Poisson .rsample() / reparameterization

1

Why doesn’t torch’s Poisson have an rsample method? Can’t I have the gradients flow through it like in other distribution through reparameterization?