Data_block API and add_test for lm

fastai users
#1

I couldn’t verify that show_batch includes test data. If I create a language model databunch (data_lm) from csv and a language model learner like this:

data_lm = (TextList.from_csv(path, 'train_valid.csv', cols='text')
                   .random_split_by_pct(valid_pct=0.2, seed=None) 
                   #.split_from_df(col='is_valid') 
                   .label_for_lm()
                   .add_test(df_test['text']) # check
                   .databunch(bs=32))
learn = language_model_learner(data_lm, pretrained_model=URLs.WT103_1)

does the text cols of the test set (df_test) participate in fine-tuning (learn.fit_one_cycle())?

Update: found I had to pass ds_type (train by default), so I guess test set takes part in model fine-tuning.