OCDocker.OCScore.Utils.DescriptorAggregateBaselines module

Simple row-wise baseline scores for OCScore comparisons.

Two families are supported:

  • Descriptor aggregates (descriptor_aggregate): mean/median/max/min over all model input features (same matrix passed to the network, after export scaler).

  • SF consensus (sf_consensus): mean/median/max/min over scoring-function columns only (Vina, Gnina, Smina, PLANTS, ODDT, …), matching historical SimpleConsensus semantics.

OCDocker.OCScore.Utils.DescriptorAggregateBaselines.evaluate_descriptor_aggregate_baselines_on_fold(feature_matrix, validation_indices, labels, groups, *, metric_names, bedroc_alpha=20.0, infer_higher_is_better, aggregators=('mean', 'median', 'max', 'min'))[source]

Evaluate descriptor-row aggregates on one validation fold.

Parameters:

  • feature_matrix (ndarray)

  • validation_indices (ndarray)

  • labels (ndarray)

  • groups (ndarray | None)

  • metric_names (Sequence[str])

  • bedroc_alpha (float)

  • infer_higher_is_better (Callable[[ndarray, ndarray], bool])

  • aggregators (Sequence[str])

Return type:

dict[str, dict[str, float]]

OCDocker.OCScore.Utils.DescriptorAggregateBaselines.evaluate_descriptor_aggregates_by_group(feature_matrix, validation_indices, labels, groups, *, metric_names, bedroc_alpha=20.0, infer_higher_is_better, aggregators=('mean', 'median', 'max', 'min'))[source]

Per-receptor metrics for descriptor-row aggregates.

Parameters:

  • feature_matrix (ndarray)

  • validation_indices (ndarray)

  • labels (ndarray)

  • groups (ndarray)

  • metric_names (Sequence[str])

  • bedroc_alpha (float)

  • infer_higher_is_better (Callable[[ndarray, ndarray], bool])

  • aggregators (Sequence[str])

Return type:

Mapping[str, DataFrame]

OCDocker.OCScore.Utils.DescriptorAggregateBaselines.evaluate_sf_consensus_baselines_on_fold(dataframe, validation_indices, scoring_columns, labels, groups, *, metric_names, bedroc_alpha=20.0, infer_higher_is_better, aggregators=('mean', 'median', 'max', 'min'))[source]

Evaluate SF-only row aggregates on one validation fold.

Parameters:

  • dataframe (DataFrame)

  • validation_indices (ndarray)

  • scoring_columns (Sequence[str])

  • labels (ndarray)

  • groups (ndarray | None)

  • metric_names (Sequence[str])

  • bedroc_alpha (float)

  • infer_higher_is_better (Callable[[ndarray, ndarray], bool])

  • aggregators (Sequence[str])

Return type:

dict[str, dict[str, float]]

OCDocker.OCScore.Utils.DescriptorAggregateBaselines.evaluate_sf_consensus_by_group(dataframe, validation_indices, scoring_columns, labels, groups, *, metric_names, bedroc_alpha=20.0, infer_higher_is_better, aggregators=('mean', 'median', 'max', 'min'))[source]

Per-receptor metrics for SF-only row aggregates.

Parameters:

  • dataframe (DataFrame)

  • validation_indices (ndarray)

  • scoring_columns (Sequence[str])

  • labels (ndarray)

  • groups (ndarray)

  • metric_names (Sequence[str])

  • bedroc_alpha (float)

  • infer_higher_is_better (Callable[[ndarray, ndarray], bool])

  • aggregators (Sequence[str])

Return type:

Mapping[str, DataFrame]

OCDocker.OCScore.Utils.DescriptorAggregateBaselines.format_descriptor_aggregate_scorer(aggregator)[source]

Return display key for a descriptor-row aggregate (e.g. desc_mean).

Parameters:

aggregator (str)

Return type:

str

OCDocker.OCScore.Utils.DescriptorAggregateBaselines.format_sf_consensus_scorer(aggregator)[source]

Return display key for an SF-only row aggregate (e.g. sf_mean).

Parameters:

aggregator (str)

Return type:

str

OCDocker.OCScore.Utils.DescriptorAggregateBaselines.row_aggregate_feature_scores(feature_matrix, aggregators=('mean', 'median', 'max', 'min'))[source]

Aggregate each row of a feature matrix with simple reducers.

Parameters:

  • feature_matrix (np.ndarray) – Shape (n_rows, n_features).

  • aggregators (Sequence[str], optional) – Reducers to apply. Supported: mean, median, max, min.

Returns:

One score vector per aggregator name (unprefixed keys).

Return type:

dict[str, np.ndarray]

OCDocker.OCScore.Utils.DescriptorAggregateBaselines.row_aggregate_sf_scores(dataframe, scoring_columns, row_indices=None, aggregators=('mean', 'median', 'max', 'min'))[source]

Aggregate each row across scoring-function columns only.

Parameters:

  • dataframe (pd.DataFrame) – Reduced DUDEz (or full) table containing SF columns.

  • scoring_columns (Sequence[str]) – Scoring-function column names (e.g. vina_vina).

  • row_indices (np.ndarray | None, optional) – Optional positional row subset (iloc), matching export split indices. When None, use all rows.

  • aggregators (Sequence[str], optional) – Row reducers to apply.

Returns:

One score vector per aggregator name (unprefixed keys).

Return type:

dict[str, np.ndarray]

OCDocker.OCScore.Utils.DescriptorAggregateBaselines.scorer_type_for_baseline_name(scorer)[source]

Map a baseline scorer label to its scorer_type, or None if not a baseline.

Parameters:

scorer (str)

Return type:

str | None