Releases: google-deepmind/distrax
Releases · google-deepmind/distrax
Distrax 0.0.2
Changelog [v0.0.2]
Resolves the issue with requirements in v0.0.1.
Merged pull requests:
- Add Hidden Markov Model utility and example. #37 (copybara-service[bot])
- Broadcast tensors manually when sampling from certain distributions. #31 (copybara-service[bot])
- Fix
experimental_shard_axis_names
. #30 (copybara-service[bot]) - Implement required batch shape methods for TFP compatible bijectors. #29 (copybara-service[bot])
- TFP compatibility fix. #26 (copybara-service[bot])
- Update
requirements.txt
. #25 (copybara-service[bot]) - Modify the implementation of the forward and inverse methods for the rational quadratic spline to improve speed when using TPUs. #24 (copybara-service[bot])
- Reduce number of samples in
log_stddev_normal_test.py
. #22 (copybara-service[bot]) - Add more tests in
tfp_compatible_bijector
. #20 (copybara-service[bot]) - Increase tolerance in rational-quadratic spline test. #19 (copybara-service[bot])
- Allow the inner bijector to act on arrays rather than just scalars. #16 (copybara-service[bot])
- Fix a bug in computing the forward and inverse values at the edges of the range in RationalQuadraticSpline. Numerically x and y values at the edges are the same so it doesn't cause any issues, but it was a conceptual bug. #14 (copybara-service[bot])
- Expose math.multiply_no_nan #13 (copybara-service[bot])
- Implemented KL divergence for the Transformed distribution. #12 (copybara-service[bot])
- Updated
flow.py
. #11 (copybara-service[bot]) - Updated
flow.py
. #10 (copybara-service[bot]) - Add installation instructions. #2 (copybara-service[bot])
* This Changelog was automatically generated by github_changelog_generator
Distrax 0.0.1
Changelog
This is a first GitHub release of Distrax!