Outputs

Outputs are organized by project run folder:

<output_dir>/<MODEL_NAME>_<TARGET>/

Core Artifacts

• best_hyperparameters.json: Best hyperparameters from sweep
• final_model/: Serialized trained model
• final_model_checkpoints/: Training checkpoints

Evaluation Artifacts

• evaluate/metrics.json: Evaluation metrics
• evaluate/predictions.parquet: predictions over test chromosome
• figures/scatter_test.png: True vs predicted scatter
• evaluate/embeddings_umap.parquet: UMAP embedding table
• figures/embeddings_umap.png: UMAP figure

SHAP Artifacts

• shap/shap_mean_abs.csv: Mean absolute SHAP by feature and target
• figures/shap_clustermap.png: Mean |SHAP| heatmap across cofactors and targets
• shap/spatial_shap_by_offset_<target>.csv: Spatial SHAP summary by genomic offset per target
• figures/shap_spatial_heatmap_<target>.png: Spatial SHAP heatmap (cofactor × offset) per target
• figures/shap_offset_line_<target>.png: SHAP importance profile by genomic offset per target

Logs

• <PROJECT>.log: Run-level pipeline log
• logging/: TensorBoard files (if enabled)