I put the key as the first argument for ELBO loss (in general most random functions and classes in the package take key as the first argument, so this is just for consistency).
I renamed epochs to steps, as in train_flow an epoch is composed of multiple steps, so it felt unclear to have epoch mean a single step.
@Tennessee-Wallaceh Could you give this a quick look and check you are happy?
A few minor changes:
train_flow
an epoch is composed of multiple steps, so it felt unclear to have epoch mean a single step.@Tennessee-Wallaceh Could you give this a quick look and check you are happy?