Ref: Get_preds gives different result from predict
In fastai v1, after we load a previously exported learner to predict, we need to set the ds’ processor with the learner’s, because we want to use the same parameters used in training for categorification and normalization.
data_test = (TabularList.from_df(df, cat_names=cat_names, cont_names=cn,
procs=[Categorify, Normalize],
processor=learner.data.processor)
.split_none()
.label_from_df(cols=dep_var))
In fastai v2, do we still need to do the same? If so, how?