callcut.evaluation.EventMetrics

class callcut.evaluation.EventMetrics(n_ground_truth, n_predicted, tp, fp, fn, precision, recall, f1)

Event-level detection metrics.

These metrics evaluate detection at the call/event level: each ground truth call is either matched to a prediction (true positive) or missed (false negative), and each prediction is either matched (true positive) or a false alarm (false positive).

Parameters:

n_ground_truthint

Total number of ground truth events.

n_predictedint

Total number of predicted events.

tpint

True positives (correctly matched predictions).

fpint

False positives (predictions without a matching ground truth).

fnint

False negatives (ground truth events without a matching prediction).

precisionfloat

Precision = TP / (TP + FP). Of the predicted calls, what fraction are real.

recallfloat

Recall = TP / (TP + FN). Of the real calls, what fraction were detected.

f1float

F1 score = 2 * precision * recall / (precision + recall). Harmonic mean of precision and recall.

Examples

>>> metrics = EventMetrics(
...     n_ground_truth=10,
...     n_predicted=8,
...     tp=7,
...     fp=1,
...     fn=3,
...     precision=0.875,
...     recall=0.7,
...     f1=0.778,
... )
>>> metrics.precision
0.875