search

IBM / lale

Library for Semi-Automated Data Science
https://lale.readthedocs.io
Apache License 2.0
334 stars 82 forks source link

DOC: environment requirements #1366

Closed rootsmusic closed 7 months ago

rootsmusic commented 8 months ago

I'm trying to understand the difference between core and full flavors, but I don't see any requirements.txt or pyproject.toml. What are their dependencies and which versions?

hirzel commented 7 months ago

The dependencies and their versions are documented in setup.py.

rootsmusic commented 7 months ago

No @hirzel, some versions are documented in setup.py. For example, I want to watch this repo's releases for when one will support numpy 2.0.0.

shinnar commented 7 months ago

I think there may be some confusion. The setup.py file @hirzel linked to is indeed where we have our required dependencies. As can be seen in that file, there is no upper bound on numpy, and in fact, as of version 0.8.0 of lale, we do indeed support numpy 2.0.0.

The difference between "full" and "core", i.e. lale and lale[full] is that "core" depends on all the functionality needed for core lale functionality, which includes wrappers for many scikit-learn operators.

Lale also includes operator wrappers for operators in a number of third party libraries, such as aif360 and snapml. These will only work if the underlying library is installed. lale[full] installs all of these libraries as well, so lale operator wrappers should "just work". This is helpful for getting started, although more advanced users may want more control over their set of dependencies.

rootsmusic commented 7 months ago

Ok @shinnar, pandas's next release (2.2.2) will support numpy 2.0.0. Please document what are the intended uses for lale[fairness] and lale[tutorial]. Thanks.

shinnar commented 7 months ago

they give you additional dependencies for users interested in wrappers for