Why are you trying to do this? If you want to do predictions / add a test set you can use the DataLoaders .test_dl method. So:
learn.dls.test_dl(test_df)
If you’r doing it just for the sake of it, I guess you would create the two single ‘DataLoader’ objects with the appropriate sources and combine them to a ‘DataLoaders’ object (yes, those a two different things ); the book should go into that, if I remember correctly. (edit: it does in chapter 19)
Ah ok, I mistook that . You do this by passing a splits parameter when creating the TabularPandas. Check the tutorial on tabular data here. Then dls.train holds the training and dls.valid the validation dataloader.