ML icon indicating copy to clipboard operation
ML copied to clipboard

Sparse matrix factorisation

Open TheCelavi opened this issue 1 year ago • 3 comments

Are there any plans to include some algorithm for sparse matrix factorisation (recommendations)? Thanks!

TheCelavi avatar Nov 11 '24 08:11 TheCelavi

Hi @TheCelavi, we have no plans to offer sparse matrix factorization in the near future. Having that said, I would love to implement sparse matrix factorization - perhaps in Rubix ML 3.0+ after we move to NumPower and then add support for sparse arrays. It's not a matter of want but a matter of funding and community support.

Is this something that you would be able to contribute @TheCelavi?

andrewdalpino avatar Dec 26 '24 17:12 andrewdalpino

I have stumbled upon this: https://github.com/swoole/phpy and started to using https://surprise.readthedocs.io/en/stable/index.html via PHP. I have really good experience with this extension so far, so it seams kinda redundant to write PHP counterpart as it can be seamlessly used with PHP.

TheCelavi avatar Dec 27 '24 09:12 TheCelavi

Hey @TheCelavi thanks for showing me that PHPY library. It looks like we could possibly use that to call PyTorch or even NumPy under the hood to reimplement some of the Learners in Rubix ML. I'm very interested to hear more about your experience so far. How much overhead is added by zval to PyObject conversion? Can you send tensors from PHP to the GPU through Python? Anything else we should know about?

Thanks, appreciate your help!

andrewdalpino avatar Dec 28 '24 21:12 andrewdalpino