Tabular data with test set

I am trying to figure out how to specify a test set with TabularDataBunch.from_df. There doesn’t appear to be an argument for test_df anymore in the function signature, though the docs still say that a test_df can be optionally specified.

Looking at the source code for from_df, it doesn’t appear to either handle a test set at all.

What is the recommended way to get test predictions with the tabular learner? I can call learn.predict on each row in the test dataframe, however this is quite slow for large test sets.

Edit: Using fastai 1.0.30

1 Like

This has been answered in this topic.
Basically: learn how to use the data block API to customize the behavior of the factory methods to your need :wink: