 0.500 0 diff --git a/diagnostics/2026-11-06-test-split-reproduction/scripts/run_seeds.py b/diagnostics/2026-11-06-test-split-reproduction/scripts/run_seeds.py new file mode 100644 index 0000000000000000000000000000000000000000..4ac7cf1cfe94855ae221d997011a6f3ad07c5f93 --- /dev/null +++ b/diagnostics/2026-11-06-test-split-reproduction/scripts/run_seeds.py @@ -0,0 +1,92 @@ +"""Diagnostic seed runs on the 80/20 train/test split (scores from the untouched test set). + +Usage: run_seeds.py [seeds] + +configs.json maps a name to Hyperparameters overrides. A config may set +"first_layer": "dense_then_mask" to swap in the upstream notebook's first layer +(ReLU(x W + b) followed by a fixed multiplication by the biological mask) for this +run only. biolmnet/model.py is not modified; the swap happens in memory here. +Resumable: (config, seed) pairs already in out.jsonl are skipped. +""" +import json +import math +import os +import pickle +import sys +from dataclasses import asdict, replace +from pathlib import Path + +import numpy as np +import torch +from sklearn.metrics import recall_score +from torch import nn + +import biolmnet.training as training + +WORKSPACE, CONFIGS, OUT = Path(sys.argv[1]), json.loads(Path(sys.argv[2]).read_text()), Path(sys.argv[3]) +SHARD, SHARDS = int(sys.argv[4]), int(sys.argv[5]) +SEEDS = [int(x) for x in sys.argv[6].split(",")] if len(sys.argv) > 6 else [0, 1, 2, 3, 7, 42, 123, 2024, 5, 11] +torch.set_num_threads(int(os.environ.get("THREADS", "2"))) + + +class DenseThenMask(nn.Module): + """Upstream notebook first layer: ReLU(x W + b) @ M, with M fixed (not trained).""" + + def __init__(self, mask_t: torch.Tensor): + super().__init__() + n_inputs = mask_t.shape[1] # BioMaskedLinear stores mask as [hidden, inputs] + self.dense = nn.Linear(n_inputs, n_inputs) + bound = math.sqrt(3.0 / n_inputs) # LeCun uniform, as in the notebook + nn.init.uniform_(self.dense.weight, -bound, bound) + nn.init.zeros_(self.dense.bias) + self.register_buffer("mask", mask_t.T.contiguous()) # [inputs, hidden] + + def forward(self, inputs): + return torch.relu(self.dense(inputs)) @ self.mask + + +original_make_model = training._make_model +workspace = pickle.loads(WORKSPACE.read_bytes()) +names = workspace.label_names +done = set() +if OUT.exists(): + done = {(r["config"], r["seed"]) for r in map(json.loads, OUT.read_text().splitlines()) if r} + +jobs = [(c, s) for c in CONFIGS for s in SEEDS] +for index, (config, seed) in enumerate(jobs): + if index % SHARDS != SHARD or (config, seed) in done: + continue + overrides = dict(CONFIGS[config]) + first_layer = overrides.pop("first_layer", "masked_weights") + + def make_model(ws, hp, _variant=first_layer): + model = original_make_model(ws, hp) + if _variant == "dense_then_mask": + for branch in (model.gene_branch, model.dna_branch): + branch.biological = DenseThenMask(branch.biological.mask) + branch.biological_activation = nn.Identity() # ReLU already applied before the mask + return model + + training._make_model = make_model + hp = replace(training.Hyperparameters(), seed=seed, **overrides) + try: + result = training.train(workspace, hp) + except Exception as exc: + row = {"config": config, "seed": seed, "error": repr(exc)} + else: + m = result.bundle.metrics + vp = result.test_predictions + counts = vp["predicted"].value_counts().reindex(names, fill_value=0).astype(int).tolist() + row = { + "config": config, "seed": seed, "first_layer": first_layer, "hyperparameters": asdict(hp), + "split": m["evaluation_split"], "test_samples": m["test_samples"], + "accuracy": m["accuracy"], "f1_macro": m["f1_macro"], "f1_weighted": m["f1_weighted"], + "precision_macro": m["precision_macro"], "recall_macro": m["recall_macro"], + "epochs_completed": m["epochs_completed"], + "pred_counts": counts, "collapsed": int((np.array(counts) > 0).sum()) == 1, + "recall": recall_score(vp["observed"], vp["predicted"], labels=names, average=None, zero_division=0).tolist(), + } + with OUT.open("a") as handle: + handle.write(json.dumps(row) + "\n") + print(f"[{SHARD}] {config} seed={seed} " + (row.get("error") or f"acc={row['accuracy']:.3f} f1w={row['f1_weighted']:.3f}"), flush=True) +print(f"SHARD {SHARD} DONE", flush=True) diff --git a/diagnostics/2026-11-06-test-split-reproduction/scripts/summary.py b/diagnostics/2026-11-06-test-split-reproduction/scripts/summary.py new file mode 100644 index 0000000000000000000000000000000000000000..d065af45b8795c4d204163253f3adab80dbcd747 --- /dev/null +++ b/diagnostics/2026-11-06-test-split-reproduction/scripts/summary.py @@ -0,0 +1,46 @@ +"""Summarise run_seeds.py output (test-set metrics, mean ± std over seeds). +Usage: summary.py [ ...]""" +import json +import sys +from pathlib import Path + +import pandas as pd + +# BioLM-NET results reported in the PSB 2026 paper (80/20 split, mean of 10 test runs). +PAPER = { + "scTrioseq2": "paper: accuracy 0.95, F1 0.95, precision 0.95, recall 0.96", + "BRCA": "paper: F1 0.82", + "COAD": "paper: accuracy 0.87, recall 0.80", + "GBM": "paper: deep models underperformed classical ones (no number in text)", +} + +pd.set_option("display.width", 250) +pd.set_option("display.max_colwidth", 60) +for path in sys.argv[1:]: + rows = [json.loads(line) for line in open(path)] + errors = [r for r in rows if "error" in r] + df = pd.DataFrame([r for r in rows if "error" not in r]) + dataset = Path(path).stem + print(f"\n=== {dataset} ({PAPER.get(dataset, '')})") + if df.empty: + print("no finished runs") + continue + g = df.groupby("config") + + def fmt(col): + return g[col].mean().map("{:.3f}".format) + " ± " + g[col].std().map("{:.3f}".format) + + summary = pd.DataFrame({ + "seeds": g.size(), + "accuracy": fmt("accuracy"), + "macro F1": fmt("f1_macro"), + "weighted F1": fmt("f1_weighted"), + "precision": fmt("precision_macro"), + "recall": fmt("recall_macro"), + "macro F1 min/max": g.f1_macro.min().map("{:.3f}".format) + " / " + g.f1_macro.max().map("{:.3f}".format), + "collapsed": g.collapsed.sum(), + "_sort": g.f1_macro.mean(), + }).sort_values("_sort", ascending=False).drop(columns="_sort") + print(summary.to_string()) + for r in errors: + print("ERROR", r["config"], r["seed"], r["error"][:150]) diff --git a/tests/test_core.py b/tests/test_core.py index 1568f3a91fbb967a124aa08961a747143d97cfd2..f8e2792fca7c062e7c7c1c048b456bd0d1845f11 100644 --- a/tests/test_core.py +++ b/tests/test_core.py @@ -18,10 +18,12 @@ from biolmnet.data import ( ) from biolmnet.model import BioMaskedLinear, BioLMNet from biolmnet.training import ( - DEFAULT_VALIDATION_FRACTION, + DEFAULT_TEST_FRACTION, + EARLY_STOPPING_FRACTION, Hyperparameters, predict, - split_train_validation, + split_train_validation_test, + stratified_split, train, ) @@ -252,102 +254,116 @@ _FAST = Hyperparameters( ) -def test_default_split_is_75_training_25_validation() -> None: - assert DEFAULT_VALIDATION_FRACTION == 0.25 - assert Hyperparameters().validation_fraction == 0.25 - labels = np.repeat(np.array([0, 1]), 50) - train_index, validation_index = split_train_validation( - labels, Hyperparameters().validation_fraction, Hyperparameters().seed - ) - assert len(train_index) == 75 - assert len(validation_index) == 25 +def _three_way(labels, hyperparameters=None): + hp = hyperparameters or Hyperparameters() + return split_train_validation_test(labels, hp.test_fraction, hp.early_stopping_fraction, hp.seed) + + +def test_default_split_is_80_training_20_test() -> None: + assert DEFAULT_TEST_FRACTION == 0.2 + assert EARLY_STOPPING_FRACTION == 0.125 + labels = np.repeat(np.array([0, 1]), 100) + fit_index, validation_index, test_index = _three_way(labels) + assert len(test_index) == 40 # 20% untouched test + assert len(fit_index) + len(validation_index) == 160 # 80% training part + assert len(validation_index) == 20 # 1/8 of the training part = 10% of all samples def test_split_partitions_every_sample_exactly_once() -> None: labels = np.repeat(np.array([0, 1, 2]), [40, 25, 18]) - train_index, validation_index = split_train_validation(labels, 0.25, 42) - assert len(train_index) + len(validation_index) == len(labels) - assert set(train_index).isdisjoint(validation_index) - assert sorted(np.concatenate([train_index, validation_index])) == list(range(len(labels))) + parts = _three_way(labels) + assert sum(len(part) for part in parts) == len(labels) + assert set(parts[0]).isdisjoint(parts[1]) + assert set(parts[0]).isdisjoint(parts[2]) + assert set(parts[1]).isdisjoint(parts[2]) + assert sorted(np.concatenate(parts)) == list(range(len(labels))) def test_split_preserves_class_proportions() -> None: - labels = np.repeat(np.array([0, 1, 2]), [120, 60, 20]) # 60 / 30 / 10 - train_index, validation_index = split_train_validation(labels, 0.25, 42) + labels = np.repeat(np.array([0, 1, 2]), [240, 120, 40]) # 60 / 30 / 10 expected = np.array([0.6, 0.3, 0.1]) - for partition in (train_index, validation_index): + for partition in _three_way(labels): observed = np.bincount(labels[partition], minlength=3) / len(partition) - np.testing.assert_allclose(observed, expected, atol=0.02) + np.testing.assert_allclose(observed, expected, atol=0.03) def test_split_is_reproducible_for_a_fixed_seed() -> None: - labels = np.repeat(np.array([0, 1]), [30, 34]) - first = split_train_validation(labels, 0.25, 42) - second = split_train_validation(labels, 0.25, 42) - other_seed = split_train_validation(labels, 0.25, 7) - np.testing.assert_array_equal(first[0], second[0]) - np.testing.assert_array_equal(first[1], second[1]) - assert set(first[1]) != set(other_seed[1]) + labels = np.repeat(np.array([0, 1]), [60, 68]) + first = _three_way(labels) + second = _three_way(labels) + other_seed = split_train_validation_test(labels, 0.2, 0.125, 7) + for a, b in zip(first, second): + np.testing.assert_array_equal(a, b) + assert set(first[2]) != set(other_seed[2]) def test_split_rejects_partitions_too_small_to_stratify() -> None: with pytest.raises(ValueError, match="at least two samples"): - split_train_validation(np.array([0, 0, 0, 1]), 0.25, 42) - # 4 classes x 2 samples: 25% leaves 2 validation samples for 4 classes. + stratified_split(np.array([0, 0, 0, 1]), 0.2, 42) + # 4 classes x 2 samples: 20% leaves 2 test samples for 4 classes. with pytest.raises(ValueError, match="one sample per class"): - split_train_validation(np.repeat(np.arange(4), 2), 0.25, 42) + stratified_split(np.repeat(np.arange(4), 2), 0.2, 42) -def test_training_uses_paired_rows_and_train_only_scaler() -> None: +def test_metrics_come_from_untouched_test_rows_with_train_only_scaler() -> None: workspace = _tiny_workspace() - hyperparameters = replace(_FAST, validation_fraction=0.25) - train_index, validation_index = split_train_validation( - workspace.labels, hyperparameters.validation_fraction, hyperparameters.seed - ) - # Shift only the validation rows: a scaler fitted on all samples would move. - workspace.gene_expression[validation_index] += 100.0 - workspace.dna_methylation[validation_index] -= 100.0 + fit_index, validation_index, test_index = _three_way(workspace.labels, _FAST) + # Shift the held-out rows: a scaler fitted on anything but the training rows would move. + held_out = np.concatenate([validation_index, test_index]) + workspace.gene_expression[held_out] += 100.0 + workspace.dna_methylation[held_out] -= 100.0 - result = train(workspace, hyperparameters) + result = train(workspace, _FAST) bundle = result.bundle - assert bundle.metrics["training_samples"] == 36 - assert bundle.metrics["validation_samples"] == 12 - np.testing.assert_array_equal( - result.validation_predictions["sample_row"].to_numpy(), validation_index + assert bundle.metrics["evaluation_split"] == "test" + assert (bundle.metrics["training_samples"], bundle.metrics["validation_samples"], bundle.metrics["test_samples"]) == ( + 33, + 5, + 10, ) - expected_labels = [workspace.label_names[value] for value in workspace.labels[validation_index]] - assert result.validation_predictions["observed"].tolist() == expected_labels + # Reported predictions are exactly the test rows, with their own labels. + np.testing.assert_array_equal(result.test_predictions["sample_row"].to_numpy(), test_index) + expected_labels = [workspace.label_names[value] for value in workspace.labels[test_index]] + assert result.test_predictions["observed"].tolist() == expected_labels # Both omics scalers come from the same training rows only. - np.testing.assert_allclose( - bundle.gene_mean, workspace.gene_expression[train_index].mean(axis=0), rtol=1e-5 - ) - np.testing.assert_allclose( - bundle.gene_scale, workspace.gene_expression[train_index].std(axis=0), rtol=1e-5 - ) - np.testing.assert_allclose( - bundle.dna_mean, workspace.dna_methylation[train_index].mean(axis=0), rtol=1e-5 - ) - np.testing.assert_allclose( - bundle.dna_scale, workspace.dna_methylation[train_index].std(axis=0), rtol=1e-5 - ) + np.testing.assert_allclose(bundle.gene_mean, workspace.gene_expression[fit_index].mean(axis=0), rtol=1e-5) + np.testing.assert_allclose(bundle.gene_scale, workspace.gene_expression[fit_index].std(axis=0), rtol=1e-5) + np.testing.assert_allclose(bundle.dna_mean, workspace.dna_methylation[fit_index].mean(axis=0), rtol=1e-5) + np.testing.assert_allclose(bundle.dna_scale, workspace.dna_methylation[fit_index].std(axis=0), rtol=1e-5) - # Validation rows are transformed with that training-fitted scaler: the - # stored validation probabilities match predict(), which uses the bundle's scaler. + # Test rows are transformed with that training-fitted scaler: the stored test + # probabilities match predict(), which uses the bundle's scaler. rescored = predict( - pd.DataFrame(workspace.gene_expression[validation_index], columns=bundle.gene_features), - pd.DataFrame(workspace.dna_methylation[validation_index], columns=bundle.dna_features), + pd.DataFrame(workspace.gene_expression[test_index], columns=bundle.gene_features), + pd.DataFrame(workspace.dna_methylation[test_index], columns=bundle.dna_features), bundle, ) probability_columns = [f"P({label})" for label in workspace.label_names] np.testing.assert_allclose( - result.validation_predictions[probability_columns].to_numpy(), + result.test_predictions[probability_columns].to_numpy(), rescored[probability_columns].to_numpy(), atol=1e-5, ) +def test_test_set_never_influences_training_or_checkpoint() -> None: + clean = _tiny_workspace() + _, _, test_index = _three_way(clean.labels, _FAST) + corrupted = _tiny_workspace() + corrupted.gene_expression[test_index] = 1e3 + corrupted.dna_methylation[test_index] = -1e3 + corrupted.labels = corrupted.labels.copy() + + first = train(clean, _FAST).bundle + second = train(corrupted, _FAST).bundle + # Same loss curve, same early-stopping decisions, same weights: the test rows played no part. + assert first.history == second.history + for key, value in first.model.state_dict().items(): + assert torch.equal(value, second.model.state_dict()[key]), key + + def test_class_weights_use_training_labels_only(monkeypatch) -> None: import biolmnet.training as training @@ -361,16 +377,17 @@ def test_class_weights_use_training_labels_only(monkeypatch) -> None: monkeypatch.setattr(training.nn, "CrossEntropyLoss", spy) workspace = _tiny_workspace() # 24 control / 24 case: all-sample weights are [1, 1]. # Force a lopsided split so training-only weights clearly differ from all-sample ones. - train_index = np.concatenate([np.arange(0, 22), np.arange(24, 30)]) - validation_index = np.setdiff1d(np.arange(48), train_index) + fit_index = np.concatenate([np.arange(0, 22), np.arange(24, 30)]) + rest = np.setdiff1d(np.arange(48), fit_index) + validation_index, test_index = rest[::2], rest[1::2] monkeypatch.setattr( - training, "split_train_validation", lambda *args: (train_index, validation_index) + training, "split_train_validation_test", lambda *args: (fit_index, validation_index, test_index) ) train(workspace, _FAST) - counts = np.bincount(workspace.labels[train_index], minlength=2).astype(np.float32) + counts = np.bincount(workspace.labels[fit_index], minlength=2).astype(np.float32) assert counts.tolist() == [22.0, 6.0] - np.testing.assert_allclose(captured["weight"], len(train_index) / (2 * counts), rtol=1e-6) + np.testing.assert_allclose(captured["weight"], len(fit_index) / (2 * counts), rtol=1e-6) def test_results_page_receives_training_history_plot() -> None: @@ -381,16 +398,41 @@ def test_results_page_receives_training_history_plot() -> None: result = train(_tiny_workspace(), _FAST) figure = refresh_results(result.bundle, {})[-1] traces = {trace.name: trace for trace in figure.data} - assert set(traces) == {"Training", "Validation"} + assert set(traces) == {"Training", "Validation (early stopping)"} history = result.bundle.history assert list(traces["Training"].y) == [row["training_loss"] for row in history] - assert list(traces["Validation"].y) == [row["validation_loss"] for row in history] + assert list(traces["Validation (early stopping)"].y) == [row["validation_loss"] for row in history] # Artifacts saved without history still render the rest of the page. result.bundle.history = [] assert refresh_results(result.bundle, {})[-1] is None +def test_results_labels_say_test_for_new_models_and_validation_for_old_artifacts() -> None: + from app import refresh_results + + result = train(_tiny_workspace(), _FAST) + stats_html = refresh_results(result.bundle, {"macro_auc": 0.9})[0] + for label in ("Test accuracy", "Test macro F1", "Test macro AUC"): + assert label in stats_html + + # An artifact saved before the test split has no evaluation_split: keep its honest label. + del result.bundle.metrics["evaluation_split"] + stats_html = refresh_results(result.bundle, {})[0] + assert "Validation accuracy" in stats_html + assert "Test accuracy" not in stats_html + + +def test_example_dataset_defaults_to_sctrioseq2() -> None: + import app + + dropdown = next( + block for block in app.demo.blocks.values() + if getattr(block, "elem_id", None) == "dataset-select" + ) + assert dropdown.value == "scTrioseq2" + + def test_activation_choices_reach_model_config_and_artifact(tmp_path) -> None: hyperparameters = replace( _FAST,