nexuml_library.scenarios.evaluation.anomaly¶
Evaluation spec builders for anomaly detection scenarios.
decision_rule_spec(
score_key: str = "anomaly_score",
decision_key: str = "decision",
fit_mask_key: str | None = None,
fit_label_key: str | None = None,
) -> list[LayerSpec]
Return a decision-rule layer that produces decision_key from score_key.
classification_metrics_spec(
score_key: str = "anomaly_score",
decision_key: str = "decision",
label_key: str = "anomaly",
metrics: list[str] | None = None,
fit_mask_key: str | None = None,
fit_label_key: str | None = None,
) -> dict[str, list[LayerSpec]]
Return a pipeline stage dict containing a decision rule and ClassificationMetrics.
A DecisionRulePipelineLayer is prepended so that the decision key is
produced from score_key before ClassificationMetrics consumes it.
anomaly_evaluation_spec(
label_key: str = "anomaly",
group_keys: list[str] | None = None,
output_score_key: str = "anomaly_score",
decision_key: str | None = "decision",
dcase_metric_axes: dict[str, str] | None = None,
max_fpr: float = 0.1,
) -> EvaluationSpec
Evaluation-only spec for anomalous sound detection.
Returns only the reporting algorithm (anomaly_evaluator). Score-producing components (group_distance, calibration, reduction) are pipeline layers.
Returns:
| Name | Type | Description |
|---|---|---|
EvaluationSpec |
EvaluationSpec
|
Evaluation specification with anomaly evaluator algorithm. |