Alexey Radul

You mean that the Travis build fails? It's reasonable to assume that's unrelated to this PR. Is everything in good shape otherwise?

@csuter Oh yes, you're right.

@inhandan You need to add a bazel BUILD target for the new file, like this: https://github.com/tensorflow/probability/blob/master/tensorflow_probability/python/distributions/BUILD#L149 Also add a dependency from "distributions" to it, here: https://github.com/tensorflow/probability/blob/master/tensorflow_probability/python/distributions/BUILD#L103 If you want to...

Reopening on request.

@davmre The reported problem with @tf.function still happens. Are we expecting to fix that? Also, @jeffpollock9: do you need this to work in TF2 under @tf.function? For the record, here's...

For the record, PR #503 was an attempt here; reading the conversation there, it was close, but didn't quite make it over the line. Could be picked up to complete...

No idea. Never heard of `rlwrap` before -- sounds like a nice tool. So, we don't know anything concrete about why this doesn't work; if you investigate and find something,...

I can't reproduce the compiler bug. All three of the examples you have in `stats-tests.dx` seem to work fine: ``` chol_cov_mat : Fin 2=>Fin 2=>Float = [[0.2, 0.], [-0.3, 0.1]]...

Looks like a legit bug, which I minimized and refiled as #1333, but I don't know how to fix it.

By inspection, I expect `IDiv` to have the same problem.