YCY

A handy option when the user want to have more control on the final result, sounds good to me. We can add some parameters (like EQ/stereo width/loudness... etc) in the...

Have you tried adjusting the beta value? Which beta divergence are you using? You can try beta=2 which is more numerically stable in my experience.

@chadD123321 I have updated this package recently, could you help try again on the newest version? Docs and installation: https://pytorch-nmf.readthedocs.io/en/latest/ Please install directly from github (the pypi version has some...

Hi @jonnor , > Could one connect a neural network to the NFM module, and learn them at the same time? Any example code or tips on how to do...

> Hi @yoyololicon - thank you for the response and example code! In this framework, the loss would be something that compares the output of the NMF (decomposed and re-composed)?...

@schonkopf If the `Vshape` argumet (the first argument) is specified, it will overrided other arguments like `W` and `H`. To use custom template tensor, in your case, you have to...

Hi @schonkopf, In the newest commit e4563b2 I add an warning note in class docstring. Also, I changed the behavior of `trainable_*` arguments to only take effect when template tensor...

@aichifandefan Do you use conda? `fast_transformer` needs full installation of cuda toolkit to build, and conda only provide a partial installation. I can run the project succesfully with cuda 11.3,...

Maybe you can try run more steps and increase the learning rate. The final loss should be around 2.6 in my experience.

Hi @ZhimaoLin, good question! You're right. It's not the same as in the paper. It's just a different interpretation of how convolution works. In the original definition of convolution, given...