Hello, I noticed that the cosine similarity function calls safe_sparse_dot, and makes it produce a dense output. Would it be a good idea to expose the dense_output argument of safe_sparse_dot in the signature for cosine_similarity?
I recently had an example where I'm computing the cosine simiilarity of two matrices that are 96% sparse, and the output is only 50% sparse. So maybe the output is not always sparse enough to justify making it sparse, but at least in my case it doesn't raise MemoryErrors. Here's what I'm suggesting - https://gist.github.com/jaidevd/3715846bb729d153694b Thanks -- JD
------------------------------------------------------------------------------
_______________________________________________ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
