While I understand that the databunch concept is to cleanly work through a pipeline of preparation & modeling, I would like to access training and validation DataFrames and perhaps other constituents from
dbunch, which is created like this:
dbunch = (TabularList.from_df(df, path=Path.cwd(), cat_names=cats, cont_names=contins, procs=procs) .random_split_by_pct(0.3) .label_from_df(cols='target', label_cls=FloatList).databunch())
To send to other libraries like scikit-learn’s RandomForestregressor. Is there not an attribute like
dbunch.train_df that will do the job?