I am working with tabular data where the prediction is numerical, not categorical.
I was both having bad results and errors when trying to select any metric different than accuracy
. After much search, I found this comment that explains that you must pass label_cls=FloatList
to label_from_df
for it to be treated as numerical.
Hence:
data = (TabularList.from_df(df, path=path, cat_names=cat_names, cont_names=cont_names, procs=procs)
.split_by_idx(list(range(0,last_index)))
.label_from_df(cols=dep_var,label_cls=FloatList)
.databunch())
Posting as a standalone with the search terms I looked for, so others can find it