{
  "model_id": "2026_W41_unit",
  "task": "unit-encoding",
  "model_class": "UnitAutoencoder",
  "vintage": "2026_W41",
  "training": {
    "random_seed": 0,
    "training_size": 572,
    "training_units": 56497,
    "validation_size": 75,
    "validation_units": 7671,
    "testing_size": 149,
    "testing_units": 15880
  },
  "environment": {
    "python": "3.11.9",
    "xgboost": "3.2.0",
    "scikit_learn": "1.8.0",
    "numpy": "2.3.5",
    "pandas": "3.0.5",
    "torch": "2.7.1+cu118",
    "ephysatlas": "0.9.0",
    "iblatlas": "1.2.0"
  },
  "method": "gmm",
  "granularity": "unit",
  "artifacts": {
    "autoencoder": "autoencoder.pt",
    "config": "config.json",
    "scaler": "shared_latent_scaler.joblib",
    "gmm": "global_gmm.joblib",
    "context_transform": "context_transform.joblib",
    "context_weights": "context_weight_model_bundle.pt",
    "knn_bank": "knn_bank.npz",
    "component_features": "component_feature_expectations.npz",
    "stats": "preprocessing/unit_stats.npz",
    "split": "split.json",
    "context": [
      "agea_vol_pca.npy",
      "merfish_vol_pca.npy"
    ],
    "results": [
      "results/summary.json"
    ]
  },
  "inputs": {
    "index": [
      "pid",
      "cluster"
    ],
    "columns": [
      "x",
      "y",
      "z"
    ],
    "modalities": [
      "waveform",
      "acg",
      "stpc"
    ]
  },
  "outputs": {
    "kind": "continuous",
    "columns": [
      "depolarisation_slope",
      "recovery_slope",
      "repolarisation_slope",
      "spatial_spread_um",
      "tip_val",
      "spike_width_secs",
      "predepolarisation_width_secs",
      "spike_amplitude",
      "peak_to_trough_ratio_log",
      "polarity"
    ],
    "feature_order_sha256": "8536d8c5c99959110df046905224cf91de680434287153131d3ba85eaed968af",
    "latent_dim": 60
  },
  "config": {
    "architecture": {
      "modalities": [
        "waveform",
        "acg",
        "stpc"
      ],
      "modality_latent_dim": 20,
      "latent_dim": 60,
      "gmm_components": 25,
      "gmm_covariance_type": "full",
      "knn_k": 20
    },
    "context": {
      "n_cell_pcs": 50,
      "n_gene_pcs": 50,
      "conditions": "mixture weights, and which training exemplars represent each component in the phenotype readout; component means/covariances are global",
      "source_model": "int-brain-lab/ea-encoder-channel"
    },
    "preprocessing": {
      "waveforms": "per-unit max-abs normalized, 20 channels centred on the peak channel",
      "stats_file": "preprocessing/unit_stats.npz",
      "mirror_x_to_single_hemisphere": true,
      "mirror_x_sign": -1.0
    },
    "readout": {
      "method": "context_local_members",
      "description": "the molecular context sets the mixture weights gamma_k(x) and selects which training exemplars represent each component: its members nearest to x in a context key, shrunk toward all of its members -- further as the key leaves the training data, and entirely without context",
      "key_dim": 3,
      "key_ridge_alpha": 1000.0,
      "neighbours": 4096,
      "shrinkage": 10.0,
      "off_data_quantile": 0.99,
      "absent_context": "global member means"
    }
  },
  "data_source": {
    "backend": "s3-ibl",
    "project": "ibl_neuropixel_brainwide_01_2026_W41",
    "requires_one": true
  }
}