v0.4.0
Contributors: Trevor James Smith (@Zeitsperre), Jan Haacker (@j-haacker), Éric Dupuis (@coxipi).
Changes
xsdbanow supports Python3.13. Metadata and CI have been adjusted. (PR/105).- Unpinned
numpyand raised minimum supported versions of a few scientific libraries. (PR/105). - More code that needed to be ported from
xclimhas been added. This includes mainly documentation, as well as testing utilities and a benchmark notebook. (PR/107).
Fixes
- For
fastnanquantile,POT, andxclimhave been added to a newextrasinstall recipe. All dependencies can be installed using the$ python -m pip install xsdba[all]command. Documentation has been added. (PR/105). - Several small
dask-related issues (chunking behaviour, dimension order when broadcasting variables, lazy array preservation) have been fixed. (GH/112, GH/113, PR/114). xsdba.processing.escorenow correctly handles all-nan slices. (GH/109, PR/108).xsdbanow uses directlyoperatorinstead of usingxarray's derivedget_opfunction. A refactoring inxarrayhad changed the position ofget_opwhich caused a bug. (PR/120).- For more than 1000 quantiles,
fastnanquantileis not used anymore, as it would throw an error. (GH/119, PR/123). Groupernow throws an error ifgroup='time'is used withwindow>1. (GH/104, PR/122).
Internal changes
toxhas been configured to test Python3.10 builds againstnumpy >=1.24.0,<2.0in the GitHub Workflow pipeline. Passing thenumpykeyword totox($ tox -e py3.10-numpy) will adjust the build. (PR/105).- Authorship and Zenodo metadata have been updated. Order of contributions is now developers followed by contributors in alphabetical order. (PR/116).
MBCn.adjustnow re-performs the check onrefandhistto ensure they have compatible time arrays (the check is done a second time inadjustsincerefandhistare given again). (PR/118).- Updated
docsdependencies to usesphinx>=8.2.2. (PR/133).