def calculate_metric(self):
return compute_multi_label_classification_metrics(
[
LabelListPrediction(scores, pred, expect)
for scores, pred, expect in zip(
self.all_scores, self.all_preds, self.all_targets)
],
self.label_names,
self.calculate_loss(),
)
def test_compute_multi_label_classification_metrics(self) -> None:
roc_auc_dict = {"label1": 1.0, "label2": 0.25, "label3": 0.0}
metrics = compute_multi_label_classification_metrics(PREDICTIONS,
LABEL_NAMES,
loss=5.0)
self.assertAlmostEqual(metrics.roc_auc, 1.25 / 3)
for k, v in metrics.per_label_soft_scores.items():
metric_value = getattr(v, "roc_auc", None)
self.assertAlmostEqual(metric_value, roc_auc_dict[k])
def calculate_metric(self):
return compute_multi_label_classification_metrics(
[
LabelListPrediction(scores, pred, expect)
for scores, pred, expect in zip(
self.all_scores, self.all_preds, self.all_targets)
],
self.label_names,
self.calculate_loss(),
average_precisions=(not self.is_memory_efficient),
recall_at_precision_thresholds=self.recall_at_precision_thresholds,
)
