Building a TabularDataBunch from different data frames


I’m trying to train a tabular model.
I have two data frames - one for train and the other for validation.

how can I create a TabularDataBunch using them?
I tried appending them, then using TabularDataBunch.from_df()
but they are too big to handle such operation…

Thanks! :slightly_smiling_face:

I believe it’s from_dfs IIRC

Thank you :slight_smile:

you mean there’s a function called from_dfs()?
it’s not a method of TabularDataBunch \ DataBunch

No, I believe it’s TabularDataBunch.from_dfs

Edit: sorry, v1 doesn’t have that, and neither does v2 :frowning:

What you could do is if it is too much memory, in fastai2 I’ve utilized NumPy if you can manage to pre-process the data beforehand (Grab your procs on the training DataFrame, then apply them to your valid DataFrame), this is also something you could do without the numpy too, just some FFT :slight_smile: