Distrax 0.1.43

v0.1.43

Install dependencies before running version checks in the release wor…

Distrax 0.1.42

v0.1.42

Install dependencies before running version checks in the release wor…

Distrax 0.1.41

What's Changed

• Remove references to deprecated jax aliases by @copybara-service in #243
• Add log_survival_function for Laplace distribution. by @copybara-service in #251

Full Changelog: v0.1.4...v0.1.

Distrax 0.1.5

What's Changed

• Remove references to deprecated jax aliases by @copybara-service in #243
• Add log_survival_function for Laplace distribution. by @copybara-service in #251

Full Changelog: v0.1.4...v0.1.5

Distrax 0.1.4

What's Changed

• Update categorical.py by @cyprienc in #236
• Drop python 3.7 and 3.8

New Contributors

• @cyprienc made their first contribution in #236

Full Changelog: v0.1.3...v0.1.4

Version v0.1.3

What's Changed

• Adjust tolerance in gumbel_test.py. by @copybara-service in #135
• Add an abstract class for linear bijectors. by @copybara-service in #136
• Expose linear bijectors, shift bijector, and general multivariate normal. by @copybara-service in #138
• Add missing symbols in all and put list in alphabetical order. by @copybara-service in #139
• Added log_cdf method for the Gamma distribution and modified some tests in gamma_test.py. by @copybara-service in #141
• Fix test_convert_seed. by @copybara-service in #142
• Implement the Beta distribution. by @copybara-service in #140
• Migrate RLax squashed gaussian to use Distrax. Explicitly broadcast shapes in Distrax scalar affine to avoid rank promotion errors. by @copybara-service in #143
• Remove randomness in laplace_test.py. by @copybara-service in #144
• Implement the Dirichlet distribution. by @copybara-service in #146
• Add survival and log-survival function to distrax.Distribution base class. by @copybara-service in #148
• Implement survival and log-survival function for the Logistic distribution. by @copybara-service in #150
• Implement survival and log-survival function for the Normal distribution. by @copybara-service in #151
• Raise ValueError when input parameters are not valid for Bernoulli, Multinomial, and Categorical. by @copybara-service in #153
• Remove RTOL from all distribution tests. by @copybara-service in #152
• Documentation changes to LogStddevNormal. by @copybara-service in #154
• Improve numerical stability of prob/log_prob computation of distributions.Quantized. by @copybara-service in #149
• Improve the tests in beta_test.py. by @copybara-service in #156
• Remove the median of the Bernoulli distribution. by @copybara-service in #157
• Change tolerance of sample stddev test in bertnoulli_test.py. by @copybara-service in #160
• Delete references to removed jax.core attributes. by @copybara-service in #166
• Update .pylintrc by @copybara-service in #170
• Add the von Mises distribution to distrax. by @copybara-service in #169
• Allow Distribution to work with ArrayTree events. by @copybara-service in #175
• Explicitly separate JAX and non-JAX data during Jittable serialization. by @copybara-service in #177
• Add PyPI badge showing the latest released version. by @copybara-service in #179
• Add vmap support for Categorical, Normal, and Independent. by @copybara-service in #180
• Use jax.tree_util.$tree_fn instead of deprecated jax.$tree_fn alias. by @copybara-service in #181
• Compare with slightly less numerical precision. by @copybara-service in #183
• Avoid NaN gradients from Categorical KL and entropy. by @copybara-service in #190
• Add categorical-uniform distribution. by @copybara-service in #199
• Add categorical-uniform distribution. by @copybara-service in #205
• Add clipped distributions to distrax. by @copybara-service in #210
• Lower the requirements to test the VonMises distribution. by @copybara-service in #217
• Increase tolerance of test in gumbel_test.py. by @copybara-service in #221
• Prepare inverse transformation for JAX jit==pjit migration. by @copybara-service in #223
• Fix TypeErrors for python < 3.10. Fixes #224. by @copybara-service in #225

Full Changelog: v0.1.2...v0.1.3

Distrax 0.1.2

What's Changed

• Implement TFP's base measure interface for distrax distributions and bijectors. by @copybara-service in #104
• Add a matrix property to affine bijectors that returns a matrix representing their linear part. by @copybara-service in #114
• Fix categorical KL computation when the second distribution has zero probability on entries where the first distribution also has zero probability. by @copybara-service in #113
• Implement the DiagAffine bijector. by @copybara-service in #115
• Implement the MultivariateNormalFromBijector distribution. by @copybara-service in #116
• Implement the MultivariateNormalDiag distribution as an instance of a MultivariateNormalFromBijector. by @copybara-service in #117
• Implement the MultivariateNormalTri distribution. by @copybara-service in #118
• Implement the MultivariateNormalDiagPlusLowRank distribution. by @copybara-service in #119
• Add tests for KL computations between different Multivariate Gaussian distributions. by @copybara-service in #123
• Demote TriangularAffine to TriangularLinear. by @copybara-service in #125
• Add tests for same_as method to increase test coverage. by @copybara-service in #126
• Demote DiagAffine to DiagLinear. by @copybara-service in #127
• Demote DiagPlusLowRankAffine to DiagPlusLowRankLinear. by @copybara-service in #128
• Updated 2-distributions tests for the MultivariateNormalDiagPlusLowRank. by @copybara-service in #131
• Adjust test tolerances. by @copybara-service in #133
• Expose the full-covariance multivariate normal distributions, and upgrade version. by @copybara-service in #134

Full Changelog: v0.1.1...v0.1.2

Distrax 0.1.1

What's Changed

• Add an absolute tolerance in multinomial_test to unblock an XLA optimization. by @copybara-service in #74
• Change the default dtype of discrete distributions to unqualified int. by @copybara-service in #77
• Add a straight-through gradient wrapper method. by @copybara-service in #70
• Expose straight_through_wrapper. by @copybara-service in #82
• Update Distrax citation. by @copybara-service in #83
• Ensure forward compatibility with Chex. by @copybara-service in #84
• Relax test tolerances in multinomial_test. by @copybara-service in #85
• Update requirements and allow new versions of JAX. by @copybara-service in #87
• Implement a triangular affine bijector, with tests. by @copybara-service in #88
• Implement LowerUpperTriangularAffine as a composition of two TriangularAffines. by @copybara-service in #89
• Reduce testing time by disabling JAX optimizations. by @copybara-service in #94
• Remove the old venv directory before testing the package. by @copybara-service in #95
• Initial Gumbel distribution and bijector. by @kashif in #36
• Add the Gumbel distribution. by @copybara-service in #96
• Delete unnecessary cdf definition. by @copybara-service in #98
• Update the tests for the Tanh and Sigmoid bijectors. by @copybara-service in #97
• Delete stale "pylint: disable". by @copybara-service in #99
• Added the GumbelCDF bijector originally developed in #36. by @copybara-service in #100
• Add an affine bijector whose weight matrix is a low-rank perturbation of a diagonal matrix. by @copybara-service in #90
• Perform validity checking of probability distribution when sampling from Categorical / OneHotCategorical, and return -1 instead of an invalid sample if normalized probability distribution is invalid. by @copybara-service in #105
• Use a safer log_prob in KL divergence between categoricals. by @copybara-service in #102
• Clarify behaviour of the Transformed distribution in the docstring. by @copybara-service in #109

Full Changelog: v0.1.0...v0.1.1

Distrax 0.1.0

Support for Python 3.6 has been dropped as per JAX deprecation policy. Please upgrade to a supported Python version.

Distrax 0.0.3

It is the latest version compatible with Python 3.6. See deepmind/optax#222 for more details.

Full Changelog

Closed issues:

• ImportError: cannot import name 'partial' from 'jax.util' #60
• Problem with shapes in simple transformed distributions #42
• Issue with jax version in requirements.txt #38
• Duplication/Forking vs Collaboration #35
• Tanh numerical instability #7

Merged pull requests:

• Temporarily disable py3.9 tests. #69 (copybara-service[bot])
• Minor updates to docstring of bijectors. #68 (copybara-service[bot])
• Minor updates to slicing tests. #67 (copybara-service[bot])
• Add note for Tanh and Sigmoid. #66 (copybara-service[bot])
• Support slicing on more distributions. #65 (copybara-service[bot])
• [JAX] Update JAX users in preparation for a change that makes iteration over a JAX array return JAX arrays, instead of NumPy arrays. #63 (copybara-service[bot])
• Fix typo in PRNGKey. #62 (copybara-service[bot])
• Fix import errors & freeze the latest compatible JAX version. Fixes #60 #61 (copybara-service[bot])
• Add slice to more distrax distributions. #59 (copybara-service[bot])
• Add slice operation to Distrax Distribution and implement for TFP, MVN and Categorical. #58 (copybara-service[bot])
• Update tests.yml workflow. #57 (copybara-service[bot])
• Add test.sh for launching CI tests on a local machine. #56 (copybara-service[bot])
• Use TFP nightly builds in CI tests. #55 (copybara-service[bot])
• Fix type of event_shape_tensor in Distrax-to-TFP distribution adapter. #54 (copybara-service[bot])
• Uncomment test cases. #53 (copybara-service[bot])
• Fix breakage in the TFP-to-Distrax distribution adapter. #52 (copybara-service[bot])
• Disable use-list-literal refactoring lint tests. #51 (copybara-service[bot])
• Remove use-dict-literal from the set of refactoring lint tests. #50 (copybara-service[bot])
• Adjust tolerance locally to fix failing test. #48 (copybara-service[bot])
• Change explicit string formats to f-strings to mitigate github pylint failures. #47 (copybara-service[bot])
• Replace uses of jax.partial and jax.util.partial with functools.partial, in preparation for removing jax.partial and jax.util.partial. #46 (copybara-service[bot])
• Remove users of jax.api.* symbols, in preparation for removing the deprecated jax.api name. #44 (copybara-service[bot])
• Implement a numerically stable version of the quadratic formula used in the rational quadratic spline bijector. #43 (copybara-service[bot])
• Distrax: Enable Categorical to be used with a single element. #41 (copybara-service[bot])
• Replace use of deprecated jax test utility #40 (copybara-service[bot])
• initial gamma distribution #8 (kashif)

* This Changelog was automatically generated by github_changelog_generator