search

pyro-ppl / pyro

Deep universal probabilistic programming with Python and PyTorch
http://pyro.ai
Apache License 2.0
8.56k stars 986 forks source link

Add continuous values for data in the Bernoulli test fixtures #564

Closed neerajprad closed 6 years ago

neerajprad commented 7 years ago

Ref. #539.

We should either throw an error if the data to be scored is continuous for the Bernoulli distribution or have some continuous samples for testing if we would like to allow this.

If we go with the second option (consistency with tensorflow could be a reason), we should have our test fixtures reflect this by having test data in the range [0, 1].

martinjankowiak commented 6 years ago

@neerajprad @fritzo can we close this because it's being tracked in distributions issues in probtorch?

neerajprad commented 6 years ago

Closing; we can close most distribution issues (except those that track building wrappers/shims on the Pyro side).