Changelog¶

Version 0.3.2¶

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

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.

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.