Welcome to pytorch-metrics’s documentation!¶
Implemented metrics¶
Note on memory efficiency¶
Most metrics can reduce their internal state when the user calls update, however some metrics like ExplainedVariance requires access to the full set of targets and predictions when compute is called. These metrics therefore store all targets and predictions passed to update and are therefore not memory efficient. Wheather or not a metric is memory efficient, is stored in the boolean variable Metric.memory_efficient. If this is false, use the metric with care.