OCDocker.OCScore.Utils.DUDEzScaling module

Explicit DUDEz feature scaling policies for staged OCScore transfer.

PDBbind regression fits a train-only StandardScaler. DUDEz screening must apply a documented scaling strategy so transferred encoders see compatible inputs.

class OCDocker.OCScore.Utils.DUDEzScaling.DUDEzScalingConfig(strategy='pdbbind_scaler', strict=True)[source]

Bases: object

Configuration for DUDEz feature scaling during staged screening.

Parameters:

  • strategy (str) – Scaling strategy name.

  • strict (bool, optional) – When True, forbid implicit unscaled features during transfer, by default True.

strategy: Literal['pdbbind_scaler', 'dudez_train_scaler', 'none_prestandardized'] = 'pdbbind_scaler'
strict: bool = True
OCDocker.OCScore.Utils.DUDEzScaling.scale_dudez_features(X, *, train_idx, val_idx, test_idx, config, selected_features, pdbbind_scaler=None)[source]

Scale DUDEz feature matrices according to config.

Parameters:

  • X (np.ndarray) – Full DUDEz feature matrix before splitting.

  • train_idx (np.ndarray) – Training row indices.

  • val_idx (np.ndarray) – Validation row indices.

  • test_idx (np.ndarray) – Test row indices.

  • config (DUDEzScalingConfig) – Scaling policy.

  • selected_features (Sequence[str]) – Feature column order used for modeling.

  • pdbbind_scaler (StandardScaler | None, optional) – Train-fitted PDBbind scaler from the regression stage.

Returns:

Scaled train/val/test arrays, metadata dict, and optional fitted DUDEz scaler.

Return type:

tuple

OCDocker.OCScore.Utils.DUDEzScaling.scaling_config_to_dict(config)[source]

Serialize scaling config for JSON artifacts.

Parameters:

config (DUDEzScalingConfig)

Return type:

dict[str, Any]