Questions about evaluating metric ACC

[yichenwang231]

Thanks for your work, when I reproduced the classification problem, I found that the results obtained by the validation set and the test set during the training process were inconsistent, and the results of the test set after the training was completed were also inconsistent with the results obtained during the training process. https://github.com/thuml/Time-Series-Library/blob/843ee18239b895b3018e7240002e85a0433e7f8b/exp/exp_classification.py#L128 https://github.com/thuml/Time-Series-Library/blob/843ee18239b895b3018e7240002e85a0433e7f8b/exp/exp_classification.py#L129 When I evaluate the validation set twice like this, the results will be different vali_loss, val_accuracy = self.vali(vali_data, vali_loader, criterion) vali_loss1, val_accuracy1 = self.vali(vali_data, vali_loader, criterion) print(val_accuracy,val_accuracy1 ) I don't know if it's due to randomness or other questions, but I'm looking forward to your answers.Thank you.