#Changelog
All notable changes to the PySparse package will be documented in this file.
- Fixed bug - chunk size being compared to a tuple instead of an integer
- Fixed problem with small arrays with large chunksizes
- Fixed ongoing multi-index issues
- Accidentally left njit commented out!
- Progress bar fixed to always finish with 100%
- Fixed single-index bug at axis=0
- Fixed multi-indexing bug
- Fixed further bug around multi-index slicing, and in doing so, tidied up some unnecessary Class code
- Fixed bug whereby the first value was being lost when converting to_sparse
- Fixed bug whereby certain values were being lost in to_sparse due to a dictionary error
- Added .shape attribute alongside .dense_shape and .coords_shape
- Fixed bug with indexing 1D data
- README updated appropriately
- Small dependencies bug, now resolved
- Additional "coords_dictionary" file included, which replaces the previous find_indices function and introduces significant loading speed-up of 2-3x. This version is not backwards compatible with arrays encoded using v0.*
- N/A
- README tidied up a little.
- One of the core to_sparse functions are now jitted - this introduces a slight speed-up. Reading from slow I/O formats (.h5) still takes a long time to convert to sparse.
- tqdm now replaced with an in-house progress bar to reduce dependencies
- find_indices now replaced with a static coords_dictionary.pkl file.
- N/A
- N/A
- Further optimised indexing through use of parallel find_indices search, with x2 speed-up.
- N/A
- load_sparse() can now take either a single path to the parent directory of the three arrays, or individual paths to each
- if using a single path to parent directory to load the sparse arrays, the files must be kept named as the standard 'sparse_data.npy', 'sparse_coords.npy', and 'dense_shape.npy' filenames
- Fixed issues with indexing for 1D arrays / rows where the values are all zero
- to_sparse() default behaviour will now convert an array in one go if the array is already all in memory (i.e., np.ndarray), or else via chunksize if the array is memory-mapped (np.memmap).
- N/A
- N/A
- Reduced dependency for numba to <0.57 due to numba/numpy errors.
- Fixed missing parameter in __calc_sparse_shape()
- Changed init.py for a cleaner interface - just two options! (to_sparse and load_sparse)
- to_sparse() now creates directory as part of savepath
- N/A
- N/A
- N/A
- Re-uploaded due to broken compile
- N/A
- N/A
- Fixed issues with dependences
- N/A
- N/A
- Initial release of
PySparse
- N/A
- N/A
- N/A