https://pytorch-ignite.ai/tutorials/beginner/01-getting-started/ does it run training loop twice?

[inikishev]

https://pytorch-ignite.ai/tutorials/beginner/01-getting-started

its says this

trainer = create_supervised_trainer(model, optimizer, criterion, device)

val_metrics = {
    "accuracy": Accuracy(),
    "loss": Loss(criterion)
}

train_evaluator = create_supervised_evaluator(model, metrics=val_metrics, device=device)
val_evaluator = create_supervised_evaluator(model, metrics=val_metrics, device=device)

Does that mean after trainer runs training epoch it will run train_evaluator through the entire train dataset to calculate losses again? Or am I misunderstanding something?

[vfdev-5]

Below you can see the following:

So, yes, on each completed epoch we run train_evaluator.run(train_loader) on entire training dataloader to compute average loss value. This is optional and helps to see model's generalization capability. For large datasets we can also create a special train_eval_loader of size of validation dataset.

Keep also in mind that the average loss indicated during the training is not the one computed separately over the training set as during the training the model is updated on each iteration.

[inikishev]

ok thanks😊