Skip to content

Latest commit

 

History

History
82 lines (52 loc) · 3.03 KB

CHANGELOG.md

File metadata and controls

82 lines (52 loc) · 3.03 KB
Raw

Changelog

Version 0.6.0

  • chore: Remove Python 3.8 (EOL).
  • precommit: Replace docformatter with ruff's formatter.

Version 0.5.3

  • Coerce numpy.generic instances to dtype in the SparseNdarray constructor, for consistency with NumPy functions.
  • Avoid cast to float when operating on uint64 indices in a SparseNdarray.

Version 0.5.2

  • Support dtype= and copy= arguments in __array__(), as required by NumPy v2.
  • Added getters for the RegularTick properties, for convenience.

Version 0.5.1

  • Add support for more numpy functions to dense and sparse DelayedArrays.

Version 0.5.0

  • Switch to Grid classes to control how iteration is performed. This supercedes the previous chunk_shape() generic.

Version 0.4.0

  • Added a buffer_size= option to the apply* functions.
  • Provide converter generics to SciPy classes, with methods for SparseNdarray and sparse DelayedArrays.
  • Converted th to_*_array() functions into generics.
  • Correctly handle zero-extent arrays in the apply* functions.
  • Added methods to compute basic statistics (e.g., sum, variance) from a SparseNdarray.
  • Added a copy method for SparseNdarray.
  • Use block processing to compute basic statistics from a DelayedArray.
  • Do not require Fortran storage order from extract_dense_array().
  • The subset= argument is now mandatory in extract_*_array() calls.
  • Bugfix to respect changes in index dtype upon SparseNdarray concatenation.

Version 0.3.5

  • Support masking throughout the various operations and methods.

Version 0.3.4

  • Fixes to maintain support for Python 3.8.

Version 0.3.3

  • Support the delayed unary logical_not operation.
  • Added utilities like choose_block_shape_for_iteration() and apply_over_blocks() to iterate over the dimensions or blocks.

Version 0.3.2

  • Bugfix for row-wise combining of 2-dimensional SparseNdarrays.

Version 0.3.1

  • Added a wrap() method that dispatches to different DelayedArray subclasses based on the seed.

Version 0.3.0

  • Replace the __DelayedArray methods with generics, for easier extensibility to classes outside of our control.
  • Restored the SparseNdarray class, to provide everyone with a consistent type during sparse operations.
  • Adapted extract_array() into the extract_dense_array() generic, which now always returns a (Fortran-order) NumPy array.
  • Added the extract_sparse_array() generic, which always returns a SparseNdarray object for sparse arrays.
  • Added the is_sparse() generic, which determines whether an object is sparse.
  • Minor fixes to the repr() method for DelayedArray objects.
  • scipy is no longer required for installation but will be used if available.

Version 0.2.3

  • Added a chunk_shape() generic to identify the "best" direction for iterating over the matrix.
  • Added an easy way to compute iteration widths over the desired dimension.
  • Corrected the reported dtype from a delayed Cast.

Version 0.0.3

  • separate dense and sparse matrix classes

Version 0.0.1

  • initial classes for H5 backed matrices