search

AutoViML / featurewiz

Use advanced feature engineering strategies and select best features from your data set with a single line of code. Created by Ram Seshadri. Collaborators welcome.
Apache License 2.0
588 stars 87 forks source link

Unpin requirements? #112

Open ecederstrand opened 5 months ago

ecederstrand commented 5 months ago

featurewiz currently has restrictions on e.g. pandas<2.0, scipy<1.11.0 and xgboost<=1.6.2. These packages are getting a bit outdated. Is there anything holding back versions for these packages?

AutoViML commented 5 months ago

Hi Erik: @ecederstrand 👍 thank you for trying out featurewiz.

pandas 2.0 is a horrible release. They broke 1000's of libraries that have been working forever. Do you really want to upgrade to pandas 2.0 just for the sake of keeping up with your neighbors? I can fix featurewiz to handle pandas 2.0 but once your machine is upgraded to pandas-2.0, 1000's of other libraries will stop working. Are you ready to take that risk??

Scipy 1.11.1 was released in June 28, 2023 - that's less than a year old. How is that outdated??

Can you please answer the questions above and I will then act upon your recommendation... Thanks Ram

ecederstrand commented 5 months ago

Hi Ram,

I don't really have an opinion on whether pandas 2.0 is good or bad. I just start seeing package versions that require pandas >= 2, and if I want to use featurewiz along with those packages, then I can't use the latest versions.

I'm not saying that scipy 1.11.0 is outdated, just that newer versions have come out since and featurewiz does not advertise support for those. For example, newer versions of SciPy support Python 3.12, which has nice features that I would like to take advantage of.

My use case is maintaining a virtualenv with a lot of packages (1.403 to be exact) that are used by scientists in a big analytics platform. We try to keep versions updated regularly because our users want that, and featurewiz is one of the four packages we have installed that does not yet support pandas 2.

brandonros commented 1 month ago

This is preventing me from using this otherwise very cool looking library. All sorts of build errors.

bwhartlove commented 1 month ago

I'll second these notions. Trying to install on Python3.12 and held up because of the scikit-learn pin that does not support Python3.12.

rajshah4 commented 1 month ago

I wanted to include this in a feature selection tutorial, but I'm having trouble running it. This library uses Dask, which requires panda>2.0.

AakashKeswani commented 2 weeks ago

@AutoViML any decision on the pandas requirement upgrade? It is blocking me from using it with other libraries that require newer versions