EarlyStopping

class behavenet.fitting.training.EarlyStopping(patience=10, min_epochs=10, delta=0)

Bases: object

Stop training when a monitored quantity has stopped improving.

Adapted from: https://github.com/Bjarten/early-stopping-pytorch/blob/master/pytorchtools.py

Methods Summary

on_val_check(epoch, curr_loss)

Check to see if loss has begun to increase on validation data for current epoch.

Methods Documentation

on_val_check(epoch, curr_loss)

Check to see if loss has begun to increase on validation data for current epoch.

Rather than returning the results of the check, this method updates the class attribute should_stop, which is checked externally by the fitting function.

Parameters

• epoch (int) – current epoch

• curr_loss (float) – current loss