Deploy Space: update app.py

app.py

@@ -1,5 +1,6 @@
 import json
 import os
+import re
 from pathlib import Path
 
 import gradio as gr
@@ -17,6 +18,11 @@ from src.utils import (
     dataset_section_title,
     get_grouped_dfs,
     group_datasets_by_domain,
+    load_baseline_rank_history,
+    load_baseline_rank_history_weekly,
+    load_forecast_snapshots,
+    make_forecast_plot,
+    make_rank_trend_plot,
     prepare_ranks_table,
     prepare_values_table,
 )
@@ -72,25 +78,55 @@ def _dataset_tables(grouped: dict, datasets: list[str]) -> tuple[list, list]:
     return values_tables, rank_tables
 
 
+def _get_model_slugs(results_path: str) -> list[str]:
+    """Return subdirs of results_path that contain a forecasts/ subdirectory."""
+    slugs = []
+    for p in sorted(Path(results_path).iterdir()):
+        if p.is_dir() and (p / "forecasts").exists():
+            slugs.append(p.name)
+    return slugs
+
+
+def _build_forecast_choices(results_path: str) -> dict[str, dict[str, dict]]:
+    """Build {model_slug: {dataset_name: snapshot}} for all models."""
+    choices: dict[str, dict[str, dict]] = {}
+    for slug in _get_model_slugs(results_path):
+        snaps = load_forecast_snapshots(results_path, slug)
+        if snaps:
+            choices[slug] = snaps
+    return choices
+
+
+def _short_dataset_key(dataset_name: str) -> str:
+    """Extract the base dataset key from a leaderboard dataset name like 'ds/freq/term'."""
+    return dataset_name.split("/")[0]
+
+
 def refresh_leaderboard() -> tuple:
     grouped = get_grouped_dfs(RESULTS_PATH)
     status = load_online_status(RESULTS_PATH)
     datasets = grouped["datasets"]
+    daily_df = load_baseline_rank_history(RESULTS_PATH)
 
     outputs: list = [
-        prepare_values_table(grouped["overall_values"]),
-        prepare_ranks_table(grouped["overall_ranks"]),
+        prepare_values_table(grouped["overall_values"]),    # [0]
+        prepare_ranks_table(grouped["overall_ranks"]),      # [1]
+        daily_df,                                           # [2] daily table
+        load_baseline_rank_history_weekly(RESULTS_PATH),    # [3] weekly table
+        make_rank_trend_plot(daily_df),                     # [4] rank trend plot
     ]
     value_tables, rank_tables = _dataset_tables(grouped, datasets)
-    outputs.extend(value_tables)
-    outputs.extend(rank_tables)
-    outputs.append(build_status_line(status))
+    outputs.extend(value_tables)    # [5 .. 5+N-1]
+    outputs.extend(rank_tables)     # [5+N .. 5+2N-1]
+    outputs.append(build_status_line(status))  # [-1]
     return tuple(outputs)
 
 
 grouped_initial = get_grouped_dfs(RESULTS_PATH)
 DATASETS: list[str] = grouped_initial["datasets"]
 DOMAIN_GROUPS = group_datasets_by_domain(DATASETS, RESULTS_PATH)
+FORECAST_CHOICES = _build_forecast_choices(RESULTS_PATH)
+MODEL_SLUGS = list(FORECAST_CHOICES.keys())
 INITIAL_OUTPUTS = refresh_leaderboard()
 
 demo = gr.Blocks(css=custom_css)
@@ -104,6 +140,7 @@ with demo:
     dataset_idx = 0
 
     with gr.Tabs(elem_classes="tab-buttons"):
+        # ── Overall Tab ──────────────────────────────────────────────────────
         with gr.TabItem("Overall"):
             gr.Markdown(
                 "Geometric mean of **absolute** metrics across all datasets. Lower is better.",
@@ -122,6 +159,36 @@ with demo:
                 label="Ranks (average per-dataset rank, lower is better)",
             )
 
+            gr.Markdown(
+                "**Baseline daily rank history** · average rank across all datasets per day (lower is better)",
+                elem_classes="markdown-text",
+            )
+            baseline_history_df = gr.Dataframe(
+                value=INITIAL_OUTPUTS[2],
+                interactive=False,
+                wrap=True,
+                label="Baseline rank history — by day",
+            )
+            gr.Markdown(
+                "**Baseline weekly rank history** · average of daily ranks per ISO week (lower is better)",
+                elem_classes="markdown-text",
+            )
+            baseline_weekly_df = gr.Dataframe(
+                value=INITIAL_OUTPUTS[3],
+                interactive=False,
+                wrap=True,
+                label="Baseline rank history — by week",
+            )
+            gr.Markdown(
+                "**Baseline rank trend** · MASE / CRPS / MAE / RMSE rank over time",
+                elem_classes="markdown-text",
+            )
+            rank_trend_plot = gr.Plot(
+                value=INITIAL_OUTPUTS[4],
+                label="Rank trend over time",
+            )
+
+        # ── Domain Tabs ───────────────────────────────────────────────────────
         for domain, domain_datasets in DOMAIN_GROUPS.items():
             with gr.TabItem(domain):
                 for dataset in domain_datasets:
@@ -131,7 +198,7 @@ with demo:
                     )
                     dataset_value_dfs.append(
                         gr.Dataframe(
-                            value=INITIAL_OUTPUTS[2 + dataset_idx],
+                            value=INITIAL_OUTPUTS[5 + dataset_idx],
                             interactive=False,
                             wrap=True,
                             show_label=False,
@@ -143,12 +210,63 @@ with demo:
                     )
                     dataset_rank_dfs.append(
                         gr.Dataframe(
-                            value=INITIAL_OUTPUTS[2 + len(DATASETS) + dataset_idx],
+                            value=INITIAL_OUTPUTS[5 + len(DATASETS) + dataset_idx],
                             interactive=False,
                             wrap=True,
                             show_label=False,
                         )
                     )
+
+                    # ── Forecast plot for this dataset ──────────────────────
+                    if MODEL_SLUGS:
+                        ds_key = _short_dataset_key(dataset)
+                        # Find which models have a snapshot for this dataset
+                        available_models = [
+                            slug for slug in MODEL_SLUGS
+                            if any(
+                                _short_dataset_key(k) == ds_key
+                                for k in FORECAST_CHOICES.get(slug, {})
+                            )
+                        ]
+                        if available_models:
+                            gr.Markdown(
+                                "**Forecast: true vs predicted** · select model",
+                                elem_classes="dataset-heading",
+                            )
+                            model_dd = gr.Dropdown(
+                                choices=available_models,
+                                value=available_models[0],
+                                label="Model",
+                                interactive=True,
+                            )
+
+                            def _get_initial_snapshot(slug, dk=ds_key):
+                                snaps = FORECAST_CHOICES.get(slug, {})
+                                for name, snap in snaps.items():
+                                    if _short_dataset_key(name) == dk:
+                                        return snap
+                                return None
+
+                            forecast_plot = gr.Plot(
+                                value=make_forecast_plot(_get_initial_snapshot(available_models[0])),
+                                label="True vs Predicted",
+                            )
+
+                            def _make_update_fn(dk):
+                                def _update(slug):
+                                    snaps = FORECAST_CHOICES.get(slug, {})
+                                    for name, snap in snaps.items():
+                                        if _short_dataset_key(name) == dk:
+                                            return make_forecast_plot(snap)
+                                    return None
+                                return _update
+
+                            model_dd.change(
+                                fn=_make_update_fn(ds_key),
+                                inputs=[model_dd],
+                                outputs=[forecast_plot],
+                            )
+
                     dataset_idx += 1
 
     with gr.Accordion("About", open=False):
@@ -165,6 +283,9 @@ with demo:
     refresh_outputs = [
         overall_values_df,
         overall_ranks_df,
+        baseline_history_df,
+        baseline_weekly_df,
+        rank_trend_plot,
         *dataset_value_dfs,
         *dataset_rank_dfs,
         status,