What’s New¶
v0.4.1 (2019-02-02)¶
- Fixed build regression in readthedocs
v0.4.0 (2019-02-02)¶
- Moved
recursive_diff
,recursive_eq
andncdiff
to their own package recursive_diff - Fixed bug in
proper_unstack()
where unstacking coords with dtype=datetime64 would convert them to integer - Mandatory flake8 in CI
v0.3.0 (2018-12-13)¶
- Changed license to Apache 2.0
- Increased minimum versions: dask >= 0.19, pandas >= 0.21, xarray >= 0.10.1, pytest >= 3.6
- New function
proper_unstack()
- New functions
recursive_diff
andecursive_eq
- New command-line tool
ncdiff
- Blacklisted Python 3.7 conda-forge builds in CI tests
v0.2.2 (2018-07-24)¶
- Fixed segmentation faults in
to_csv()
- Added conda-forge travis build
- Blacklisted dask-0.18.2 because of regression in argtopk(split_every=2)
v0.2.1 (2018-07-22)¶
- Added parameter nogil=True to
to_csv()
, which will switch to a C-accelerated implementation instead of pandas to_csv (albeit with caveats). Fixed deadlock in to_csv as well as compatibility with dask distributed. Pandas code (when using nogil=False) is not wrapped by a subprocess anymore, which means it won’t be able to use more than 1 CPU (but compression can run in pipeline). to_csv has lost the ability to write to a buffer - only file paths are supported now. - AppVeyor integration
v0.2.0 (2018-07-15)¶
- New function
xarray_extras.csv.to_csv()
- Speed up interpolation for k=2 and k=3
- CI: Rigorous tracking of minimum dependency versions
- CI: Explicit support for Python 3.7
v0.1.0 (2018-05-19)¶
Initial release.