Hello! I am working on an image classifier, and I am trying to see how it performs given subsets of the same dataset. Is there a way I could use the same test set? Would that be silly? Here is my code.
To see the performance of a training subset (batch) I think you can do
# loss_func=CrossEntropyLoss() x, y = dls.train.one_batch() preds = learn.model(x) learn.loss_func(preds,y)