Handling NAN values in NMF

[PolarBean]

There is this fantastic pull request in sklearn which does not apply the loss value to parts of the matrix which have nan values. scikit-learn/scikit-learn#8474
Is this possible in pytorch-NMF at present?

[yoyololicon]

Hi @PolarBean , thanks for asking.
I have a very brief look at the PR you mentioned.
It looks like it's a weighted version of NMF, and can be handled with masked tensors.
https://pytorch.org/tutorials/prototype/maskedtensor_overview
I'll have a proper look when I have the time, but it sounds doable.

[PolarBean]

very interesting, thank you!