AttributeError: 'AmortizedPSIKT' object has no attribute `logits`.

[plzhan]

Hi mlcolab!

Thank you for open-sourcing this amazing work! I’m deeply interested in the algorithm and have been trying to reproduce it for learning. While implementing it, I encountered an issue and would appreciate your guidance.

At Line 1083:

loss_cat = -gmvae_loss.entropy(self.logits, self.probs) - numpy.log(0.1)

Confusion:

• The variables self.logits and self.probs in AmortizedPSIKT are not defined in the current class or its parent classes, which results in AttributeError: 'AmortizedPSIKT' object has no attribute `logits`.

I'm still learning this architecture and suspect that self.probs and self.logits might be outputs from the GMVAE and Ornstein-Uhlenbeck process respectively. Apologies for my limited understanding—I couldn't fully reconstruct the missing parts, or I might have missed something obvious.

Any hints would be greatly appreciated!

Best regards,
zhanpl01@outlook.com