Changes
- Ensure Future value is in da.from_delayed task graph @TomAugspurger (#11896)
- Fix annotations passed to delayed @fjetter (#11893)
- migrate delayed unpack_collections @fjetter (#11881)
- Remove Pub/Sub references from docs @jrbourbeau (#11891)
- Ensure only classes without custom init are singletons @fjetter (#11886)
- Remove custom initializers for delayed expressions @fjetter (#11888)
- Fix persisting multiple DFs at the same time @fjetter (#11887)
- Avoid always parsing list inputs to
DataFrame.isinas object type numpy arrays @mroeschke (#11869) - Unskip pandas-dev cov/corr tests @TomAugspurger (#11873)
- Hlg blockwise fix @fjetter (#11871)
- Ensure annotations for HLG objects are properly generated @fjetter (#11866)
- Factor out singleton logic from base Expr class @fjetter (#11868)
- Ensure HLGs are using dependencies properly in optimization @fjetter (#11859)
- Ensure dictionaries tokenize deterministically @fjetter (#11867)
- Ensure default dask scheduler only compute what's needed @fjetter (#11861)
- Faster tokenization of pd.RangeIndex @fjetter (#11863)
- Update link to Quansight in community doc @pavithraes (#11860)
- Relax tolerance in autocorr test @TomAugspurger (#11857)
- Use
map_blocksinarray.storeto avoid materialization and dropping of annotations @fjetter (#11844) - Ensure repartition does not trigger memory size computation during lowering (i.e. on the scheduler) @fjetter (#11855)
- Support args and kwargs for rolling aggregations @fjetter (#11856)
- Remove nightly
h5pyfromupstreamCI job @jrbourbeau (#11847) - Ensure HLGExpr tokenize uniquely @fjetter (#11849)
- Do not inject median in describe for pandas 3 @fjetter (#11846)
- Fixed
Expr.__setattr__for subclasses @TomAugspurger (#11845) - Wrap HLGs in an Expr to avoid Client side materialization @fjetter (#11736)
See the Changelog for more information.