Jean Kossaifi

Thanks for the push! I agree that we want to move toward an API that provides an argument `solver` so we can specify, e.g. 'ALS' or 'NN-ALS'. However, currently it's...

Maybe this is the occasion to introduce a class `Parafac`, implementing `fit` and `transform`. We could then have a nice consistent API: ```python cp = Parafac(rank=10, solver='ALS') # or 'NN-ALS',...

Yes, this is what I mentioned: in the Tucker case it is easy to transform, we simply return the core. However, in the CP case, we only get the weights...

Thanks, I'll have a look in the coming days.

Thanks for spotting @bakhtos, this should be fixed! I agree, it would be safer to raise an issue rather than silently change the rank. Would be great if you wanted...

It would be nice to have a TensorLy implementation, would you like to take a crack at it? The required functions are already available in TensorLy As a side note,...

Did you get it all to work?

Based on my experience with other frameworks, I believe this would involve some rewriting but I have not worked on this, especially since eager mode is now the default mode...

Great, looking forward to it @craymichael

Hi @craymichael, If the integration was transparent it would be ideal but since it requires wrapping things (e.g. in tf.while_loop, etc) it might make more sense to simply write a...