Questions about test sets (i.e. not training or validation sets)

Is there a recommended way to make 3 sets of data (training, validation, and test) via the FastAI API?

I know of sklearn’s “train_test_split” method, but this only gives us two groups of data, what’s the recommended way to turn this into 3 sets? Should I just call “train_test_split” twice (i.e. once to get training and test, and another to split my previous “training” into my real training set and my validation set)?

And what % of data should we reserve for the test set? I know with the validation set, the standard % is 20% of your data. But what about the test set?

1 Like

Check this out by @muellerzr

1 Like