Metrics

A metric is a function that is used to judge the performance of your model.

Metric functions are similar to loss functions, except that the results from evaluating a metric are not used when training the model. Note that you may use any loss function as a metric.

Available metrics

For classification

• Accuracy
• categorical_accuracy
• binary_accuracy
• ROC_AUC
• F1_Score

For regression

• MeanSquaredError
• MeanAbsoluteError
• RootMeanSquaredError

Custom metrics

Any callable object can be custom metrics.

For example,

def mse(y_pred, y_true):
    return torch.mean((y_pred - y_true) ** 2)

Or

class MSE(object):
    def __call__(y_pred, y_true):
        return torch.mean((y_pred - y_true) ** 2)

If you want to name the metric with an abbr. or alias, you can subclass keras4torch.metrics.Metric and implement get_abbr() method. Another way to do the same thing is providing a dict to compile() when you config the model.

model.compile(
    optimizer='adam',
    loss='mae',
    metrics={'mse': MeanSquaredError()}
)

Then the model will use 'mse' as the abbr. of MeanSquaredError in the logger.