Metrics

A collection of methods for the evaluation of classifiers.

@author: J. Cid-Sueiro, A. Gallardo-Antolin

src.domain_classifier.metrics.binary_metrics(preds, labels, sampling_probs=None)

Compute performance metrics based on binary labels and binary predictions only

Parameters

• preds (np.array) – Binary predictions

• labels (np.array) – True class labels

• sampling_probs (np.array) – Sampling probabilities. It is used to compute performance metrics as weighted averages

Returns

eval_scores – A dictionary of evaluation metrics.

Return type

dict

src.domain_classifier.metrics.print_metrics(m, roc=None, title='', data='', print_unweighted=True)

Pretty-prints the given metrics

Parameters

• m (dict) – Dictionary of metrics (produced by the binary_metrics() method)

• roc (dict or None, optional (default=None)) – A dictionary of score-based metrics. It is used to print AUC.

• data (str, optional (default=””)) – Identifier of the dataset used to compute the metrics. It is used to compose the text title

• print_unweighted (boolean (default=True)) – If True, unweighted metrics are printed in addition to the weighted metrics

src.domain_classifier.metrics.score_based_metrics(scores, labels, sampling_probs=None)

Computes score-based metrics

Parameters

• scores (np.array) – Score values

• labels (np.array) – Target values

• sampling_probs (np.array) – Sampling probabilities. It is used to compute performance metrics as weighted averages

Returns

eval_scores – A dictionary of evaluation metrics.

Return type

dict