Are there easy way to make learner calculate metric on training set as well?

same as title. The default behavior is only calc metric on validation set.

Learner.get_preds() first argument is the dataloader index (“0” for training and “1” for validation). So you can easily get the predictions on your test observations and then compute the metrics at the end end of the training cycle. But to get metrics on the training data during the training process would be much harder (although certainly possible).