PyTensor Files

What's Changed

Bugfixes 🐛

• Fix broadcast/upcast loss in local_add_of_sparse_write and local_read_of_write by @frankbuckley in https://github.com/pymc-devs/pytensor/pull/2350
• Fix numba failure on gradient of chained batched Convolve1d by @juanitorduz in https://github.com/pymc-devs/pytensor/pull/2375

Documentation 📖

• Add API docs for pytensor.graph.traversal by @williambdean in https://github.com/pymc-devs/pytensor/pull/2401

Maintenance 🔧

• Numba: speed up advanced indexing with non-scalar core dims by @ricardoV94 in https://github.com/pymc-devs/pytensor/pull/2345
• Inline SymbolicOps at the start of the fusion pass so their bodies can fuse by @ricardoV94 in https://github.com/pymc-devs/pytensor/pull/2346
• Numba: keep layout information through DimShuffle by @ricardoV94 in https://github.com/pymc-devs/pytensor/pull/2348
• Fix MergeOptimizer's dead guard against merging destroyed variables by @velochy in https://github.com/pymc-devs/pytensor/pull/2355
• Numba: don't materialize index broadcast by @ricardoV94 in https://github.com/pymc-devs/pytensor/pull/2353
• Numba: faster Alloc implementation by @ricardoV94 in https://github.com/pymc-devs/pytensor/pull/2356
• Add MLX dispatch and a Metal kernel for gammainc and gammaincc by @jessegrabowski in https://github.com/pymc-devs/pytensor/pull/2347
• Bump numba upper bound to <=0.67.0 (main) by @github-actions[bot] in https://github.com/pymc-devs/pytensor/pull/2372
• Numba: solve triangular C-contiguous B without copy by @jessegrabowski in https://github.com/pymc-devs/pytensor/pull/2359
• Add MLX dispatch and a Metal kernel for betainc by @jessegrabowski in https://github.com/pymc-devs/pytensor/pull/2368
• Fix test incompatibilities with numpy 2.5 by @juanitorduz in https://github.com/pymc-devs/pytensor/pull/2376
• Add MLX dispatch for CholeskySolve by @jessegrabowski in https://github.com/pymc-devs/pytensor/pull/2377
• Rewrite log(sqr(x)) to 2 * log(abs(x)) to avoid underflow by @jessegrabowski in https://github.com/pymc-devs/pytensor/pull/2380
• Numba: back-port compilation speedup for large tuple/ number of arguments by @velochy in https://github.com/pymc-devs/pytensor/pull/2361
• Fix docstrings that document parameters the functions do not have by @VenishPaneliya in https://github.com/pymc-devs/pytensor/pull/2408
• Numba Join: better memory and runtime performance by @velochy in https://github.com/pymc-devs/pytensor/pull/2362

New Contributors

• @frankbuckley made their first contribution in https://github.com/pymc-devs/pytensor/pull/2350
• @VenishPaneliya made their first contribution in https://github.com/pymc-devs/pytensor/pull/2408

Full Changelog: https://github.com/pymc-devs/pytensor/compare/rel-3.3.0...rel-3.3.1