OCDocker.OCScore.Utils.ExternalBlindEvaluation module

External blind evaluation for exported OCScore models.

Loads frozen preprocessing and model artifacts, applies them once to an external dataset, and writes predictions plus a provenance-rich evaluation report. No feature reduction, Optuna, or fitting occurs on blind rows.

class OCDocker.OCScore.Utils.ExternalBlindEvaluation.ExternalBlindConfig(export_dir, blind_csv, output_dir, label_column=None, group_column='receptor', kind_column='kind', device='cpu', pdbbind_export_dir=None, forbidden_dataset_hashes=None, command=None)

Bases: object

Configuration for external blind evaluation.

Parameters:

• export_dir (str | Path)

• blind_csv (str | Path)

• output_dir (str | Path)

• label_column (str | None)

• group_column (str | None)

• kind_column (str | None)

• device (str)

• pdbbind_export_dir (str | Path | None)

• forbidden_dataset_hashes (Sequence[str] | None)

• command (list[str] | None)

export_dir: str | Path

blind_csv: str | Path

output_dir: str | Path

label_column: str | None = None

group_column: str | None = 'receptor'

kind_column: str | None = 'kind'

device: str = 'cpu'

pdbbind_export_dir: str | Path | None = None

forbidden_dataset_hashes: Sequence[str] | None = None

command: list[str] | None = None

OCDocker.OCScore.Utils.ExternalBlindEvaluation.run_external_blind_evaluation(config)

Run one-shot external blind evaluation using a frozen export bundle.

Parameters:

config (ExternalBlindConfig)

Return type:

dict[str, Any]