v0.2.0

Changelog

Breaking Changes

• Minimal python version is now 3.10 in all packages.
• cuet.TensorProduct and cuet.EquivariantTensorProduct now require inputs to be of shape (batch_size, dim) or (1, dim). Inputs of dimension (dim,) are no more allowed.
• cuex.IrrepsArray is an alias for cuex.RepArray.
• cuex.RepArray.irreps and cuex.RepArray.segments are not functions anymore. They are now properties.
• cuex.IrrepsArray.is_simple is replaced by cuex.RepArray.is_irreps_array.
• The function cuet.spherical_harmonics is replaced by the Torch Module cuet.SphericalHarmonics. This was done to allow the use of torch.jit.script and torch.compile.

Added

• Add an experimental support for torch.compile. Known issue: the export in c++ is not working.
• Add cue.IrrepsAndLayout: A simple class that inherits from cue.Rep and contains a cue.Irreps and a cue.IrrepsLayout.
• Add cuex.RepArray for representing an array of any kind of representations (not only irreps like before with cuex.IrrepsArray).

Fixed

• Add support for empty batch dimension in cuet (cuequivariance_torch).
• Move README.md and LICENSE into the source distribution.
• Fix cue.SegmentedTensorProduct.flop_cost for the special case of 1 operand.

Improved

• No more special case for degree 0 in cuet.SymmetricTensorProduct.

List of Pull Requests

• full_tp descriptor by @mitkotak in #26
• Add support for zero batch by @mariogeiger in #27
• Documentation: add pytorch version requirement by @mariogeiger in #36
• Avoid calling irreps.dim and logger in forward (fix for #32) by @mariogeiger in #35
• List as inputs by @mariogeiger in #28
• pre-commit by @mariogeiger in #42
• test on python==3.9 by @mariogeiger in #43
• Compatibility with jit.script and torch.compile: COMPLETE by @borisfom in #40
• Uniformize the python version requirements by @mariogeiger in #47
• fix sh l=0 by @mariogeiger in #49
• cue.IrrepsAndLayout, cue.EquivariantTensorProduct, cuex.RepArray by @mariogeiger in #46
• Working torch.jit.script() and torch.compile() support by @borisfom in #44
• fix documentation errors due to latest merged PR by @mariogeiger in #54
• CI: test doc in PR by @mariogeiger in #55
• Update CHANGELOG.md and fix few documentation details by @mariogeiger in #57
• Attempt for fix issue 53 by @mariogeiger in #56
• Fix SegmentedTensorProduct.flop_cost for 0 inputs (1 operand) by @mariogeiger in #59
• Test batch size zero (needs !77 in backend to be merged) by @mariogeiger in #58
• Removing tensor lists from all APIs by @borisfom in #61
• Fix tests by @mariogeiger in #62
• Added export test for variable batch by @borisfom in #65
• Prepare for next release by @mariogeiger in #64
• Fix shared weights for IWeightedSymmetricTPDispatcher by @mariogeiger in #63
• Update version number by @mariogeiger in #67
• Update headers with new year by @mariogeiger in #66
• Update CHANGELOG.md by @mariogeiger in #70
• Mini fixes in the doc by @mariogeiger in #71

New Contributors

• @mitkotak made their first contribution in #26
• @borisfom made their first contribution in #40

Full Changelog: v0.1.0...v0.2.0

cuEquivariance v0.1.0

First Beta Release