Open mtsokol opened 2 weeks ago
I think I need some guidance which specific files I should modify to add sparse
test dependency.
Thanks for making a PR!
Before working more on the code side I think it would be worth expanding on the use-case and motivation for this change. Maybe there is some document or blog post that explains how this fits into the whole PyData ecosystem. The reason I am saying all this is that adding a new dependency to scikit-learn is one of the hardest things to do. The default answer is "no" - and even if it isn't a straight no, then there are a lot of questions and discussions. So if you have some links to things people can read, find answers and inform the debate that would be great.
In addition to @betatim, I would mention that we recently add support for scipy sparse arrays in addition to the scipy sparse matrices. From what I foresee in the long term is that the scipy sparse arrays implementation should allow us to remove any specific sparse Python code (not the specialize Cython one).
So I'm wondering if this is wise to add support for another sparse container that go away from this goal while we could request our user to make the conversion before to provide us the data, isn't it?
Thanks everyone, I've opened a discussion thread over here: https://github.com/scikit-learn/scikit-learn/discussions/29064
Hi!
This PR adds a simple
scipy.sparse
dispatch for https://github.com/pydata/sparse (transformingpydata/sparse
arrays toscipy.sparse
matrices).Same efforts to support
pydata/sparse
input and convert toscipy.sparse
were completed inscipy.sparse.linalg
andscipy.sparse.csgraph
: