Repository navigation
Supported environments and dependencies
This release was tested in CI and/or nightly testing with the following language versions:
- Python: 3.10, 3.11, 3.12, 3.13, 3.14.5+
- Chapel: 2.5.0, 2.6.0, 2.7.0, 2.8.0, 2.9.0, 2.10.0
Key Changes
I/O Improvements
-
Updated Arkouda to use the community
ParquetMason package rather than its own custom version (#5510, #5525) -
Added support for
tag_dataon the complete data set read byak.read_parquet(), providing the ability to track which file each row was read from (#5506)
Performance Optimizations
-
Added support for aggregated creation of sparse matrices for better performance (#5527)
-
Improved the parallelization of
ak.diff()to improve its performance (#5508) -
Improved the performance of array slices by specializing for non-strided ranges and using whole array operations (#5509, #5519)
Sparse Improvements
-
Added support for a
to_scipy_sparse()routine that converts a sparse matrix to SciPy's representation for the purposes of comparisons and correctness checking (#5527) -
Extended sparse matrices to support
realfloating point types (#5531)
Changes to Testing / CI
-
Added new benchmarks for tracking the performance of:
-
Updated a few tests to remove the
xfail(expected to fail) marker now that they're passing (#5504) -
Added new tests to ensure that Parquet
tag_datasupport continues working (#5506) -
Added a new CI check to help with branch protection (#5521)
-
Updated a numpy API test to avoid direct accesses to
__dict__(#5524) -
Fixed a mypy error caused by stricter pyzmq type checking (#5528)
Portability Improvements
-
Added support for deploying Arkouda on multiple nodes of HPE Cray EX systems using containers (#5523)
-
Improved documentation for Mac Anaconda users (#5522)
-
Fixed a syntax warning that showed up in Python 3.14 (#5520)
Code Refactorings and Cleanups
-
Refactored
ak.apply()to be more concise now that all supported Chapel versions support Python interoperability (#5502) -
Refactored a few log statements that didn't match the style of others (#5503)
Contributors
Thanks to everyone who contributed to this release:
For questions, issues, or upgrade guidance, please open an issue on GitHub.
Auto-generated release notes
What's Changed
- Add chapel 2.9 to the build for CI images by @jabraham17 in #5499
- Refactor two log statements that did not match the style by @jabraham17 in #5503
- Remove xfail from now passing tests by @jabraham17 in #5504
- Refactor ak.apply by @jabraham17 in #5502
- Add benchmarks for ak.diff and slicing ops by @jabraham17 in #5507
- Implement
tag_dataforak.read_parqueton the whole data set by @jabraham17 in #5506 - Improve parallelization of diff by @jabraham17 in #5508
- Optimize array slice performance by @jabraham17 in #5509
- Revert optimization for negOne slice by @jabraham17 in #5519
- Add CI check for status check by @jabraham17 in #5521
- Fix syntax warning about return in finally by @jabraham17 in #5520
- Add CI for Chapel 2.9.0 by @jabraham17 in #5500
- Add line to add Anaconda to PATH in MAC_INSTALL.md by @eric-vo in #5522
- Add containerfiles for HPE-Cray-EX by @arezaii in #5523
- Remove direct access of dict in numpy test by @jabraham17 in #5524
- Migrate Parquet support to Mason package by @eric-vo in #5510
- Add parquet-clean target by @jabraham17 in #5525
- Fix mypy error caused by stricter pyzmq type checking by @jabraham17 in #5528
- Add benchmark for 1 trillion row challenge by @eric-vo in #5526
- Add support for aggregation in the creation of sparse matrices by @benharsh in #5527
- Reorder benchmarks to avoid impacting performance by @jabraham17 in #5529
- Make sparse benchmark problem size scale with numNodes by @benharsh in #5530
- Support for floating point values in sparse matrices by @benharsh in #5531
New Contributors
Full Changelog: v2026.07.15...v2026.10.14