What's Changed

• Rerun new version of black on the repo by @ordabayevy in #605
• Scale factors across plate dims in partial_sum_product by @ordabayevy in #606
• Bump to version 0.4.6 by @ordabayevy in #608

Full Changelog: 0.4.5...0.4.6

Minor release supporting numpyro 0.11.0

Minor release supporting jax 0.4.

What's Changed

• Update black by @fritzo in #583
• Extend partial_sum_product() to cases tractable for Gaussians by @fritzo in #584
• update classifiers by @anukaal in #586
• Fix provenance tensor to work with PyTorch 1.11 by @fritzo in #588
• Update to PyTorch 1.11.0 by @fritzo in #589
• Bump to version 0.4.3 by @fritzo in #590

New Contributors

• @anukaal made their first contribution in #586

Full Changelog: 0.4.2...0.4.3

Enhancements

• #543 Added ProvenanceTensor for torch backend
• #543 Added Constant funsor
• #549 Added funsor.recipes.forward_fiter_backward_rsample()
• #553 Added a Precondition interpretation for Gaussians, plus funsor.recipes.forward_fiter_backward_precondition()
• #572 Support Funsors in pprint.pprint() and pdb

Breaking changes

• #568 Switch to sqrt(precision) representation for Gaussian

Minor release updating to PyTorch 1.9

• Added new sum product algorithms (contributed by @ordabayevy)
• Added many new conjugacy patterns
• Automated op registration and made adding new ops easier
• #419 Support ExtendedDistribution
• #432 Support IndependentConstraint
• #397 Allow .reduce(op, vars_not_in_inputs)
• #430 Add Tuple Funsor
• #409 Change sarkka_bilmes_product interface

• #380 Drop support for Python 3.5 and PyTorch <=1.5
• #352 New syntax for funsor.domain.Domains: bint(n) -> Bint[n], reals(*shape) -> Reals[shape]
• #369 New StatefulInterpretation pattern to simplify context-dependent interpretations

• Supports multiple backends: numpy (default), torch, and jax. To change the backend, use funsor.set_backend("torch") or funsor.set_backend("jax").

• Adds dependency on pytest