Ancient Fastai Machine Learning Functions

I just started the ML for practitioners and found out an issue there is no more existence of some useful functions for pre-processing in the latest version of fastai. So, i tried to create a pip package to solve this issue
Moreover, I have published the first pip package

!pip install fastai_ml
from fastai_ml.basics import *

Right now i have added only functions that had been used in first lecture

train_cats(), fix_missing(), numericalize(), scale_vars(), proc_df()

1 Like

Very nice! I will say though, they do exist just they changed names. It’s now Normalize, FillMissing, and Categorify :wink:

but i think now they are for DataBunch (not for panda’s DataFrame) and for apply these you have to separate cat and cont names etc.
Please correct me if I am wrong :thinking:

You do and don’t. If you check the docs we can get the cat and cont names quickly by get_cat_cont (or something similar) and we can also function call them normally. (Though in practice you may want to decide these yourself as sometimes you may want a numerical to be categorical if it’s a one-hot representation, etc)

And they can be applied to a dataframe too! :slight_smile:

Thanks for your guidance :heart_eyes:… As i just started with the ML course so I’m getting trouble in applying these methods. Now I’ll check for sure. Thank you for your support

1 Like

The new upcoming version of the course will be a mix, which should help some. Lesson 4 on the new lecture (Practical Deep Learning for Coders) goes into tabular too :slight_smile: I’d recommend Intro ML for the concepts and then PDLC for a newer application of those approaches/concepts