results/summary.json

{
  "model_name": "context_weights_k25_knn20",
  "vintage": "2026_W41",
  "architecture": {
    "modalities": [
      "waveform",
      "acg",
      "stpc"
    ],
    "modality_latent_dim": 20,
    "joint_latent_dim": 60,
    "gmm_components": 25,
    "gmm_covariance_type": "full",
    "gmm_geometry": "global means and covariance matrices",
    "context_conditioning": "mixture weights gamma, and the TRAIN exemplars representing each component",
    "context_definition": "50 MERFISH PCs + 50 AGEA PCs",
    "knn_k": 20,
    "knn_policy": "distance-weighted empirical TRAIN-exemplar projection in standardized latent space",
    "phenotype_readout": {
      "method": "context-local members: gamma_k(x) weights the mean of component k's TRAIN members nearest to x in a context key, shrunk toward all its members",
      "key_dim": 3,
      "key_ridge_alpha": 1000.0,
      "neighbours": 4096,
      "shrinkage": 10.0,
      "off_data_quantile": 0.99,
      "absent_context": "global member means"
    }
  },
  "split_counts": {
    "train_units": 56497,
    "validation_units": 7671,
    "test_units": 15880,
    "train_pids": 572,
    "validation_pids": 75,
    "test_pids": 149
  },
  "heldout_basic_test": {
    "model": "context_weights_k25_knn20",
    "split": "held-out test PIDs",
    "reconstruction": {
      "waveform_mse": 0.000788584933616221,
      "acg_mse": 4.7604778956156224e-05,
      "stpc_mse": 0.006178959738463163,
      "n_test_units": 4096
    },
    "latent_nll": {
      "conditional_k25": -26.648546443872384,
      "unconditional_k25": -26.095386854718793,
      "conditional_improvement": 0.5531595891535908,
      "lower_is_better": true
    },
    "knn20": {
      "k": 20,
      "nearest_distance_mean": 3.551875114440918,
      "nearest_distance_median": 3.415438652038574,
      "kth_distance_mean": 4.589000225067139,
      "kth_distance_median": 4.397944450378418,
      "effective_neighbors_mean": 19.91122817993164,
      "effective_neighbors_median": 19.925037384033203
    },
    "sanity_checks": {
      "all_metrics_finite": true,
      "conditional_beats_unconditional": true
    }
  },
  "invariants": [
    "TEST PIDs are never used for AE/GMM/context training or as kNN retrieval exemplars.",
    "The GMM means and covariance matrices are global.",
    "Molecular context sets the GMM mixture weights gamma_k(x) and selects which TRAIN exemplars represent each component in the phenotype readout; the conditional latent density depends on it only through gamma_k(x).",
    "The kNN stage changes phenotype projection only; it does not change the conditional latent density.",
    "All spatial modeling uses the configured single-hemisphere mirroring convention."
  ]
}