Releases: dask-contrib/dask-awkward
Releases · dask-contrib/dask-awkward
2023.2.2
What's Changed
Misc
- [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci in #170
Bug fixes
- Fix layer order assumption in optimization by @douglasdavis in #173
- Protect against empty project_columns argument by @douglasdavis in #175
Full Changelog: 2023.2.1...2023.2.2
2023.2.1
Major update
Refactoring of the necessary columns optimization (#148): this was a very large PR that touched a lot of the code base. The dask-awkward configuration was changed to go along with the new optimization that is based on running the task graph on awkward typetracer arrays.
New
Other
- Some pre-commit/CI updates
2023.2.0
New
- Register with
dask.sizeof
(#142) - Use length zero arrays of metadata for metadata tracking when necessary (#143, #162)
- Configuration parameter
awkward.raise-failed-meta
(default isFalse
) controls whether an exception is raised if the the metadata of a new collection cannot be determined automatically. (#162) - Support integer dtype arrays as the
where
parameter ofgetitem
calls (#158) - More awkward namespace operations supported (#146)
Fixes
- Remove need for explicit computation in
dak.with_name
(#150) - Upstream
UnknownScalar
removal (#152) - Relax partition checking for dak.concatenate (#161)
- Remove xfail from a distributed test (#149)
Other
- License as nested text (#157)
- Some pre-commit and CI updates
2023.1.0
New features
- Add
dak.concatenate
withaxis>=0
support. (#125) - Add
dak.unzip
support (#136) - Add
dak.where
support (#133) - Add
dak.with_field
support (#137)
Fixes
- Remove
flatten_records
keyword argument from reducers (remove deprecation warning from upstreamawkward
) (#139) - Fix getitem optimization such that only starting fields can be discovered (#131)
Other
- More documentation (#132)
- Some pre-commit updates
2022.12.0
This is the first official release of dask-awkward!
Please find the documentation at https://dask-awkward.readthedocs.io/