You can clone with
HTTPS or Subversion.
Bug fix release 0.4.1. Closes #9.
The low-level functions nanstd_3d_int32_axis1 and
nanstd_3d_int64_axis1, called by bottleneck.nanstd(),
wrote beyond the memory owned by the output array if
arr.shape == 0 and arr.shape > arr.shape,
where arr is the input array.
Bump development status to beta and version to 0.5.0dev.
Step release date to today! Bump version to 0.4.0.
Make bottleneck work with Cython 0.14.1.
Previously used Cython 0.13.
Bug fixes and unit tests for size zero input arrays.
Six functions faster for axis=None:
median, nanmedian, nanargmax, nanargmin, rankdata,
New function: nanrankdata()
Make new rankdata function support 32 bit OS.
Added new function: rankdata()
Added two optional inputs to benchmark: shapes and nans.
Remove group_nanmean() function.
Bottleneck is dropping the group-by functions
Add infrastructure to create 32 and 64 bit OS source releases.
All functions faster; less overhead in low-level functions.
Bump version to 0.4.0dev.
Set v0.3 release date to today!
Doc work and bump development status from pre-alpha to alpha.
Moving window functions now default to axis=-1 instead of 0.
Yet another speed up of the selector functions.
Less overhead in selector functions. Plus:
Removed window parameter from moving window
selector functions signature. Not needed.
Made all high-level functions faster. Less overhead.
Added two new functions: move_std(), move_nanstd().
Cut unit test time in half.
Added two new functions: move_sum(), move_nansum().
Improve behavior of move_min and move_max wrt NaNs:
This change slows the code but gives the results that
NumPy users would expect: if any NaNs appear in the
window then the result is NaN for that window.
Added a new function: move_mean().
Added new function: nansum().
Bug: benchmark with interger dtype.
Make benchmark output more compact.
The benchmark output was too long and getting
longer as new functions were added.
Merge branch 'reuse'
Test that no slow functions creep in where they don't belong.
Reuse functions (details below):
The low-level cython functions for median and
nanmedian are the same for int dtype since ints
cannot contain NaNs. Same goes for move_nanmin
and move_nanmax. So reuse functions.
Can now specify dtype and axis in benchmark.
Ascending minima algorithm copyright correction.