Clarification Needed on Model Selection Strategy Across Epochs

[suni482]

I am currently looking into the implementation details of the model training process, particularly focusing on the model saving mechanism as delineated in the code. In if block, on line 489, it is observed that the model is persistently saved at the conclusion of each training epoch. However, the methodology employed for the selection of the optimal model based on the test/validation set performance remains unclear.

Could you kindly provide an elucidation on the criteria or algorithm used for identifying the most effective epoch based on the validation/test set? This clarification will significantly aid in understanding the overall model selection strategy within the training loop.

Thank you for your assistance.

[callsys]

Instead of choosing the model that achieves the best performance based on the performance of the validation/test set, we choose the model saved with the last epoch. If the performance of the model is evaluated after each training epoch, it might help us to choose a better model.