Releases: dask/dask
Releases · dask/dask
2024.12.1
Changes
- Fix map_overlap bug where rechunking and trim=False caused inconsistent chunkings @phofl (#11605)
- Avoid reference to bound method in NestedContainer @hendrikmakait (#11608)
- Avoid constructing
NestedContainer
s in case of trivial inputs @hendrikmakait (#11600) - Avoid legacy implementation in read-csv @phofl (#11603)
- Remove legacy DataFrame import @phofl (#11604)
- asarray ignores dtype for array inputs @crusaderky (#11586)
- Add back LLM chatbot to Dask docs @dchudz (#11594)
- Avoid creating trivial DataNodes in graph conversion @hendrikmakait (#11598)
- Don't wrap keys in
TaskRef
inAlias
@hendrikmakait (#11597) - Bump JamesIves/github-pages-deploy-action from 4.6.9 to 4.7.2 @dependabot (#11593)
- Migrate dask array creation routines to task spec @jrbourbeau (#11582)
- Migrate most of dask array random to task spec @jrbourbeau (#11581)
- Do not use local function in
array.push
@fjetter (#11576)
See the Changelog for more information.
2024.12.0
Changes
- Revert "Add LLM chatbot to Dask docs (#11556)" @dchudz (#11577)
- Automatically rechunk if array in to_zarr has irregular chunks @phofl (#11553)
- Blockwise uses
Task
class @fjetter (#11568) - Migrate rechunk and reshape to task spec @phofl (#11555)
- Cache svg-representation for arrays @dcherian (#11560)
- Fix empty input for containers @fjetter (#11571)
- Convert
Bag
graphs to TaskSpec graphs during optimization @fjetter (#11569) - add LLM chatbot to Dask docs @dchudz (#11556)
- Add support for Python 3.13 @phofl (#11456)
- Fuse data nodes in linear fusion too @phofl (#11549)
- Migrate slicing code to task spec @phofl (#11548)
- Speed up ArraySliceDep tokenization @phofl (#11551)
- Fix fusing of p2p barrier tasks @phofl (#11543)
- Remove infra/mentions of GPU CI @charlesbluca (#11546)
- Temporarily disable gpuCI update CI job @jrbourbeau (#11545)
- Use BlockwiseDep to implement map_blocks keywords @phofl (#11542)
- Remove optimize_slices @phofl (#11538)
- Make reshape_blockwise a noop if shape is the same @phofl (#11541)
- Remove read-only flag from open_arry in open_zarr @phofl (#11539)
- Implement linear_fusion for task spec class @phofl (#11525)
- Remove recursion from TaskSpec @fjetter (#11477)
- Fixup test after dask-expr change @phofl (#11536)
- Bump codecov/codecov-action from 3 to 5 @dependabot (#11532)
- Create dask-expr frame directly without roundtripping @phofl (#11529)
- Add
scikit-image
nightly back to upstream CI @jrbourbeau (#11530) - Remove
from\_dask\_dataframe
import @phofl (#11528) - Ensure that from_array creates a copy @phofl (#11524)
- Simplify and improve performance of normalize chunks @phofl (#11521)
- Fix flaky nanquantile test @phofl (#11518)
- Fix tests for new
read\_only
kwarg inzarr=3
@phofl (#11516)
See the Changelog for more information.
2024.11.2
Changes
- Remove only_refs parsing option for TaskSpec @fjetter (#11511)
- Fix upstream ci pandas Series repr error @phofl (#11514)
- Implement
nanpercentile
for dask arrays @phofl (#11505) - Bump JamesIves/github-pages-deploy-action from 4.6.8 to 4.6.9 @dependabot (#11512)
- Add fuse method for TaskSpec @fjetter (#11509)
See the Changelog for more information.
2024.11.1
2024.11.0
Changes
- Add changelog for Dask release @phofl (#11502)
- Minor updates to optional dependencies table @jrbourbeau (#11503)
- Add push for ffill like operations @phofl (#11501)
- Remove func packing for TaskSpec @fjetter (#11496)
- Make tokenization for vindex more efficient @phofl (#11493)
- Cut down runtime of einstein summation test @phofl (#11499)
- Improve test runtime for test_rot90 @fjetter (#11498)
- Disable low level optimization for TaskSpec in Bags @fjetter (#11495)
- Add automatic rechunking to sliding-window-view @phofl (#11479)
- Add
load_stored
kwarg todask.array.store
. @dcherian (#11465) - Fix quantile error in two dimensions @phofl (#11489)
- Bump conda-incubator/setup-miniconda from 3.0.4 to 3.1.0 @dependabot (#11490)
- Update map_blocks docstring @phofl (#11491)
- Fix einsum with empty arrays @phofl (#11488)
- Implement non gil-blocking quantile method @phofl (#11473)
- Use internal keyword for trimming in map_overlap to reduce graph size @phofl (#11486)
- minor dask order refactor @fjetter (#11467)
- Remove empty tasks from map_overlap @phofl (#11483)
- Fixup auto chunks calculation if single chunk goes below 1 @phofl (#11485)
- Fix CI after pandas upstream changes @phofl (#11482)
- Make sure that block_id and block_info don't create extra tasks @phofl (#11484)
- Use repeat to build nearest boundary @j2bbayle (#9666)
- Remove dead code from make_blockwise @fjetter (#11478)
- Patch auto-chunks calculation for rioxarray @phofl (#11480)
- Skip legacy test because of flaky warning @phofl (#11475)
- Unskip a few dask-expr tests @phofl (#11474)
- Keep chunk sizes consistent in einsum @phofl (#11464)
- Improve how normalize_chunks squashes together chunks when "auto" is set @phofl (#11468)
- Fix resolve_aliases when multiple aliases are in graph @phofl (#11469)
- Avoid cyclic import in dask.array @hendrikmakait (#11472)
- Unskip daraframe test @phofl (#11471)
- Improve dask.order performance for large graphs @fjetter (#11466)
- Ensure that
slice(None)
just maps the keys @phofl (#11450) - Fix
Task.__repr__()
of unpickled object @pentschev (#11463) - Use TaskSpec in local dask execution @fjetter (#11378)
- Adjust accuracy in test_solve_triangular_vector @fjetter (#11461)
- Update Aggregation docstring to better reflext the input argument of … @guillaumeeb (#11459)
- Implement fuse option for delayed objects @phofl (#11441)
- Deprecate legacy dask dataframe implementation @phofl (#11437)
- Fix na casting behavior for groupby.agg with arrow dtypes @phofl (#11118)
- Fix behavior of keys_in_tasks for TaskSpec nodes @fjetter (#11445)
- Convert dtype to int instead of np.uint8 for visualising large task graphs @phofl (#11440)
- TaskSpec: Ensure dependencies are not mutated @fjetter (#11438)
- Full support for task spec in dask.order @fjetter (#11347)
See the Changelog for more information.
2024.10.0
Changes
- Ensure
broadcast_shapes()
returns integers, not NumPy scalars. @trexfeathers (#11434) - (fix): sparse indexing @ilan-gold (#11430)
- Task Spec: Ensure arrays are allowed as arguments @fjetter (#11432)
- Ensure that recursively calling tokenize respects ensure_deterministic @fjetter (#11431)
- Task spec: ensure kwargs can have dependencies @fjetter (#11429)
- Explicitly list setuptools as a build dependency in conda recipe @charlesbluca (#11427)
- Zarr-Python 3 compatibility @jhamman (#11388)
- Avoid exponentially increasing taskgraph in overlap @phofl (#11423)
- Unxfail fixed test @phofl (#11424)
- Ensure numba tokenization does not use slow pickle path @fjetter (#11419)
- Tasks - Remove sequence dict classes @fjetter (#11377)
- Bump JamesIves/github-pages-deploy-action from 4.6.4 to 4.6.8 @dependabot (#11408)
- Switch from mambaforge to miniforge in CI @jrbourbeau (#11409)
See the Changelog for more information.
2024.9.1
Changes
- Improve error message for incorrect columns order in meta information @dbalabka (#11393)
- Update gpuCI
RAPIDS_VER
to24.12
@github-actions (#11407) - Bump jacobtomlinson/gha-anaconda-package-version from 0.1.3 to 0.1.4 @dependabot (#11405)
- Switch to using
zarr.open\_array
instead of using thezarr.Array
constructor @jhamman (#11387)
See the Changelog for more information.
2024.9.0
Changes
- Revert "Improve normalize_chunks calculation for "auto" setting" @jrbourbeau (#11385)
- Bump peter-evans/create-pull-request from 6 to 7 @dependabot (#11380)
- Add a Task class to replace tuples for task specification @fjetter (#11248)
- Reduce overhead in tokenize @fjetter (#11373)
- Improve normalize_chunks calculation for "auto" setting @phofl (#11354)
- Bump
bokeh
minimum version to 3.1.0 @jrbourbeau (#11375) - Move
tokenize
to dedicated submodule @fjetter (#11371) - Ensure process_runnables is not too eager in the presence of multiple splits @fjetter (#11367)
- Use
np.min\_scalar\_type
in shuffle @jrbourbeau (#11369) - Write indexing arrays into dask graph to reduce size for multiple xarray variables @phofl (#11362)
- Cast indexer to minimal dtype in shuffle @phofl (#11364)
- Reduce memory usage of dask.order @fjetter (#11361)
- Bump JamesIves/github-pages-deploy-action from 4.6.3 to 4.6.4 @dependabot (#11366)
- precommit autoupdate @fjetter (#11360)
See the Changelog for more information.
2024.8.2
Changes
- Release 2024.8.2 @jrbourbeau (#11359)
- Add changelor entries for shuffle, vindex and blockwise_reshape @phofl (#11350)
- Ensure persisted collections are released without GC @fjetter (#11348)
- Update zoom link for dask meeting @scharlottej13 (#11357)
- Add more docstring examples for
normalize\_chunks
@Illviljan (#11271) - Choose automatically between tasks-based and p2p rechunking @hendrikmakait (#11337)
- Implement blockwise reshape @phofl (#11328)
- Make rechunking in shuffle more intelligent to distribute unevenly if necessary @phofl (#11326)
- Increase visibility of GPU CI updates @charlesbluca (#11345)
- Update
numpy
andpyarrow
versions in install docs @jrbourbeau (#11340) - Fixup dask and distributed dependencies @phofl (#11338)
- Bump
numpy>=1.24
andpyarrow>=14.0.1
minimum versions @jrbourbeau (#11331) - Add
crick
back to Python 3.11+ CI builds @jrbourbeau (#11335) - Preserve chunksizes in vindex @phofl (#11330)
- Fix
dask.array.fft
mismatch with Numpy's interface (add support fornorm
argument) @joanrue (#10665) - Pass additional parameters to
rechunk_p2p
@hendrikmakait (#11319) - Fix docstring formatting for map_overlap @Tao-VanJS (#11332)
- Fix NumPy overflowing for prod on 2.0 @phofl (#11327)
- tensordot: ensure
axes
are positive / add tests for negative axes @joanrue (#10812) - Fix map_overlap with new_axis @dstansby (#11128)
See the Changelog for more information.
2024.8.1
Changes
- Ensure pickle does not change tokens @fjetter (#11320)
- Add changelog entry for reshape and ordering improvements @phofl (#11324)
- Rename chunksize-tolerance option @phofl (#11317)
- Upgrade gpuCI and fix Dask Array failures with "cupy" backend @rjzamora (#11309)
- Implement automatic rechunking for shuffle @phofl (#11311)
- Ensure we test against numpy 2 in CI @jrbourbeau (#11182)
- Revert "Test ordering on distributed scheduler (#11310)" @fjetter (#11321)
- Test ordering on distributed scheduler @fjetter (#11310)
- Add tests to cover more cases of new reshape implementation @phofl (#11313)
- order: Choose better target for branches with multiple leaf nodes @phofl (#11303)
- order: ensure runnable tasks are certainly runnable @fjetter (#11305)
- Fix upstream numpy build @phofl (#11304)
- Make shuffle a no-op if possible @phofl (#11291)
- Keep chunksize consistent in reshape @phofl (#11273)
- Enable slicing with only one unknonw chunk @phofl (#11301)
- Link to dask vs spark benchmarks on dask docs @scharlottej13 (#11289)
- Fix slicing for masked arrays @phofl (#11300)
- array: fix
asarray
for array input withdtype
@lucascolley (#11288) - array: add constants @lucascolley (#11287)
- Ignore typing of return value @phofl (#11286)
- Remove automatic resizing in reshape @phofl (#11269)
- API: expose
np
dtypes indask.array
namespace @lucascolley (#11178) - Drop support for Python 3.9 @phofl (#11245)
See the Changelog for more information.