NVIDIA/cuml v26.10.00 on GitHub

What's Changed

🚨 Breaking Changes

Deprecate get_feature_names in favor of get_feature_names_out by @jcrist in #8480
FIX Remove spurious IsolationForest arguments by @betatim in #8486
Rewrite OneHotEncoder and OrdinalEncoder by @jcrist in #8490
Only coerce 1 column 2D outputs to Series for predict*/fit_predict methods by @jcrist in #8512
Raise NotFittedError from unfitted IsolationForest methods by @betatim in #8546
Migrate stream APIs from rmm::cuda_stream_view to cuda::stream_ref by @bdice in #8529

🐛 Bug Fixes

Update container image in build GHA workflow by @viclafargue in #8396
Temporarily exclude GB300 from testing by @KyleFromNVIDIA in #8431
Preserve dtype for NumPy array-protocol inputs by @viclafargue in #8432
Re-enable testing on GB300 by @KyleFromNVIDIA in #8452
Fix sporadic KBinsDiscretizer mismatches with scikit-learn by @viclafargue in #8467
Fix devcontainer cache version updates by @bdice in #8485
Fix support for pytorch inputs by @jcrist in #8491
Add cuTile build dependencies to CUDA 13.3 pip devcontainers by @csadorf in #8560
Fix index misalignment in char_wb ngram padding by @adityaanikam in #8429
Return Euclidean distances from KMeans transform by @Hashim1999164 in #8539
Fix IsolationForest export for scikit-learn 1.10 by @csadorf in #8589
Validate HDBSCAN min_cluster_size by @csadorf in #8588
Harden ARIMA backing-buffer size arithmetic by @csadorf in #8586
Exclude libkvikio from libcuml wheel repair by @csadorf in #8611
Fix KNN neighbor buffer layout bugs by @csadorf in #8619
Fix sparse t-SNE neighbor count clamping by @csadorf in #8621
Check kwargs for sparse input as well in ProxyBase by @jcrist in #8622
Fix KMeans weighted inertia/score and Dask label ordering by @viclafargue in #8618
Handle zero-update numerical stagnation in the SVM solver by @viclafargue in #8609
Implement multiclass SVC with non-uniform class weights by @viclafargue in #8592

📖 Documentation

Update README for NVIDIA cuML branding by @csadorf in #8464
DOCS Replace references to the old GitHub org by @betatim in #8471
Align documentation with NVIDIA cuML branding by @csadorf in #8476
Enable public docs features in CI by @bdice in #8533
Include inherited methods in class API docs by @jcrist in #8553
Use latest intersphinx inventories on main by @csadorf in #8558
Update cuml.accel benchmark results and presentation by @csadorf in #8517
Clarify HDBSCAN min_samples semantics by @viclafargue in #8567
Make intersphinx references release-aware by @bdice in #8549
Consolidate developer documentation in Sphinx by @csadorf in #8608
Consolidate package README and use latest documentation links by @csadorf in #8680

🚀 New Features

More get_feature_names_out support by @jcrist in #8511

🛠️ Improvements

Update RAPIDS.cmake to log source of rapids-cmake by @arhag23 in #8338
Forward merge release/26.08 into main by @jcrist in #8404
Clean up stale cuML multi-GPU build options by @viclafargue in #8137
Simplify RF lower-bound and index types by @RAMitchell in #8383
enforce 'yamllint' checks by @jameslamb in #8405
Register EmpiricalCovariance benchmark algorithm by @csadorf in #8120
Refactor UMAP's optimize_batch_kernel by @jinsolp in #8415
Remove deprecations for 26.10 dev cycle by @jcrist in #8435
Detect and error when the cmake cached and current conda env differ by @betatim in #8035
X-ORG-8423: Publish API docs to docs.nvidia.com by @josephine-wolf-oberholtzer in #8424
Remove deprecated cuml.fil by @jcrist in #8437
Remove the deprecated convert_dtype kwarg by @jcrist in #8436
Simplify dask serializers by @jcrist in #8443
Fix callback handling in cuml.accel by @betatim in #8433
Remove deprecated legacy output types by @jcrist in #8439
Replace rapids-xgboost with xgboost in the CI. by @trivialfis in #8440
Remove cuml.experimental by @jcrist in #8444
Reject negative inputs in chi-squared kernels by @fallintoplace in #8438
Remove ForestInference notebook by @jcrist in #8445
Restore linalg error mode after KernelRidge fallback by @fallintoplace in #8442
X-ORG-8423: Enable docs version picker by @josephine-wolf-oberholtzer in #8449
Filter out FutureWarning from rapids-xgboost by @chyunsu3 in #8453
Skip flaky RandomForestRegressor compatibility check by @csadorf in #8458
Support float64 y natively in KNeighborsRegressor by @jcrist in #8456
ensure nightly builds always produce new packages, expand 'changed-files' lists by @jameslamb in #8448
Document metric-specific **kwds parameters in pairwise_distances by @mmustafasenoglu in #8459
Update CODEOWNERS team names for rapidsai->NVIDIA migration by @josephine-wolf-oberholtzer in #8463
Distributed RF: add gtests, handle empty partitions, fix bug in leaf output by @RAMitchell in #8394
Update to rapids-logger 0.3 by @bdice in #8447
Update onnxruntime tests by @jcrist in #8462
Update PR issue status automation post-migration by @josephine-wolf-oberholtzer in #8473
Support dataframe-like inputs in reflection machinery by @jcrist in #8478
Remove stale single-GPU test xfails by @csadorf in #8450
enforce 'rstcheck' checks by @jameslamb in #8408
wheels: enforce 'abi3audit' checks by @jameslamb in #8474
Use CUML_KERNEL for remaining kernels by @ahmedtaha100 in #8465
Support fitted IsolationForest conversion to scikit-learn by @JulienAu in #8483
Add sklearn.cluster.HDBSCAN acceleration to cuml.accel by @viclafargue in #8472
ci: avoid triggered-by-schedule and triggered-by-merge builds cancelling each other by @jameslamb in #8489
Expand get_feature_names_out support by @jcrist in #8500
X-ORG-8423: publish-api-docs version-map uses vars by @josephine-wolf-oberholtzer in #8494
Fix incorrect async initialisations by @betatim in #8518
Remove all decorator usage of dask.delayed by @jcrist in #8516
CI Use the CUDA sticky errors plugin for cuml.accel tests by @betatim in #8519
CI Mark scikit-learn example as xfail due to upstream bug by @betatim in #8523
enh: rename HDBSCAN CUDA headers by @Rajkaran-122 in #8503
pre-commit: update 'cython-lint' to 0.21.1 by @jameslamb in #8526
Remove native model from Isolation Forest estimators by @betatim in #8493
Add support for joblib 1.6.0 by @jcrist in #8535
Add OneHotEncoder to cuml.accel by @jcrist in #8528
test: isolate library wheel smoke test by @bdice in #8406
Propagate feature names in transform/fit_transform/inverse_transform by @jcrist in #8542
Make cupy.ndarray pickle patch version conditional by @jcrist in #8550
Update link references for rapidsai -> nvidia by @jcrist in #8552
Update Treelite to 4.7.2 by @chyunsu3 in #8522
OneHotEncoder fallback to CPU on bytes inputs by @jcrist in #8551
Adopt CUDA stream compatibility accessors by @bdice in #8564
fix(deps): remove rapidsai from channel resolution to avoid rapids-xgboost by @gforsyth in #8596
make 'numpydoc' dependency conda-only by @jameslamb in #8556
A few fixes for sklearn 1.9.1 by @jcrist in #8613
Rewrite cuml.feature_extraction by @jcrist in #8575
wheels: use system 'tileiras', search site-packages for CMake targets by @jameslamb in #8573
Add cuml.accel support for sklearn.ensemble.IsolationForest by @adityaanikam in #8477
Some fixes for recent stream changes by @jcrist in #8645
Fix scikit-learn examples on sklearn 1.9.1 by @jcrist in #8647
Update several rapids docs links to the nvidia docs equivalent. by @ncclementi in #8640
Fix test_spectral_clustering_pickle by @jcrist in #8651
Add IsolationForest to the API documentation by @betatim in #8653
Restore rapidsai channel for 26.10 by @bdice in #8658
Update Dask RF to use the new distributed algo by @chyunsu3 in #8466
CI Make dataset fetching more robust by @betatim in #8667
Remove "rapids" verbage by @jcrist in #8672
CI Pin breathe to v4 as the new release contains a bug by @betatim in #8699
X-ORG-410: Integrate archived docs into version switcher by @josephine-wolf-oberholtzer in #8726

New Contributors

@arhag23 made their first contribution in #8338
@mmustafasenoglu made their first contribution in #8459
@ahmedtaha100 made their first contribution in #8465
@JulienAu made their first contribution in #8483
@Rajkaran-122 made their first contribution in #8503
@adityaanikam made their first contribution in #8429
@Hashim1999164 made their first contribution in #8539

Full Changelog: v26.10.00a...v26.10.00