Can you please shed more light on how the pearson correlation of the predicted X0 and X1 computed?

Are they measured against the given X0 and X1 (ground truth samples)?

I would like to know that as well, but I don’t think Xeek Team is keen on sharing that.

Just a note - you don’t use X0 and X1 to generate samples from CVAE, so there is no ground truth to measure against. My guess is that you take linear approximation of all X0 curves and then check if any of the generated ones has Paerson’s < 0.9 against this linear approximation. Then separately same check for X1 curves.

Thanks for the insights. I understand that generated samples are as a result of the given_y and not X. X0 and X1 values have been used to train the CVAE however with y as the condition data. Ground truth is probably not the ideal word as I have used