def extract_output(
self, accumulator: _Matrices
) -> Dict[metric_types.PlotKey,
metrics_for_slice_pb2.MultiLabelConfusionMatrixAtThresholds]:
pb = metrics_for_slice_pb2.MultiLabelConfusionMatrixAtThresholds()
for threshold in sorted(accumulator.keys()):
matrix = pb.matrices.add(threshold=threshold)
for k in sorted(accumulator[threshold].keys()):
matrix.entries.add(
actual_class_id=k.actual_class_id,
predicted_class_id=k.predicted_class_id,
false_negatives=accumulator[threshold][k].false_negatives,
true_negatives=accumulator[threshold][k].true_negatives,
false_positives=accumulator[threshold][k].false_positives,
true_positives=accumulator[threshold][k].true_positives)
return {self._key: pb}
def extract_output(
self, accumulator: _Matrices
) -> Dict[metric_types.PlotKey,
metrics_for_slice_pb2.MultiLabelConfusionMatrixAtThresholds]:
pb = metrics_for_slice_pb2.MultiLabelConfusionMatrixAtThresholds()
for threshold in sorted(accumulator.keys()):
# Convert -epsilon and 1.0+epsilon back to 0.0 and 1.0.
if threshold == -_EPSILON:
t = 0.0
elif threshold == 1.0 + _EPSILON:
t = 1.0
else:
t = threshold
matrix = pb.matrices.add(threshold=t)
for k in sorted(accumulator[threshold].keys()):
matrix.entries.add(
actual_class_id=k.actual_class_id,
predicted_class_id=k.predicted_class_id,
false_negatives=accumulator[threshold][k].false_negatives,
true_negatives=accumulator[threshold][k].true_negatives,
false_positives=accumulator[threshold][k].false_positives,
true_positives=accumulator[threshold][k].true_positives)
return {self._key: pb}