Skip to content

Releases: yarrow-id/diagrams

PyPI version 0.0.3.1

17 Jul 20:27
Compare
Choose a tag to compare

Changes since 0.0.2.2

New Features

  • Optic functors yarrow.functor.optic
  • Numpy-specific diagram layering yarrow.numpy.layer
  • New class AbstractIndexedCoproduct, which replaces SegmentedFiniteFunction
  • New functions:
    • Diagram.half_spider
    • FiniteFunction.cumsum
    • FiniteFunction.interleave
    • FiniteFunction.cointerleave
  • Operations class now implements __len__ and __iter__ for convenience

Refactoring (API-breaking changes)

  • Replaced all uses of SegmentedFiniteFunction with IndexedCoproduct
  • Move numpy-backed diagrams into yarrow.numpy
  • Tidied up yarrow.functor.functor and (documentation)

PyPI version 0.0.2.2

06 Jun 14:36
Compare
Choose a tag to compare

Main changes are one feature and one bugfix:

  • Adds Diagram.tensor_list method helper. This is O(n) but doesn't speed up to O(log n) the parallel case. Useful if you don't need to run on the GPU.
  • Fix a bug in FrobeniusFunctor which crashed for functors that are not identity-on-objects, and adds a test.

pypi-0.0.2.1

25 May 10:40
Compare
Choose a tag to compare

Minor bugfixes:

  • Computing the universal map for coequalizers of bipartite multigraphs crashed for non-finite signatures
  • pi and pi finite functions of bipartite multigraphs had finite codomains, meaning that coproducts could incorrectly raise an error

Other changes:

  • Added a Type value to array backends, corresponding to the underlying array type. For the numpy backend, this is just np.ndarray.
  • A slightly stricter constructor for finite functions which passes through the underlying table unchanged if its type is equal to the array backend's Type.

PyPI version 0.0.2

17 May 23:37
Compare
Choose a tag to compare

PyPI version 0.0.2 adds the CuPy array backend and proof-of-concept GPU accelerated diagrams.