ANLP_WS24_CA2/EarlyStopping.py

28 lines
985 B
Python

import torch
class EarlyStopping:
def __init__(self, patience=5, verbose=False):
self.patience = patience
self.verbose = verbose
self.counter = 0
self.best_score = None
self.early_stop = False
def __call__(self, val_loss, model):
score = -val_loss
if self.best_score is None:
self.best_score = score
self.save_checkpoint(val_loss, model)
elif score < self.best_score:
self.counter += 1
if self.counter >= self.patience:
self.early_stop = True
else:
self.best_score = score
self.save_checkpoint(val_loss, model)
self.counter = 0
def save_checkpoint(self, val_loss, model, filename='checkpoint.pt'):
if self.verbose:
print(f'Validation loss decreased ({self.best_score:.6f} --> {val_loss:.6f}). Saving model ...')
torch.save(model.state_dict(), f'checkpoints/{filename}')