Christopher Suter

Thanks @tirkarthi. For the time being, despite not supporting py2 externally, we still need to maintain compatibility for some internal users inside Google. I agree we should eventually take care...

Thanks for putting this together! Still need to take a closer look, but in the mean time can you make the linter happy? (See TravisCI results [here](https://travis-ci.org/tensorflow/probability/builds/540560791?utm_source=github_status&utm_medium=notification))

No worries! Thanks for pitching in, and congrats on your first PR :) Linters in this context perform static analysis of code changes to ensure various style guide constraints are...

@axch i don't think an import error is expected on travis

it looks like TF is masking some of the call stack. can you get a hold of the full stack trace?

This is still not showing the hidden stack frames from within TF code. try this https://www.tensorflow.org/api_docs/python/tf/debugging/disable_traceback_filtering

looks like a legit bug in kronecker linear operator code. there's a code path that doesn't set the `dim` variable at all.

For historic reasons, we've usually shipped a new stable version when TF releases one. I think the TF 2.17 aligned release just slipped through the cracks, as the usual releasers...

@srvasude it looks like the `power` parameter of the PowerTransform bijector is not permitted to be a Variable. Do you know why? https://github.com/tensorflow/probability/blob/v0.23.0/tensorflow_probability/python/bijectors/power_transform.py#L62

indeed, the Exp bijector is just implemented as power transform with power=0. we could decouple these, exp would be very simple on its own. or we could keep the static...