James Chapman

tl;dr you're right on the leaky relu theoretically speaking and probably practically too. I agree that all the eigenvalues should be positive semidefinite (everything is a covariance matrix, covariance matrices...

Yeah will do when I get a moment

A fairly common reason for this is if you are using batch size < latent dimensions. The classic DCCA, DMCCA, DGCCA algorithms will all break. I recommend giving our DCCA-EY...

Thanks @JohannesWiesner - to be honest I've documented both PRCCA and GRCCA really badly sorry. Will take a look but first glance I think you're right. Think I'll go with...

Hi, I'm just looking into this. Reproducible script would be great!

So one thing to say is in more recent versions I have changed SCCA_PMD to SPLS. Later versions should in general be much better/fewer bugs. I'd need to see where...

Looks like a scaling thing a bit like #162 - the correlations of the weights are >0.90 or higher. I'm pushing an update that will ensure that for cca_zoo models...

  putting weights on the same scale

Would be good to get a bit closer though I'll take another look at the implementation.

ok think I've found the source of the error. Will let you know the results soon.