Inconsistency between log_q and log_p in Encoder and NormalizingFlowVAE about normalizing-flows HOT 1 CLOSED

Hi @maulberto3,

Sorry for the late reply. I'll go through the other issues you raised as well in the next days.
First of all, the name for the log probability does not affect the behavior of the code, so even though there is an inconsistency, the code should still work as expected.
Whether you denote the density by p or q depends on what you use it for. Since the Encoder is primarily meant to be used in variational inference here, I agree that consistently calling it log_q seems more reasonable since the encoder is typically the distribution to be inferred.
I changed the name in my development branch and will merge this to master together with some other changes soon.
I'm closing this issue, but you are welcome to reopen it if there is still something unclear.

Best regards,
Vincent

from normalizing-flows.