Metrics

Metrics to evaluate performance and to use as loss function.

Not all metrics are suitable as loss functions. Some are not differentiable or do not return a scalar.

Metrics are computed on the training and validation sets every eval_freq steps, set in Experiment.

class optexp.metrics.Metric

Abstract base class for metrics.

Output-Target Metrics

Metrics that are computed on the output of the model and the target.

class optexp.metrics.LossLikeMetric

Abstract base class for loss-like metrics, which take inputs and labels.

class optexp.metrics.Accuracy

class optexp.metrics.CrossEntropy

class optexp.metrics.AccuracyPerClass

Accuracy per class.

Can result in large logs on problems with many classes.

class optexp.metrics.CrossEntropyPerClass

Cross entropy loss per class.

Can result in large logs on problems with many classes.