✔️ Linting Passed

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

_{Generated for commit: bbcd695. Link to the linter CI: here}

On a longer term perspective, we could discuss using a c++ implementation from the the stdlib as explained here #27593 (comment). Then the benefit for sklearn would be significant enough to reconsider deprecating murmuhash3.

If something exists in STL for Python, or C++, I suppose, why not use it? I'm not too aware of the issues/nuance here, so if you want to educate me, I'll be happy to learn.

If there is no such reason tho, we should just go for standard implementations to prevent re-inventing the wheel. In the tree/ module, we also had removed a lot of code after Cython allowed cimports of a wider range of C++ STL.

But yeah this also LGTM based on just the merits of the PR itself

If something exists in STL for Python, or C++, I suppose, why not use it? I'm not too aware of the issues/nuance here, so if you want to educate me, I'll be happy to learn.

I think the C++ lib provides an implementation of murmurhash 2 instead of murmurhash 3 we currently use in scikit-learn. We would have to check the performance impact of switching back to 2. Maybe it does not matter. Changing the hash function used in hashing vectorizers will introduce a behavior change (features will be mapped to different columns) but I think it's ok because they are arbitrary anyways.