✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: 31720d0. Link to the linter CI: here}

I think this would warrant a new entry, see: https://github.com/scikit-learn/scikit-learn/blob/main/doc/whats_new/upcoming_changes/README.md for instructions.

There's quite a few other places that need to be updated, I would suggest searching for "scikit_learn_data" in the codebase.

I believe I have covered all the files now

Thanks for the PR @norgera. This is a breaking change so I think it needs a deprecation cycle.

Indeed. Here is a link on how we want to go through when it comes to deprecation: https://scikit-learn.org/dev/developers/contributing.html#deprecation

Thank you all for the information.

I've added deprecation to use the original home directory if its detected or manually set, providing a warning + instructions to move files to new cache folder. Otherwise uses new cache folder.

Let me know if this approach is appropriate.

Thanks a lot for tackling this and also adding all the machinery for deprecation. Do we really need to do a deprecation (and introduce a new parameter)? To me the location of a cache directory feels like an internal detail of the data downloader. If the user specified a path explicitly and we somehow changed it then I'd agree we need a deprecation. But for a cache directory that is implicitly set I am less sure. But then again people might be using this to download datasets and then accessing the files they downloaded. So a better way to think of the path is as a default download location, not a cache directory (in which case we should do a deprecation).

But if we think of this as more of a data downloading tool where people can (and should) access the files - should we be using a directory that is explicitly called cache (which to me implies that it can be deleted at random or its structure changed without notice)? For example if my OS asks me to "clean up caches and temporary files?" I think it should always be safe to click yes. But for the use-case where someone uses the downloading machinery to place a file in a specific directory, that is no longer true?

Or am I making this too complicated? What do you all think about this?