PyTensor Files

What's Changed

Major Changes 🛠

• Make Sort and ArgSort axis a static Op property by @raashish1601 in https://github.com/pymc-devs/pytensor/pull/2462

New Features 🎉

• Make the assumptions system extensible by downstream libraries by @jessegrabowski in https://github.com/pymc-devs/pytensor/pull/2371
• Add initial to max and min by @anishmehta24 in https://github.com/pymc-devs/pytensor/pull/2444

Bugfixes 🐛

• Return FALSE from assumption rules only when the property is ruled out by @jessegrabowski in https://github.com/pymc-devs/pytensor/pull/2461
• Fix AdvancedSubtensor constant-index bounds check in C-backend by @AnKiT-GaRG2 in https://github.com/pymc-devs/pytensor/pull/2473
• Fix non-terminating rewrites that referenced shape of replaced output by @ricardoV94 in https://github.com/pymc-devs/pytensor/pull/2485
• Fix batch-axis AllocDiag facts and remove a redundant Cholesky rewrite by @jessegrabowski in https://github.com/pymc-devs/pytensor/pull/2491

Maintenance 🔧

• Bump numba upper bound to <=0.68.0 (main) by @github-actions[bot] in https://github.com/pymc-devs/pytensor/pull/2452
• Improve numba FusedElemwise full reduction by @ricardoV94 in https://github.com/pymc-devs/pytensor/pull/2440
• Add a generated_src_dir config option for generated source files by @anevolbap in https://github.com/pymc-devs/pytensor/pull/2448
• Return NaN from numba MvNormalRV for a non-finite covariance by @RitiGrover in https://github.com/pymc-devs/pytensor/pull/2441
• Respect numba__fastmath in Numba vectorized loops by @ilgrad in https://github.com/pymc-devs/pytensor/pull/2458
• Eagerly rewrite nested DimShuffle in unbatched Blockwise rewrite by @N-0-MAD in https://github.com/pymc-devs/pytensor/pull/2420
• Numba: SIMD-friendly maximum/minimum by @ricardoV94 in https://github.com/pymc-devs/pytensor/pull/2370
• Numba CAReduce: cap the number of specialized loop nests by @velochy in https://github.com/pymc-devs/pytensor/pull/2453
• Split positive-definite solves with vector b into Cholesky form by @jessegrabowski in https://github.com/pymc-devs/pytensor/pull/2476
• Remove obsolete macOS -ld64 linker flag by @statuser in https://github.com/pymc-devs/pytensor/pull/2432
• Stabilize slogdet computation on large Cholesky-factorized matrices by @jessegrabowski in https://github.com/pymc-devs/pytensor/pull/2471
• fix: index shape per-axis in flatten() so static shapes fold by @AnKiT-GaRG2 in https://github.com/pymc-devs/pytensor/pull/2472
• Raise a clear error when TensorType is created without a shape by @AnKiT-GaRG2 in https://github.com/pymc-devs/pytensor/pull/2486
• Add vectorize rule for Split by @jessegrabowski in https://github.com/pymc-devs/pytensor/pull/2488
• Match structural linalg rewrites through left expand_dims by @jessegrabowski in https://github.com/pymc-devs/pytensor/pull/2464
• Implement ProdWithoutZeros in JAX, PyTorch, and MLX by @jessegrabowski in https://github.com/pymc-devs/pytensor/pull/2489
• Add a TRIDIAGONAL assumption and rank it above pos in solve rewrites by @jessegrabowski in https://github.com/pymc-devs/pytensor/pull/2478
• Tweak determinant optimization and cover extensions to slogdet by @ricardoV94 in https://github.com/pymc-devs/pytensor/pull/2484
• Rewrite cho_solve quadratic forms as whitened sums of squares by @jessegrabowski in https://github.com/pymc-devs/pytensor/pull/2467

New Contributors

• @RitiGrover made their first contribution in https://github.com/pymc-devs/pytensor/pull/2441
• @ilgrad made their first contribution in https://github.com/pymc-devs/pytensor/pull/2458
• @raashish1601 made their first contribution in https://github.com/pymc-devs/pytensor/pull/2462
• @statuser made their first contribution in https://github.com/pymc-devs/pytensor/pull/2432
• @AnKiT-GaRG2 made their first contribution in https://github.com/pymc-devs/pytensor/pull/2473

Full Changelog: https://github.com/pymc-devs/pytensor/compare/rel-3.3.3...rel-3.3.4