Last epoch that beat the incumbent by min_delta.
Early stopping turns a noisy validation curve into a checkpoint decision. Tune patience, minimum improvement, overfitting onset, and noise while watching the best epoch and stop signal evolve.
Simulate trainingThe validation metric estimates behavior on held-out data. A good stopping rule tracks the best qualifying improvement, tolerates short noisy regressions, saves the best checkpoint, and stops only after patience is exhausted.
Last epoch that beat the incumbent by min_delta.
Metric attached to the checkpoint to restore.
Consecutive epochs without qualifying improvement.
Continue, save, or stop under this rule.
When validation improves enough, persist model weights, optimizer state, scheduler state, epoch, and evaluation metadata. Stopping later should restore that selected checkpoint.
A single worse epoch is weak evidence. Patience waits for repeated non-improvement, while min_delta prevents tiny fluctuations from resetting the counter forever.
Repeatedly tuning against one validation split can overfit the selection process. Keep a final test set untouched, version the split, and report the checkpoint-selection rule.