Deprecations

• Move operators specific to SuiteSparse:GraphBLAS to .ss namespaces (and deprecate previous locations)
  • Unary: erf, erfc, frexpe, frexpx, lgamma, tgamma, positioni, positioni1, positionj, positionj1
  • Binary: firsti, firsti1, firstj, firstj1, secondi, secondi1, secondj, secondj1
  • Semirings that use deprecated binary operators such as any_firsti
  • Aggregators: argmin, argmax, first, last, first_index, last_index
  • Use e.g. gb.unary.ss.erf or gb.agg.ss.argmin instead
  • Will be removed in version 2023.9.0 or later

Enhancements

• Support computation on scalars (#358)
  • gb.unary.exp(1)
  • s.ewise_add(t)
• Add conversions to and from pydata/sparse objects (#347)
• Add Matrix.from_edgelist and matrix.to_edgelist to handle (row, col) pairs or (row, col, val) triples (#374)
• Add Vector.from_pairs to handle (index, value) pairs (#374)
• Add default of 1.0 for values in Vector.from_* and Matrix.from_* methods (#375)
• Add contributor guidelines and dev-requirements.txt (#345)
• Add more to FAQ documentation (#367)
• Give informative errors for bad descriptor options (#353)

Maintenance

• Use setuptools-git-versioning for versions; remove use of versioneer.py (#362)
• Determine supported dependency versions (#368)
  • Add script scripts/check_versions.sh to easily see if there are new versions of dependencies (#366)
• Move package metadata and build info from setup.py to pyproject.toml (#366)
• Add config for PyLint (not enabled by default) (#354, #369)
• Add ruff linter to git pre-commit (#354, #373)
• Add black-jupyter to git pre-commit (#366)
• New GitHub Action just for testing imports (#357)
• New GitHub Action for linting (#354, #359)
• Make readthedocs fail if there’s a warning building the docs (#361)
• Add --no-mapnumpy to pytest options and other CI improvements (#365)
• Randomize versions of dependencies to test against (#368, #371)

This is a pre-release; see 2023.1.0

Highlights

• Add implementation-specific descriptors! (#343)
  • Descriptor customization is now fully exposed for all methods via **opts keyword arguments
  • Example: C(axb_method="gustavson", nthreads=8) << A @ B
  • Example: A.reduce_scalar().new(nthreads=4)
  • See graphblas/core/ss/descriptor.py:get_descriptor for SuiteSparse:GraphBLAS descriptors
  • Other GraphBLAS implementations can have their own custom descriptors

Expired deprecations

• mask.mask has been removed; use mask.parent instead (#344)
• matrix.ss.diag has been removed; use vector.diag or Matrix.ss.build_diag instead (#342)
• vector.ss.diag has been removed; use matrix.diag or Vector.ss.build_diag instead (#342)

Enhancements

• Update to require SuiteSparse:GraphBLAS 7.4.0 (#351)
• Remove use of GxB variadic functions (and vararg function) for increased compatibility (#351)
  • For example, we now use typed versions such as GxB_Matrix_option_get_INT32
• Remove pygraphblas conversions (such as matrix.to_pygraphblas) (#348)
  • These have not been tested or working for a long time due to pygraphblas being defunct and unsupported
• Update environment.yml for developers
  • Add sparse and ipykernel (#346)
  • Add many optional dependencies not installed by default (#350)

Bug fixes

• Handle deprecations from NumPy 1.24, such as using np.bool_ instead of np.bool8 (#349)
• Support different architectures such as ARM by removing use of variadic functions (#351)

New contributors

• We welcomed @SultanOrazbayev as a new code contributor this release 🎉

Deprecations

• Deprecate A.ss.scan_rowwise, A.ss.compactify_rowwise, A.ss.selectk_rowwise (#336)
  • And A.ss.scan_columnwise, A.ss.compactify_columnwise, A.ss.selectk_columnwise
  • Use e.g. Matrix.ss.scan(order="columnwise") instead
  • Will be removed in version 2023.7.0 or later

Enhancements

• Add GxB sort methods as e.g. A.ss.sort(op.lt, order="rowwise") (#333)
• Add vanilla recipe for index slicing (#341)
  • This converts a Python slice to a full numpy array of indices
• Improve vanilla recipe of ewise_union to be a lazy expression (#337)
• Add "complete" option to wait methods for multithreaded applications (#334)
• Add DOI from Zenodo to README (#332)
• Add environment.yml in repo to create developer environment (#340)
• Update “Development Status” in setup.py to “Production/Stable” (#339)

Bug fixes

• Fix environment for building docs so environment solving is faster (#338)

Deprecations

• Deprecate from_values and to_values for Vector and Matrix (#331)
  • Use from_coo and to_coo instead
  • This is a major deprecation!
  • from_values/to_values will be removed in version 2023.9.0 or later next year

Enhancements

• Add "suitesparse-vanilla" backend that does not use GxB methods (#307)
  • This still uses SuiteSparse:GraphBLAS, but will help us support other GraphBLAS implementations
• Improve .ss.head() by using GxB iterators (#314)
• Add from_dict and to_dict methods to Vector and Matrix (#313, #323)
• Support min and max builtins as valid operators (#322)
• Add badge for Discord chat to README (#316)
• Update to SuiteSparse:GraphBLAS 7.3.2 (#321)
  • Adds gb.ss.about["openmp"], which indicates whether it was compiled with openmp

Bug fixes

• Update binder environment (#318)
• Fix viz.draw (#320)
• Notebooks updated to not use deprecations (#327)

Breaking changes

• Move get_order, normalize_chunks, and MatrixArray to gb.core.utils
  • We think we’re happy with where things are now, so we’ll try not to move anymore

Highlights

• Update to work on osx_arm64 platform (and maybe ppc64le) (#309)
• Allow Matrix objects and expressions to be used in NetworkX 3.0 via dispatching (#311)
  • By adding __networkx_plugin__ = "graphblas", NetworkX algorithms will soon dispatch Matrix as adjacency graph to graphblas-algorithms for some algorithms

Enhancements

• Allow UDTs to be assigned to every element with update syntax, such as A << (1, 2, 3) (#299)
• Allow array UDTs to be specified more easily as e.g. "INT64[3, 3]" (#300)
• Expose SuiteSparse’s GxB_pack_HyperHash and GxB_unpack_HyperHash (#305)
  • Y = A.ss.unpack_hyperhash()
  • A.ss.pack_hyperhash(Y)
• Add Matrix methods to go to and from DCSR/DCSC (a.k.a. HyperCSR/HyperCSC) (#310)
  • A = Matrix.from_dcsr(compressed_rows, indptr, cols, values)
  • compressed_cols, indptr, rows, values = A.to_dcsc()
• Add gb.core.utils.ensure_type for easier handling of expressions (#312)
  • For example, if you want a vector but allow vector expressions, you can do: v = ensure_type(v, Vector)
• Use GxB_Matrix_reshape for matrix.ss.flatten and vector.ss.reshape and delete unnecessary code (#306)
• Add get_semiring(monoid, binaryop) to gb.semiring.get_semiring (#312)
  • get_semiring already existed in (and will stay in) gb.core.operator

Bug fixes

• Always free SuiteSparse iterator when generator exits (#302)
• Don’t show object format (such as “csr”) in expression repr (to match HTML repr) (#303)

Breaking changes

• Clean up main graphblas namespace by moving files to gb.core
  • This may break existing code that used objects in files that were moved

Enhancements

• Update to SuiteSparse:GraphBLAS 7.3.0
• Rework _repr_html_ for expressions to be nicer (#282)
• Import and export from CSR and CSC using GrB methods as e.g. Matrix.from_csr(indptr, col_indices, values) (#288)
• Import and export from awkward-array as gb.io.to_awkward(x) and gb.io.from_awkward(x) (#280)

Bug fixes

• Fix fmin and fmax issue with numba 0.56.2 and numpy 1.23 (#283)
• Remove IndexUnary and Select operators from gb.op namespace (#286)

Deprecations

• gb.io.plot moved to gb.viz.plot

Enhancements

• Update to SuiteSparse:GraphBLAS 7.2.0
• Add docs website and improve some docstrings (#269, #273, #274)
  • https://python-graphblas.readthedocs.io/en/latest/
• Add ZSTD compression option to x.ss.serialize (#277)
• Add x.ss.reshape methods to Vector and Matrix objects to change the shape while keeping all the values (#277)
  • Uses GxB_Matrix_reshape and GxB_Matrix_reshapeDup
• Add options to not return rows, columns, or values, in x.to_values (#276)
  • rows, _, values = A.to_values(rows=True, columns=False, values=True)
  • indices, _ = v.to_values(indices=True, values=False)
  • Excluding things can be faster when using SuiteSparse:GraphBLAS
• Add clear=False keyword to dup; this is like empty_like (#275)
• Add gb.viz.spy and gb.viz.datashade to plot the sparsity pattern of a matrix (#279)
  • gb.viz.spy uses matplotlib.spy to create a static plot
  • gb.viz.datashade uses datashader via hvplot to create interactive plots

Deprecations

• gb.io.from_scipy_sparse_matrix and gb.io.to_scipy_sparse_matrix (#260)
  • Use gb.io.from_scipy_sparse and gb.io.to_scipy_sparse instead.

Enhancements

• Add gb.ss.config and x.ss.config to control SuiteSparse-specific configuration options of objects (#258)
• Add gb.io.to_scipy_sparse and gb.io.from_scipy_sparse (#260)
• Use Recorder._repr_html_ for rich display (#263)
• Improved UDT to/from string (#256)
  • Serialization with UDTs is still experimental and should not be relied upon!
• Automatic upload to PyPI 🤞

• Add .ss.serialize and .ss.deserialize for Matrix and Vector objects (#200)
  • This exposes SuiteSparse:GraphBLAS (de)serialize and supports compression
  • It also supports UDTs
• Allow boolean objects to be used as value masks (#248)
  • For example, C(C < 5) << expr
• Perform select with a mask or bool argument (#251)
  • For example, A.select(B.S) and A.select(C < 5)
    • These are implemented as recipes
  • A.select(A < 5) will be rewritten to the appropriate select operation (#242)
• Assign many dense items at once, such as v[[1, 2]] = [3, 4] (#250)
  • Implemented as an efficient recipe