pdmlabs.evaluation.evaluator

pdmlabs.evaluation.evaluator#

Classes

EvaluatorInterface()

Interface for implementing custom evaluation metrics.
class pdmlabs.evaluation.evaluator.EvaluatorInterface#

Bases: ABC

Interface for implementing custom evaluation metrics.

Users can implement this interface to inject their own metrics into the PdMExperiment evaluation pipeline. Evaluators must compute metrics and return them as a dictionary. They can also optionally log artifacts (e.g., plots or CSV files) to the active MLflow run.

abstract evaluate(experiment, **kwargs) → dict#

Compute evaluation metrics and return them.

Parameters:

  • experiment – The PdMExperiment instance running the evaluation.

  • **kwargs – Additional data required for evaluation depending on the experiment flavor. Common kwargs include: - result_scores: The predicted scores/values. - result_dates: The timestamps corresponding to the predictions. - results_isfailure / result_labels: Ground-truth labels or failure indicators. - plot_dictionary: A dictionary to store plotting data if debug=True. - rtfs: Run-to-failure indicators (for RUL/SA).

Returns:

A dictionary mapping metric names (str) to their computed values.

Example: {‘my_custom_f1’: 0.85, ‘my_custom_auc’: 0.92}

Return type:

dict

Contents