Stergios

@fehiepsi Hello! Whenever you want and you are ready take a look. Thank you!

Through the prism of controlled randomness in JAX, it sounds more pleasing to have consistent samples with other libraries. I do understand what you mean but that does not change...

I understand what you mean. I am only skeptical about the situation where a small number of samples (100~1000) is needed. In this case the discrepancy is the biggest and...

Changed topic title since all distributions (or most of them) do not have an entropy method.

I am actively working on it yes. Let's colab if you want @yayami3

I am working on discrete ones now. I added entropy as a method and not a property so it matches other python modules like Distrax and TFP. I have done...

I am also adding a mode property for the distributions.

Great catch there @fehiepsi There is an issue with the gradients when using the `xlogy`. I set up a toy problem to test gradients and the `xlogy` implementation failed at...

Yeah I was blind, I see the issue. This is the clipping: ```python def entropy(self, eps=1e-9): """Calculates the entropy of the Bernoulli distribution with probability p. H(p,q)=-qlog(q)-plog(p) where q=1-p. With...

Ideas about what to return when logits are very negative in a Geometric distribution. As you can see from the code below we need to divide with p and when...