v26.08.00

What's Changed

🚨 Breaking Changes

• Switch RandomForest max_depth default to None by @csadorf in #8200
• Link to libcuvs dynamically in wheel builds by @divyegala in #8250
• Cleanup pairwise distances by @jcrist in #8268
• Deprecate convert_dtype by @jcrist in #8300
• Deprecate PorterStemmer by @jcrist in #8304
• Move Lars out of experimental by @jcrist in #8336
• CI Extend time limit for hypothesis tests by @betatim in #8346
• Deprecate cuml.tsa by @csadorf in #8344
• wheels: build CUDA 13 wheels with latest CTK (13.3.0) by @jameslamb in #8345
• Deprecate cuml.datasets.make_arima by @jcrist in #8379
• Deprecate legacy output types by @jcrist in #8389

🐛 Bug Fixes

• Fix pandas 3 string dtype compatibility and CI flakes by @csadorf in #8142
• Fix KDE score_samples symbol exports by @csadorf in #8173
• Fix hinge loss decision-function shape handling by @csadorf in #8245
• Fix clang-tidy failures in CUDA 13.2 CI by @dantegd in #8251
• Match sklearn random forest feature sampling by @RAMitchell in #8239
• Test cuml.accel upstream test suite with sklearn 1.9 by @csadorf in #8207
• Update test suite to work with new pytest release by @dantegd in #8266
• Introduce Safe Arithmetic by @divyegala in #8171
• Cleanup extract_knn_graph by @jcrist in #8272
• Fix RandomForestClassifier sklearn child metadata by @csadorf in #8291
• Fallback unsupported cuml.accel PCA and KernelRidge cases to CPU by @csadorf in #8290
• Fix FastIntDiv tparam by @aamijar in #8299
• Remove sklearn example xfails by @csadorf in #8090
• FIX: handle None transformers in ColumnTransformer get_params/set_params by @Husnain987 in #8292
• Handle OpenML checksum failures in sklearn examples by @csadorf in #8359
• Fix RF feature importances with infinite metrics by @RAMitchell in #8347
• Limit cuml.accel pytest logging to errors by @csadorf in #8360
• Fix --print-status path in benchmark CLI by @fallintoplace in #8320
• fix(clang-format): preserve SPDX comments by @bdice in #8368
• Make LabelEncoder docstring account for cuDF update by @viclafargue in #8376
• Improve multi-GPU KMeans validation by @viclafargue in #8350
• Separate sklearn example compatibility and total pass rates by @csadorf in #8409
• Make FFT t-SNE deterministic when random seed is set by @dantegd in #8123
• Fix Dask logistic regression with non-contiguous class labels by @viclafargue in #8422

📖 Documentation

• Modernize the Python developer docs by @csadorf in #7642
• Update docs to use NVIDIA Sphinx theme by @bdice in #8308
• Restructure and clarify cuml.accel documentation by @csadorf in #8367
• Document list and tuple input support by @csadorf in #8381

🚀 New Features

• Add IPython magic for setting cuml.accel log level by @jcrist in #8231
• [FEA] Add host-streaming support to single-GPU KMeans by @tarang-jain in #8248
• Isolation Forest Implementation by @dantegd in #8226

🛠️ Improvements

• Suppress intentional hardcoded version references by @csadorf in #8113
• Forward merge release/26.06 into main by @csadorf in #8138
• Improve devcontainer startup times by @trxcllnt in #8135
• Fix symbol export by @vyasr in #8037
• Fix a few lingering sklearn compat xfails by @jcrist in #8166
• Add YAML based benchmark definitions for cuML regression runs by @dantegd in #7980
• Report xfail details in Python test output by @csadorf in #8140
• Add SECURITY.md by @jameslamb in #8175
• merge release/26.06 into main by @jameslamb in #8184
• Replace RF global quantile sort with deterministic sampled quantiles by @RAMitchell in #8111
• Large (int64 indices) sparse input support by @jcrist in #8196
• RF: narrow CountBin to double for weighted-training prep by @switch527 in #8132
• fix(ci): fix configuration for breaking change notification workflow by @jameslamb in #8199
• TST Make check_estimator coverage opt-out by @betatim in #8176
• Forward merge release/26.06 into main by @csadorf in #8214
• Remove deprecated sample_weight in naive bayes/sample_weights in hinge_loss by @jcrist in #8221
• Remove deprecated probability argument to SVC by @jcrist in #8223
• Bump scikit-learn minimum to 1.6 by @csadorf in #8091
• Avoid host sync in _repr_mimebundle_ when possible by @jcrist in #8232
• Remove deprecated probability argument to LinearSVC by @jcrist in #8224
• Fix type reflection when input is an array-like by @jcrist in #8234
• Update cuml.tsa to use new validation by @jcrist in #8236
• Unify distributed RF quantile samples by @RAMitchell in #8190
• Remove numba-cuda upper bound by @brandon-b-miller in #8189
• refactor: switch to rapids-artifact-name for consistent artifact naming by @gforsyth in #8174
• Update cuML sklearn tags for scikit-learn 1.6 by @csadorf in #8227
• Refactor RF objectives by objective family by @RAMitchell in #8233
• Update multi-gpu estimators to new input validation by @jcrist in #8241
• Prepare RF bins and objectives for weights support by @RAMitchell in #8247
• A few fixups to multi-gpu implementations by @jcrist in #8249
• Remove obsolete scikit-learn 1.5 compatibility paths by @csadorf in #8253
• Remove stale sklearn example xfail by @csadorf in #8259
• Build and test with CUDA 13.3.0 by @bdice in #8246
• Remove make_monotonic and sorted_unique_labels by @jcrist in #8269
• Cleanup cuml.model_selection by @jcrist in #8270
• Fixup CI by @jcrist in #8280
• Limit high-dimensional cosine KDE property cases by @Federicorao in #8273
• Validate random forest n_streams by @fallintoplace in #8263
• Use main shared-workflows branch by @bdice in #8286
• Mark test_logreg_l1[42-csr_array] flaky by @jcrist in #8289
• Use segmented scan for RF row partitioning by @RAMitchell in #8257
• Consolidate sparse utilities by @jcrist in #8288
• Validate LinearSVC n_streams by @fallintoplace in #8282
• Add C++ RF sample weight training support by @RAMitchell in #8258
• Center RF split thresholds across empty quantile bins by @RAMitchell in #8283
• Remove xfail markers for cuml.accel examples by @jinsolp in #8297
• FIX Make __repr__ show parameters set to non-default values by @betatim in #8229
• Avoid hard-coded SHAP test outputs by @viclafargue in #8264
• CI Run cuml.accel tests in the sklearn nightly test runs by @betatim in #8219
• Simplify reflect decorator by @jcrist in #8295
• Remove cuml.internals.input_utils and remaining legacy ingest paths by @jcrist in #8293
• Remove RMM make_any_device_resource helper usage by @bdice in https://github.com/rapidsai...

v26.06.00

What's Changed

🚨 Breaking Changes

• Migrate RMM usage to CCCL MR design by @bdice in #7951
• Remove unused C API by @divyegala in #8011
• Cleanup and apply new validation to cuml.naive_bayes by @jcrist in #8051
• Apply new validation to metrics.hinge_loss by @csadorf in #8060
• Cleanup LabelEncoder by @jcrist in #8039
• Use the PTDS for (most) cupy operations by @jcrist in #8086
• Deprecate probability parameter on SVC and LinearSVC by @switch527 in #8089
• Adopt nvForest for random forest inference by @hcho3 in #8048

🐛 Bug Fixes

• Fixes for CCCL 3.4 update by @dantegd in #7949
• Set unique CUDA cache path for each pytest-xdist worker by @divyegala in #7948
• Fix IncrementalPCA: handle missing batch_size_ in sparse transform after partial_fit by @csadorf in #8010
• Drop __firstlineno__ on ProxyBase instances by @jcrist in #8045
• Fix KMeans int32 index overflow check by @csadorf in #8067
• Add early validation for n_clusters > n_samples in dask KMeans by @csadorf in #8071
• Fix CI by changing SVC intercept selection kernel dispatch by @dantegd in #8077
• Fix FIL buffer bounds assertions by @csadorf in #8069
• Fix KMeans score test tolerances by @csadorf in #8099
• Add vertex-parallel kernels to UMAP to enforce sequential behavior by @jinsolp in #7974
• Fix wrapping metaestimators in Pipeline in cuml.accel by @jcrist in #8115

📖 Documentation

• DOC Add third-party app example for cuml.accel by @betatim in #8094
• DOC Add hyper-parameter search example by @betatim in #8095
• DOC Add note about onnxruntime bug by @betatim in #8162

🚀 New Features

• Add LSMR solver to Ridge by @jcrist in #7922
• Add LSMR solver to LinearRegression by @jcrist in #7927
• Add sparse input support to ElasticNet/Lasso by @jcrist in #7943
• Support LabelEncoder in cuml.accel by @jcrist in #8082
• Add EmpiricalCovariance estimator by @csadorf in #8074

🛠️ Improvements

• Remove PoolMemoryResource wrapper by @jcrist in #7878
• fix more verify-hardcoded-version findings by @jameslamb in #7891
• Forward-merge release/26.04 into main by @csadorf in #7888
• Forward-merge release/26.04 into main by @csadorf in #7894
• Forward merge from release/26.04 by @jcrist in #7910
• Update CODEOWNERS by @jcrist in #7919
• Forward-merge release/26.04 into main by @gforsyth in #7914
• Forward merge release/26.04 into main by @jcrist in #7936
• Forward-merge release/26.04 into main by @jameslamb in #7937
• Remove deprecations from 26.04 by @jcrist in #7928
• update pip devcontainers' base image tags by @trxcllnt in #7947
• chore: bump codespell version for Python 3.14+ compatibility by @gforsyth in #7925
• Restart xdist workers on corrupted CUDA context by @jcrist in #7954
• chore: remove comment in devcontainer.json that breaks jq by @gforsyth in #7952
• A few linear model test cleanups by @jcrist in #7957
• Support recent hdbscan by @jcrist in #7956
• Re-enabling precomputed kNN on host for UMAP by @viclafargue in #7915
• Add -v option to timeout in CI scripts by @jcrist in #7966
• Mark a few cuml-accel upstream tests as flaky by @jcrist in #7967
• Update to clang 20.1.8 by @bdice in #7968
• Run all scikit-learn examples under cuml.accel in nightlies by @csadorf in #7960
• xfail ARIMA test on 'endog_hourly_earnings_by_industry_missing_exog' dataset by @jameslamb in #7972
• New input validation utilities by @jcrist in #7973
• Improve timeout verbosity and graceful shutdown for pytest CI jobs by @csadorf in #7979
• Use new input validation in cuml.linear_models/cuml.solvers by @jcrist in #7978
• Apply new validation to cuml.cluster by @jcrist in #7984
• Apply new validation to cuml.random_projection by @jcrist in #8009
• Use new input validation infrastructure for cuml.decomposition. by @csadorf in #8006
• Apply new input validation to cuml.multiclass by @mgrauer in #8017
• Deprecate max_depth=16 in RandomForest and add None support by @Nzouh in #7958
• Migrate _classification and _ranking metrics to new input validation infrastructure by @csadorf in #8012
• CI Fix cuml.svm compatibility with scikit-learn 1.9 deprecation of probability= by @betatim in #7981
• Use scikit-learn's array-api to accelerate StandardScaler by @jcrist in #8020
• Apply new input validation to cuml.kernel_ridge by @jcrist in #8007
• Apply new validation to cuml.covariance by @jcrist in #8008
• [FIL] Validate Treelite model input to prevent integer overflow and OOB memory access by @hcho3 in #8016
• Cleanup Lars, apply new validation by @jcrist in #8024
• Add MinMaxScaler, MaxAbsScaler, and PolynomialFeatures to cuml.accel by @jcrist in #8032
• Several ensemble scikit-learn compatibility improvements by @jcrist in #8023
• Remove inertia_check Param from KMeans by @tarang-jain in #8033
• Fix sample_weight handling in KernelRidge by @jcrist in #8040
• Apply new validation to cuml.svm by @jcrist in #8029
• Support subclassing proxy estimators by @jcrist in #8041
• fix(ci): resolve all zizmor findings and add zizmor pre-commit checks by @gforsyth in #8046
• Add check_cudf, and a few other validation changes by @jcrist in #8038
• Fix labeler config by @jcrist in #8047
• Apply new validation to cuml.explainer by @jcrist in #8043
• Apply new validation to metrics.pairwise_kernels by @csadorf in #8050
• Apply new validation to cuml.neighbors by @viclafargue in #8019
• Apply new input validation to metrics.regression by @csadorf in #8044
• Ensure output of classifier predict has aligned index by @jcrist in #8056
• Fix explainer hypothesis test by @jcrist in #8054
• Apply new validation to metrics.confusion_matrix by @csadorf in #8049
• Apply new validation to metrics.kl_divergence by @csadorf in #8057
• Use token.rapids.nvidia.com when issuing S3 bucket creds in devcontainers by @trxcllnt in #8036
• Apply new input validation to cuml.metrics.cluster by @csadorf in #8059
• Apply new validation to cuml.manifold by @jinsolp in #8042
• Cleanup and apply new validation to cuml._thirdparty and cuml.preprocessing by @jcrist in #8052
• Apply new validation to metrics.pairwise_distances by @csadorf in #8065
• Apply new input validation to metrics.trustworthiness by @csadorf in #8058
• A few validation fixups by @jcrist in #8066
• Use cudaStream_t instead of cuda_stream_view in cuml Cython by @vyasr in #8072
• Fix failing IncrementalPCA test by @jcrist in #8080
• Fallback on ensure_all_finite in UMAP on cuml.accel by @jcrist in #8076
• Cleanup TargetEncoder by @jcrist in https://github.com/rapidsai...

v26.04.00

What's Changed

🚨 Breaking Changes

• FIX Raise TypeError when sparse input is not supported by @betatim in #7728
• Add cuml.internals.validation, check_is_fitted checks by @jcrist in #7868
• Add support for feature_names_in_ by @jcrist in #7877
• Validate that X is 2 dimensional by @jcrist in #7889

🐛 Bug Fixes

• CI Fix tests to be compatible with new stricter cudf dtype handling by @betatim in #7762
• Replace cudf.Series.values_host with .to_numpy(). by @csadorf in #7772
• Mark two tests as flaky by @csadorf in #7787
• Mark test test_spectral_clustering_output_shape_type_and_label_count as xfail by @csadorf in #7791
• Add nvjitlink to cuda-toolkit pip extras for cusparse compatibility by @bdice in #7794
• Fix cupy-14 compatibility by @aamijar in #7814
• Fix -Wsign-compare warning in UMAP by @aamijar in #7803
• Fix thrust header by @aamijar in #7817
• Xfail some of the one hot encoder dask test. by @csadorf in #7828
• Fix Dask KMeans performance regression by @viclafargue in #7819
• Disable codecov reports by @csadorf in #7852
• Prevent unprefetchable allocation in SVM by @viclafargue in #7859
• Fix UMAP inverse_transform OOH error by @viclafargue in #7863
• Fix AttributeError when CumlArray index is a pandas RangeIndex by @viclafargue in #7865
• Avoid pickling lambda in dask LabelBinarizer pytests by @aamijar in #7898
• Fix OOM in Dask KMeans by collecting only one model after fit by @viclafargue in #7908
• Fix Dask estimators crashing when workers receive empty partitions by @viclafargue in #7807
• xfail test_umap_outliers by @jinsolp in #7938

📖 Documentation

• Fix documentation for in-notebook cuml.accel activation. by @csadorf in #7767
• [DOC] Add missing parameters to UMAP and NearestNeighbors docstrings by @silasmue in #7632
• docs updates for new RAPIDS branching strategy by @jameslamb in #7795
• DOC Restructure API documentation by @betatim in #7798
• Remove deprecated tree_method=gpu_hist with XGBoost by @hcho3 in #7858

🚀 New Features

• Add cuml.accel support for StandardScaler by @csadorf in #7766
• Add NearestNeighbors.radius_neighbors_graph by @jcrist in #7811
• Implement health checks. by @csadorf in #7854
• SpectralClustering in cuml.accel by @aamijar in #7804

🛠️ Improvements

• Fix int64 Spectral Embedding by @aamijar in #7689
• Replace thrust::tuple with cuda::std::tuple by @miscco in #7700
• CI: Standardize nightly dependencies testing by @csadorf in #7684
• Expand and improve xfail list management tooling by @csadorf in #7607
• CI: Make the cudf-pandas-integration job non-optional by @csadorf in #7704
• CI: Make the nightly dependency tests pass by @csadorf in #7703
• Add optional thread-safety guard for OpenBLAS/OpenMP on high-core systems by @dantegd in #7688
• More std::tuple fixes by @vyasr in #7707
• CI: Test CuPy nightlies by @csadorf in #7695
• Forward-merge release/26.02 into main by @csadorf in #7720
• Ignore warnings caused by sklearn/scipy incompatibility. by @csadorf in #7735
• Bump the ipython test dependencies. by @csadorf in #7737
• Drop Python 3.10 support by @gforsyth in #7731
• tighten wheel size limits, expand CI-skipping logic, other small build changes by @jameslamb in #7738
• Remove numba-cuda upper bound by @brandon-b-miller in #7706
• remove pip.conf migration code in CI scripts, update CI-skipping rules by @jameslamb in #7747
• Use verify-hardcoded-version pre-commit hook by @KyleFromNVIDIA in #7726
• Add instructions for CodeRabbit by @csadorf in #7725
• CI Use ipython 7.34.0 in "oldest" version CI by @betatim in #7743
• Remove deprecations from 26.02 by @jcrist in #7761
• Remove deprecated handle argument by @jcrist in #7751
• CI Switch to strict xfail policy for common estimator checks by @betatim in #7753
• CI Update xfail list by @betatim in #7768
• refactor: build wheels and conda packages using Python limited API by @gforsyth in #7758
• Partially revert #7715. by @csadorf in #7773
• remove gitutils by @jameslamb in #7784
• Use GHA id-token for sccache-dist auth token by @trxcllnt in #7790
• FIX Raise explicit exception for input with complex dtype by @betatim in #7729
• Add option to force serial behavior in UMAP optimize to avoid outliers by @jinsolp in #7793
• UMAP hypothesis testing by @viclafargue in #7323
• Update raft headers by @aamijar in #7752
• Remove raft math.cuh and matrix.cuh by @aamijar in #7797
• Replace uses of thrust::transform_iterator with cuda::transform_iterator by @miscco in #7808
• check-nightly-ci: update to new version by @jameslamb in #7810
• check-nightly-ci: remove testing config by @jameslamb in #7818
• Remove some dead code by @jcrist in #7825
• Drop thrust/functional it is almost not used at all by @miscco in #7823
• Add warning about the use of pickle for model deserialization to README. by @csadorf in #7829
• CI Fix testing references to take into account string dtype by @betatim in #7830
• DOC Add basic tests for ONNX support by @betatim in #7805
• refactor(limited api): add explicit wheel.py-api to pyproject.toml
  by @gforsyth in #7832
• Remove workaround for old cupy.percentile bug by @wphicks in #7789
• Update Cython lower bound pin to 3.2.2 by @vyasr in #7836
• Optimize data transfer for Pipeline in cuml.accel by @jcrist in #7835
• Allow using cuML benchmark tools on systems without cuML installed by @dantegd in #7593
• Fix integer overflow in FIL by @hcho3 in #7727
• Relax UMAP version constraint by @csadorf in #7850
• cuml.accel's pth needs to go after the numba_cuda one by @betatim in #7872
• Use PrefetchResourceAdaptor in cuml.accel by @jcrist in #7834
• Add support for Python 3.14 by @gforsyth in #7831
• Add check_features by @jcrist in #7874
• fix verify-hardcoded-version issues by @jameslamb in #7882
• Cherry-pick "Remove PoolMemoryResource wrapper" into release by @jcrist in #7881
• Update Treelite to 4.7.0 by @hcho3 in #7870
• Remove deprecation warning in TargetEncoder by @jcrist in #7892
• Remove xfailed cudf pandas tests by @jcrist in #7901
• Cap numba-cuda upper bound at <0.29.0 by @brandon-b-miller in #7900
• Remove no_bad_cuml_array_check by @csadorf in #7887
• Bump scipy minimum required version to 1.14.0. by @csadorf in #7857
• fix(devcontainers): build with Python 3.13 while dependencies are unavailable by @gforsyth in #7916
• build wheels with CUDA 13.0.x, test wheels against mix of CTK versions by @jameslamb in #7907
• Fix CDMG xfails for numpy 2.4 by @jcrist in https://github.com/rapidsai/c...

v26.02.00

What's Changed

🚨 Breaking Changes

• Simplify type reflection implementation by @jcrist in #7539
• Deprecate TotalIters for SVC/SVR by @jcrist in #7612
• Deprecate handle from public APIs by @jcrist in #7628
• Don't coerce output_type=None to global output_type in __init__ by @jcrist in #7649
• Remove unused QR decomposition MG code by @JohnZed in #7687
• Use train_test_split from scikit-learn by @betatim in #7667

🐛 Bug Fixes

• Don't link to cumlprims_mg when MG is turned off by @robertmaynard in #7583
• UMAP fix int32 overflow causing illegal mem access by @aamijar in #7587
• Explicitly initialize uninitialized SvmParameter fields in SVC constructor by @dantegd in #7604
• Use a generated dataset in place of the actual Covertype dataset. by @csadorf in #7634
• Remove reliance on remote datasets in tests by @csadorf in #7637
• Use synthetic dataset in BERTopic wheel integration tests. by @csadorf in #7644
• Refactor pointer comparison in check_ptr function for input validation by @csadorf in #7636
• Add XGBoost back to tests by @dantegd in #7653
• xfail check_estimators spectral embedding tests by @aamijar in #7677
• Update RMM memory resource APIs to use ref-based equivalents by @bdice in #7668
• Xfail test 'test_tsne_distance_metrics_on_sparse_input' by @csadorf in #7681
• Use NCCL for Dask kNN and DBSCAN by @viclafargue in #7574
• Mark some SpectralEmbedding() common test_estimators tests as flaky. by @csadorf in #7686
• Handle new sklearn LARS error message in test_typeerror_input by @csadorf in #7694
• Build.sh now support custom LIBCUML_BUILD_DIR value by @robertmaynard in #7582
• Fix UMAP outliers when random_state is given by @jinsolp in #7597
• Work-around sentence-transformer regression in BERTopic integration tests and do not fail test runs on integration tests by @csadorf in #7715
• classlabels utils refactor by @viclafargue in #7691

📖 Documentation

• DOC Rework LinearRegression documentation by @virchan in #7218
• k-means++ documented as a init option by @aamijar in #7615
• Add general XGBoost demo notebook by @dantegd in #7601
• Add security warnings to model serialization/deserialization notebook. by @csadorf in #7661

🚀 New Features

• Use CCCL's mdspan implementation by @divyegala in #7555
• TargetEncoder in cuml.accel by @aamijar in #7476

🛠️ Improvements

• Forward merge release/25.12 into main by @jcrist in #7519
• Use strict priority in CI conda tests by @bdice in #7524
• CI: Do not automatically update the release field of linked issues. by @csadorf in #7521
• Remove verbose descriptor by @jcrist in #7527
• Forward merge release/25.12 into main by @jcrist in #7531
• CI: Improve PR target release automation by @csadorf in #7511
• CI: Add or adjust job-specific timeouts for GHA test jobs by @csadorf in #7533
• Forward-merge release/25.12 into main by @csadorf in #7537
• CI: Fail PR auto assignment gracefully. by @csadorf in #7558
• Use strict priority in CI conda tests by @bdice in #7559
• Adjust CI timeouts by @jcrist in #7563
• use CUDA-suffixed XGBoost wheels by @jameslamb in #7550
• MNT: Remove unused legacy spectral.cu by @mani-builds in #7556
• Remove alpha specs from non-RAPIDS dependencies by @bdice in #7568
• Enable merge barriers by @KyleFromNVIDIA in #7579
• Update to numba-cuda >=0.22.1,<0.23.0 by @brandon-b-miller in #7549
• Forward merge release/25.12 into main by @jcrist in #7592
• Add devcontainer fallback for C++ test location by @bdice in #7596
• Add lower bound to dask-ml>=2024 by @bdice in #7600
• Forward cuml.accel logging to subprocesses by @jcrist in #7602
• Improved condense hierarchy of HDBSCAN by @jinsolp in #7459
• Isolating test_umap_outliers from precomputed kNN issues by @jinsolp in #7594
• Fix scikit-learn 1.8 test incompatibilities by @csadorf in #7589
• Prepare cuml for removal of deprecated raft apis by @aamijar in #7561
• Require scipy>=1.13.0. by @csadorf in #7605
• Remove deprecations for 26.02 by @jcrist in #7613
• Remove deprecated normalize parameter by @jcrist in #7611
• Update cuML to C++20 by @divyegala in #7619
• Avoid deprecated conversion of numpy array->scalar by @jcrist in #7618
• Updating UMAP random_state doc by @viclafargue in #7590
• Add warning in UMAP for nondeterministic behavior when build_algo=nn_descent by @jinsolp in #7620
• FIX Remove custom CUDA kernels from naive bayes estimators by @betatim in #7424
• Bypass data allocation to GPU given precomputed knn in UMAP by @jinsolp in #7566
• Attempt to find cmake-format-rapids-cmake.json in cpp/latest by @divyegala in #7626
• Optional seed spectral embedding by @aamijar in #7608
• Cleanup naive bayes by @jcrist in #7623
• Don't fail on cuda.core.experimental deprecation warnings by @jcrist in #7629
• Empty commit to trigger a build by @bdice in #7630
• Move cumlprims_mg into cuML by @dantegd in #7585
• CI Make nightly cuml.accel integration test stricter by @betatim in #7631
• Use SPDX license identifiers in pyproject.toml, bump build dependency floors by @jameslamb in #7641
• Require scikit-learn>=1.5 by @csadorf in #7606
• remove unused test dependency 'tenacity' by @jameslamb in #7645
• Add CUDA 13.1 support by @bdice in #7610
• Remove cumlprims from clang-format by @bdice in #7647
• Use cuvs spectral clustering dataset api by @aamijar in #7625
• CI: Add notebook tests to nightly tests by @csadorf in #7567
• Address sklearn deprecation warnings by @csadorf in #7371
• Remove cutlass reference in unused header by @divyegala in #7655
• Add LedoitWolf covariance estimator by @will-hill in #7548
• Update input_utils CI tests to account for cuDF update by @viclafargue in #7660
• build and test against CUDA 13.1.0 by @jameslamb in #7650
• Empty commit to trigger a build by @jameslamb in #7666
• Use main shared-workflows branch by @jameslamb in #7682
• Use saga solver in test_logistic_regression test function. by @csadorf in #7670
• Mirror input mem type for nn_descent algo in HDBSCAN by @jinsolp in #7614
• Avoid installing CCCL headers in wheels by @divyegala in #7662
• CI Add a nightly job that uses scikit-learn's nightlies by @betatim in #7621
• UMAP with int64_t spectral initialization by @aamijar in #7586
• UMAP custom init by @viclafargue in #7651
• Support older IPython versions for cuml.accel magics by @jcrist in #7675
• wheel builds: react to changes in pip's handling of build constraints by @mmccarty in #7693
• Use all-neighbors brute force in UMAP and deprecate build_kwds args to be consist...

v25.12.00

What's Changed

🚨 Breaking Changes

• Make dask an optional dependency by @jcrist in #7303
• Using int64_t for HDBSCAN and dropping int by @jinsolp in #7104
• UMAP - delay embedding allocation by @viclafargue in #7313
• Don't infer dtype from model in cuml.explainer by @jcrist in #7358
• Cleanup LinearRegression by @jcrist in #7355
• Cleanup AgglomerativeClustering by @jcrist in #7379
• Cleanup LinearSVC/LinearSVR by @jcrist in #7376
• Require CUDA 12.2+ by @jakirkham in #7408
• Several improvements to Ridge by @jcrist in #7410
• Expose n_iter_ in SVC/SVR by @jcrist in #7461
• Standardize regression predict output dtype by @jcrist in #7464
• Support non-numeric class labels everywhere by @jcrist in #7480
• Rename TSNE n_iter to max_iter by @jcrist in #7500
• Cleanup cuml.multiclass by @jcrist in #7508
• Cleanup SGD/MBSGDClassifier/MBSGDRegressor by @jcrist in #7504

🐛 Bug Fixes

• Avoid pyyaml runtime dependency. by @csadorf in #7312
• Make conversion from scikit-learn SVM to cuML more robust by @csadorf in #7324
• Revert "CI: Auto-assign priority to PRs from linked issues (#7354)" by @csadorf in #7357
• Adding checks for UMAP precomputed KNN feature by @viclafargue in #7300
• Unpin treelite and xfail some tests affected by the XGBoost/treelite incompatibility by @csadorf in #7366
• Fix issue in Dask DBSCAN by @viclafargue in #7359
• CI: Increase nightly CI check window to 14 days. by @csadorf in #7392
• Fix overflow in coordinate descent by @jcrist in #7399
• Fix output argument passed by value causing the copy constructor on owning CSR matrix by @achirkin in #7390
• Fix Dask DBSCAN tree reduction by @viclafargue in #7398
• Fixup the test_kernel_density test. by @csadorf in #7404
• Fetch rapids-cmake configuration for cmake-format in CI style checks by @bdice in #7406
• Fix string slicing dtype mismatch in PorterStemmer for cudf-pandas mode by @csadorf in #7413
• Set the same random_state value for None in cuml.dask.KMeans for all workers by @divyegala in #7417
• Init checks for Dask KMeans by @viclafargue in #7391
• Fix DBSCAN overflow by @divyegala in #7421
• Fix HDBSCAN nn-descent errors with default arguments (min_samples=None) by @viclafargue in #7402
• CI: Xfail test_umap_outliers to mitigate flaky test failure. by @csadorf in #7430
• Ensure n_components < n_cols in cuml.dask.tsvd by @divyegala in #7441
• Skip test_extract_partitions_shape by @csadorf in #7453
• refactored update-version.sh to handle new branching strategy by @rockhowse in #7455
• [BUG] Fix RandomForest Builder Sampling by @tarang-jain in #7422
• Fix the RandomForest sampling bias by @csadorf in #7449
• Pin Cython pre-3.2.0 and PyTest pre-9 by @jakirkham in #7472
• Remove SpectralEmbedding from xfail list by @csadorf in #7487
• Fix UMAP Outliers for small data by @jinsolp in #7473
• Fix PCA sign flip by @zhuxr11 in #7331
• A few small fixes for multi-gpu PCA/TruncatedSVD by @jcrist in #7560
• Remove unnecessary syncthreads in UMAP optimize by @jinsolp in #7575
• Revert precomputed kNN on CPU for UMAP by @jinsolp in #7576
• Pin scikit-learn<1.8 in test dependencies by @csadorf in #7588

📖 Documentation

• Add Spectral Embedding algorithm to README.md by @aamijar in #7348
• Fix and improve build docs by @csadorf in #7383
• Mention pip as install path in README. by @csadorf in #7414
• Use current system architecture in conda environment creation command by @bdice in #7447
• Use sklearn.datasets.fetch_covtype in example notebook by @jcrist in #7581

🚀 New Features

• Provide a static target for libcuml by @divyegala in #7351
• HDBSCAN with NN Descent build option by @jinsolp in #7339
• [FEA] Implement kfold. by @trivialfis in #7296
• [FEA] Feature Importances for Random Forests by @tarang-jain in #7275
• SpectralClustering estimator by @aamijar in #7372

🛠️ Improvements

• Update RAPIDS_BRANCH, codify changes in update-version.sh by @KyleFromNVIDIA in #7268
• Avoid usage of deprecated cudf from_pandas by @TomAugspurger in #7284
• Branch 25.12 merge branch 25.10 by @jcrist in #7291
• Update to rapids-logger 0.2 by @bdice in #7301
• Cleanup TSNE python implementation by @jcrist in #7311
• Cleanup cuml.decomposition by @jcrist in #7316
• A few HDBSCAN cleanups by @jcrist in #7319
• Run the scikit-learn test suite in parallel again by @csadorf in #7315
• A few cuml.neighbors cleanups by @jcrist in #7320
• Cleanup DBSCAN by @jcrist in #7325
• Release GIL in SpectralEmbedding by @jcrist in #7326
• Cleanup *.pxd files and cimports by @jcrist in #7327
• Accelerate linear model predict on C-ordered inputs by @jcrist in #7329
• Remove cuml.ensemble/cuml.dask.ensemble deprecated features by @jcrist in #7332
• Cleanup Ridge by @jcrist in #7330
• Add SparseRandomProjection, AgglomerativeClustering and GaussianRandomProjection to common checks by @betatim in #7307
• Enable sccache-dist connection pool by @trxcllnt in #7344
• Use pinned_host_memory_resource instead of pinned_memory_resource. by @bdice in #7340
• Propagate Dask/UCX exceptions by @viclafargue in #7308
• Add libcuml linkage smoke test by @divyegala in #7343
• Cleanup SVM by @jcrist in #7347
• Remove outdated link flags from CMakeLists.txt by @hcho3 in #7349
• Fix flaky test_logistic_regression_weighting test by @csadorf in #7321
• CI: Auto-assign priority to PRs from linked issues by @csadorf in #7354
• Reduce warnings verbosity in tests by @csadorf in #7322
• Add xgboost to benchmark utilities by @dantegd in #7350
• Use main in RAPIDS_BRANCH by @bdice in #7364
• Use main shared-workflows branch by @bdice in #7369
• Use SPDX for all copyright headers by @KyleFromNVIDIA in #7373
• Add commit to use SPDX copyright headers to git ignore blame revs. by @csadorf in #7381
• CI: Fix the auto-release assignment GA job. by @csadorf in #7380
• Cleanup ElasticNet, Lasso, and CD by @jcrist in #7382
• Add cuVS to python test dependencies by @jinsolp in #7385
• Support KernelDensity in cuml.accel by @jcrist in #7397
• Deprecate normalize option to all linear models by @jcrist in #7415
• Expose n_iter_ in LinearSVC/LinearSVR by @jcrist in #7419
• Bring back KMeansMG by @jcrist in #7420
• Remove reachability.hpp Header by @tarang-jain in #7416
• Add Out-of-Bag (OOB) Score Support to RandomForest by @csadorf in #7401
• Add Third-Party Integration Tests ...

v25.10.00

🚨 Breaking Changes

• Deprecate convert_to_* methods in favor of as_* methods in cuml.ensemble (#7254) @jcrist
• KernelDensity.score_samples output type & dtype fixes (#7240) @jcrist
• Remove get_json/get_detailed_text/get_summary_text from ensemble estimators (#7177) @jcrist
• Deprecate accuracy_metric (#7170) @jcrist
• Deprecate predict_model in cuml.ensemble/cuml.dask.ensemble (#7155) @jcrist
• Remove deprecated cuml.accel CLI options (#7110) @jcrist
• Remove deprecation warnings in 25.10 (#7109) @jcrist
• Remove deprecation warning and data_on_host option for UMAP (#7099) @jinsolp
• Fix UMAP graph thresholding (#6595) @viclafargue

🐛 Bug Fixes

• Set NCCL rpath for cu13 wheels (#7304) @divyegala
• Add NCCL library path to CMakeLists.txt for libcuml (#7281) @csadorf
• Ensure finite values passed to SpectralEmbedding (#7280) @jcrist
• Ensure sklearn roundtrip attribute consistency (#7278) @jcrist
• Fix SpectralEmbedding precomputed option with 0% sparsity (#7271) @aamijar
• Fix test reference object in test_onehot_inverse_transform_handle_unknown (#7246) @mroeschke
• Slightly increase tolerance in ridge regression test. (#7243) @csadorf
• Increase rtol for float32 in test_complement_partial_fit to reduce intermittent failures (#7237) @csadorf
• Conditionally xfail test_umap_fit_transform_score for nrows=500. (#7232) @csadorf
• Fixup StandardScaler.n_samples_seen_ (#7209) @jcrist
• Validate sample_weight in KernelDensity.fit (#7208) @jcrist
• Support non-contiguous input in input_to_host_array (#7207) @jcrist
• Set correct min_samples for HDBSCAN single_linkage (#7195) @tarang-jain
• Raise Proper Exception when HDBSCAN min_samples > number of samples (#7193) @tarang-jain
• Remove docs/handling of solver="cd" in Ridge (#7190) @jcrist
• Error nicely in KMeans if n_samples < n_clusters (#7189) @jcrist
• Further increase tolerance in test_random_seed_consistency. (#7180) @csadorf
• Use custom plugin to download test data early (#7169) @betatim
• Fix UMAP outlier issue by checking for outliers and shuffling (#7131) @jinsolp
• Note non-determinism in random projection transforms (#7129) @jcrist
• Rewrite random projection estimators (#7119) @jcrist
• Prevent CUDA issues when running UMAP with scikit-learn compatibility test suite (#7107) @viclafargue
• Fix accel profiler 2x count of some fallback methods (#7101) @jcrist
• Don't call to_output on a cupy array (#7044) @Matt711
• Add fix for devices that do not have memory resources (#6823) @viclafargue
• Fix UMAP graph thresholding (#6595) @viclafargue

📖 Documentation

• Add warning on subprocesses in cuml.accel profilers (#7290) @jcrist
• Revise the cuML docs for 25.10 (#7228) @csadorf
• Provide GitHub issue template to report CI failures. (#7178) @csadorf
• Fix cuml-accel docs redirects (#7139) @jcrist

🚀 New Features

• Update cufft defines to CUDA 13 types (#7094) @robertmaynard

🛠️ Improvements

• Fix missed deps in update-version.sh (#7285) @AyodeAwe
• Empty commit to trigger a build (#7264) @msarahan
• Enforce correct dependencies on cuVS (#7261) @divyegala
• CI Hardcode list of words to test porter stemmer with (#7258) @betatim
• Deprecate convert_to_* methods in favor of as_* methods in cuml.ensemble (#7254) @jcrist
• Revert "Extend nightlies breakage window (#7137)" (#7251) @csadorf
• Cleanup cuml.ensemble (#7249) @jcrist
• KernelDensity.score_samples output type & dtype fixes (#7240) @jcrist
• Add test for adjusted_rand_score with small inputs (#7235) @jcrist
• Use except + everywhere in cython (#7234) @jcrist
• Use rapids-xgboost and exclude libnccl from auditwheel check (#7231) @jcrist
• Add cuml.metrics.median_absolute_error (#7229) @jcrist
• Configure repo for automatic release notes generation (#7226) @AyodeAwe
• Accelerate KernelDensity.score_samples (#7219) @jcrist
• Update FIL doc to remove deprecated parameter (#7215) @hcho3
• use new cumlprims_mg repo (#7213) @jameslamb
• update nvidia-ml-py (>=12), use cuda-toolkit wheels (#7212) @jameslamb
• Expose some private attrs in cuml.accel (#7197) @jcrist
• Cleanup KMeans python layer (#7196) @jcrist
• Don't use deprecated method in StratifiedKFold (#7188) @jcrist
• DOC fix table formatting for SHAP (#7186) @sotagg
• Use branch-25.10 again (#7182) @jameslamb
• Small cleanup to cuml.dask.metrics.confusion_matrix (#7181) @jcrist
• Remove get_json/get_detailed_text/get_summary_text from ensemble estimators (#7177) @jcrist
• Move cert configuration to pytest_configure hook (#7175) @jcrist
• Remove baked in datasets from preloading plugin (#7173) @betatim
• Optimize import times a bit (#7171) @jcrist
• Deprecate accuracy_metric (#7170) @jcrist
• use 'nvidia-ml-py' instead of 'pynvml', declare 'numba-cuda' dependency pins (#7164) @jameslamb
• Deprecate predict_model in cuml.ensemble/cuml.dask.ensemble (#7155) @jcrist
• Remove deprecated cupyx.scatter_add/cupyx.scatter_max calls (#7149) @jcrist
• Fix HTML representation's fitted status (#7146) @betatim
• Replace raft calls with cuvs in DBSCAN (#7144) @aamijar
• Expose TSNE.n_iter_ fitted attribute (#7142) @jcrist
• Update rapids-dependency-file-generator (#7141) @KyleFromNVIDIA
• n_streams=1 is not required for determinism in cuml.ensemble (#7140) @jcrist
• Extend nightlies breakage window (#7137) @jcrist
• Add DBSCAN to the common tests (#7134) @betatim
• Remove UCX-Py (#7130) @pentschev
• Build and test with CUDA 13.0.0 (#7128) @jameslamb
• Nicer error message for not implemented fitted attributes in cuml.accel (#7127) @jcrist
• Use gammainc from cupyx (#7126) @jcrist
• Support kneighbors_graph for approximate algos (#7125) @jcrist
• update gputreeshap (#7121) @jameslamb
• Spectral Embedding argument affinity={"precomputed", "nearest_neighbors"} (#7117) @aamijar
• Increase wheel size cap to 525M account for newer architectures (#7116) @divyegala
• Redo memory management for NearestNeighbors (#7115) @jcrist
• FIX Make GaussianNB more resilient (#7113) @betatim
• Remove deprecated cuml.accel CLI options (#7110) @jcrist
• Remove deprecation warnings in 25.10 (#7109) @jcrist
• A few fixes to the profiler IPython magics (#7108) @jcrist
• Use build cluster in devcontainers (#7106) @trxcllnt
• Use rapids_cuda_enable_fatbin_compression (#7105) @robertmaynard
• Forward-merge branch-25.08 into branch-25.10 (#7103) @jcrist
• Remove deprecation warning and data_on_host option for UMAP (#7099) @jinsolp
• Update rapids_config to handle user defined branch name (#7098) @robertmaynard
• Add some devtools to all_* environments (#7091) @jcrist
• Configure --import-mode in pyproject.toml (#7090) @jcrist
• Support SpectralEmbedding in cuml.accel (#7087) @aamijar
• Update rapids-build-backend to 0.4.1 (#7085) @KyleFromNVIDIA
• Update HDBSCAN prediction utils to use raft::linalg::map_offset (#7075) @tarang-jain
• Improved UMAP testing and debugging (#7073) @viclafargue
• Add more estimators to the compatibility test suite (#7069) @betatim
• UMAP with new Spectral Embedding initialization (#7063) @aamijar
• Remove duplicated cuml.accel install code (#7062) @betatim
• cuml RAPIDS dependency now use the new rapids-cmake branch info (#7051) @robertmaynard
• Fix input utils test case (#7050) @emmanuel-ferdman
• Propagate cuml.accel to subprocesses (#7047) @jcrist
• Support enabling cuml.accel with env var (#7046) @jcrist
• Add basic infrastructure for common tests (#7038) @betatim
• Allow latest OS in devcontainers (#7036) @bdice
• Update build infra to support new branching strategy (#7027) @robertmaynard
• Adding appropriate casting for a more generic IdxT usage in HDBSCAN (#7023) @jinsolp
• Build libcuml wheel with static libcuvs (#6959) @hcho3
• Use GCC 14 in conda builds. (#6918) @vyasr
• Add capability to use different allocators in cuML Python benchmarks (#6903) @dantegd
• Reorganize tests and modify pytest invocation to support pytest 8 (#6876) @vyasr

[NIGHTLY] v25.12.00

🔗 Links

• Development Branch
• Compare with main branch

🚨 Breaking Changes

• Don't infer dtype from model in cuml.explainer (#7358) @jcrist
• Cleanup LinearRegression (#7355) @jcrist
• UMAP - delay embedding allocation (#7313) @viclafargue
• Make dask an optional dependency (#7303) @jcrist
• Using int64_t for HDBSCAN and dropping int (#7104) @jinsolp

🐛 Bug Fixes

• Revert "CI: Auto-assign priority to PRs from linked issues (#7354)" (#7357) @csadorf
• Make conversion from scikit-learn SVM to cuML more robust (#7324) @csadorf
• Avoid pyyaml runtime dependency. (#7312) @csadorf
• Adding checks for UMAP precomputed KNN feature (#7300) @viclafargue

📖 Documentation

• Add Spectral Embedding algorithm to README.md (#7348) @aamijar

🚀 New Features

• Provide a static target for libcuml (#7351) @divyegala

🛠️ Improvements

• Don't infer dtype from model in cuml.explainer (#7358) @jcrist
• Cleanup LinearRegression (#7355) @jcrist
• CI: Auto-assign priority to PRs from linked issues (#7354) @csadorf
• Add xgboost to benchmark utilities (#7350) @dantegd
• Remove outdated link flags from CMakeLists.txt (#7349) @hcho3
• Cleanup SVM (#7347) @jcrist
• Enable sccache-dist connection pool (#7344) @trxcllnt
• Add libcuml linkage smoke test (#7343) @divyegala
• Use pinned_host_memory_resource instead of pinned_memory_resource. (#7340) @bdice
• Remove cuml.ensemble/cuml.dask.ensemble deprecated features (#7332) @jcrist
• Cleanup Ridge (#7330) @jcrist
• Accelerate linear model predict on C-ordered inputs (#7329) @jcrist
• Cleanup *.pxd files and cimports (#7327) @jcrist
• Release GIL in SpectralEmbedding (#7326) @jcrist
• Cleanup DBSCAN (#7325) @jcrist
• Reduce warnings verbosity in tests (#7322) @csadorf
• Fix flaky test_logistic_regression_weighting test (#7321) @csadorf
• A few cuml.neighbors cleanups (#7320) @jcrist
• A few HDBSCAN cleanups (#7319) @jcrist
• Cleanup cuml.decomposition (#7316) @jcrist
• Run the scikit-learn test suite in parallel again (#7315) @csadorf
• UMAP - delay embedding allocation (#7313) @viclafargue
• Cleanup TSNE python implementation (#7311) @jcrist
• Propagate Dask/UCX exceptions (#7308) @viclafargue
• Add SparseRandomProjection, AgglomerativeClustering and GaussianRandomProjection to common checks (#7307) @betatim
• Make dask an optional dependency (#7303) @jcrist
• Update to rapids-logger 0.2 (#7301) @bdice
• Branch 25.12 merge branch 25.10 (#7291) @jcrist
• Avoid usage of deprecated cudf from_pandas (#7284) @TomAugspurger
• Update RAPIDS_BRANCH, codify changes in update-version.sh (#7268) @KyleFromNVIDIA
• Using int64_t for HDBSCAN and dropping int (#7104) @jinsolp

v25.08.00

cuML 25.08 Release Notes

🎉 What's New

⭐ Highlights

• Spectral Embedding: New algorithm for dimensionality reduction and manifold learning (#6581) @aamijar
• cuML.accel Profiler: Added profiling capabilities for Zero Code Change Acceleration (#7021) @jcrist
• cuML.accel LinearSVC/LinearSVR: New support for linear support vector classification and regression (#6866) @viclafargue
• cuML.accel set_output/get_feature_names_out: Added support for scikit-learn output configuration (#6942) @jcrist

🔧 Major Improvements

UMAP Enhancements

• Multi-GPU KNN graph building support (#7019) @jinsolp
• Improved handling of identical vectors in distance calculations (#6904) @jinsolp
• Disabled non-determinism on small datasets for better reproducibility (#7004) @viclafargue

FIL (Forest Inference Library) Improvements

• Support for wide data inference (#7014) @hcho3
• Better handling of empty categorical nodes (#6924) @hcho3
• Improved GPU context handling (#6987) @hcho3
• Restored legacy threshold behavior (#6922) @hcho3

Zero Code Change Acceleration (cuml.accel)

• Added profilers for better performance analysis (#7021) @jcrist
• Enhanced logging for proxy estimators (#6957) @csadorf
• Implemented metadata routing support (#6950) @jcrist
• New support for LinearSVC and LinearSVR (#6866) @viclafargue
• Added KernelRidge to supported algorithms (#6917) @jcrist
• Improved CLI with -c and - options support (#6852) @jcrist

Algorithm Enhancements

• DBSCAN: Now computes components_ attribute (#6976) @jcrist
• LogisticRegression: Exposed n_iter_ attribute for iteration tracking (#6911) @betatim
• RandomForest: Fixed default max_features parameter (#6862) @jcrist
• TSNE: Added fallback support for unsupported metrics (#6992) @jcrist
• Ridge: Better handling of underdetermined systems (#7003) @betatim

Developer Experience

• Testing: Enhanced CI with upstream test suites for HDBSCAN, UMAP, and other algorithms (#6995, #6989, #6986) @jcrist
• Documentation: Comprehensive updates to Python developer guide and API documentation (#6843) @csadorf
• Dependencies: Updated to CUDA 12.9 and added support for scikit-learn 1.4 (#6944, #6845) @jakirkham, @betatim

🚨 Breaking Changes

Deprecated Parameters & Functions

• UMAP: data_on_host parameter is deprecated (#6953) @jinsolp
• HDBSCAN:
  • Prediction functions in cuml.cluster namespace are deprecated (#6943) @jcrist
  • connectivity parameter is deprecated (#6936) @jcrist
• SGD Algorithms: penalty='none' is deprecated in MBSGDClassifier, MBSGDRegressor, and SGD (#6926) @jcrist
• KMeans: random_state default changed to None (#6884) @jcrist

Removed Components

• Experimental FIL: experimental.fil Python module removed (#6899) @hcho3
• Legacy FIL: Removed from libcuml (#6844) @hcho3
• CUDA 11 Support: Removed from dependencies and CI (#6847, #6885) @KyleFromNVIDIA, @dantegd
• Package Distribution: Stopped uploading to downloads.rapids.ai (#6803) @jameslamb

API Changes

• Neighbors: Ported to InteropMixin/ProxyBase (#6851) @jcrist
• Ensemble: Ported to InteropMixin/ProxyBase (#6863) @jcrist
• UMAP: Ported to InteropMixin/ProxyBase (#6840) @jcrist

🐛 Bug Fixes

Algorithm Fixes

• UMAP: Improved handling of identical vectors in UMAP distance calculations (#6904) @jinsolp
• TSNE: Relaxed tolerance for sparse input tests (#7033) @jinsolp
• RandomForest: Fixed default max_features parameter (#6862) @jcrist
• HDBSCAN: Rewrote Python wrapper for better stability (#6913) @jcrist
• Logistic Regression: Increased tolerance in Dask tests (#6848) @csadorf

Compatibility & Dependencies

• Fixed compatibility with scikit-learn 1.7.0 and Python 3.13.4 (#6865) @csadorf
• Unxfailed tests affected by numba compilation errors (#6905) @csadorf

Other

• Fixed import error in benchmark code (#6902) @dantegd

📖 Documentation Updates

User Documentation

• Supported Versions: Added comprehensive version compatibility documentation (#7040) @csadorf
• Zero Code Change Acceleration: Updated title and reorganized documentation (#7030, #7026) @csadorf, @jcrist
• UMAP: Added multi-GPU KNN graph building documentation (#7019) @jinsolp
• TSNE: Fixed FFT TSNE documentation (#6967) @jinsolp
• Limitations: Revamped cuml.accel limitations documentation (#6965) @jcrist

Developer Documentation

• Python Developer Guide: Comprehensive updates (#6843) @csadorf
• CI Workflow: Added documentation for workflow inputs (#6952) @jameslamb
• Async Operations: Removed outdated async operation section (#6980) @csadorf

🔄 Migration Guide

For Users

1. UMAP: Remove data_on_host parameter from your code
2. HDBSCAN: Update to use new prediction function signatures
3. SGD: Replace penalty='none' with appropriate alternatives
4. KMeans: Be aware that random_state=None is now the default

For Developers

1. CUDA 11: Update your development environment to CUDA 12.9
2. FIL: Update imports from experimental.fil to fil
3. Dependencies: Update to supported versions as documented

📊 Summary

This release brings significant improvements to cuML's performance, stability, and developer experience. The highlight is the new Spectral Embedding algorithm, along with major architectural improvements to HDBSCAN and UMAP. The Zero Code Change Acceleration feature continues to expand with new algorithms and better tooling.

The breaking changes are primarily focused on cleaning up deprecated APIs and improving the overall codebase structure. Users are encouraged to review the migration guide and update their code accordingly.

🔧 Internal & Technical Changes

Architecture Improvements

• HDBSCAN Migration: Migrated to cuVS cluster primitives from raft::cluster for better performance (#6560) @tarang-jain
• Module Porting: Ported cuml.neighbors, cuml.ensemble, and UMAP to InteropMixin/ProxyBase (#6851, #6863, #6840) @jcrist
• Base Class Cleanup: Removed deprecated base classes and functions (#6919, #6888) @jcrist
• Cython Optimization: De-Cythonized several modules for better maintainability (#6920) @jcrist

Performance Optimizations

• Binary Size: Reduced ARIMA kernels binary size (#6997) @jcrist
• Memory Usage: Instantiated only specific RAFT kernels (#6900, #6780) @aamijar, @divyegala
• Kernel Optimization: Used cuvs::neighbors::knn_merge_parts (#7005) @jcrist

Infrastructure & CI/CD

• Added GH_TOKEN pass-through for job summarization (#6894) @msarahan
• Fixed various CI test failures and flaky tests (#6906, #6877) @csadorf
• CUDA 12.9: Updated across all environments (#6944) @jakirkham
• Dependencies: Removed NVIDIA and Dask channels (#6935) @vyasr
• CI Images: Added versioned CI image tags (#7016) @jameslamb
• Testing: Enhanced with upstream test suites and better organization (#6995, #6989, #6986) @jcrist
• • Fixed network failure when loading MNIST dataset during spectral embedding testing (#7077) @aamijar

Code Quality

• Shell Scripts: Fixed all shellcheck warnings and errors (#6901) @gforsyth
• Linting: Updated cython-lint and fixed long lines (#6969) @jcrist
• Documentation: Comprehensive updates to developer guides (#6843) @csadorf

---

For detailed information about specific changes, please refer to the individual pull requests linked in each entry.

v25.06.00

🚨 Breaking Changes

• Deprecate device selection (#6784) @jcrist
• Remove use of legacy FIL from Python cuML (#6728) @hcho3
• Use RBC from cuVS (#6644) @divyegala
• Map Barnes-Hut to FFT for T-SNE in cuml.accel. (#6619) @csadorf
• New Estimator Proxy architecture (#6613) @jcrist
• Disable building and uploading cuml-cpu in CI (#6529) @dantegd
• FIX Propagate random state to numpy rng in make_classification (#6518) @betatim

🐛 Bug Fixes

• UMAP batch nnd test hotfix (#6826) @jinsolp
• Update forest_inference_demo.ipynb for new FIL API changes (#6824) @dantegd
• Fix to compile FIL infer_kernel with CUDA arch 1210 (#6821) @viclafargue
• Reduce solver test threshold for ElasticNet test. (#6766) @csadorf
• Use _assert_allclose function for approximate equality check. (#6763) @csadorf
• UMAP spectral initialization fallback to random on error (#6750) @aamijar
• [Fix] include_self=False for duplicate points (tied zero distances) in kneighbors_graph (#6735) @aamijar
• FIX: Set scikit-learn KMeans n_init=10 to match cuML (#6727) @csadorf
• Fix output dtype of LinearSVC.predict (#6715) @jcrist
• CUDA 12.9 Correctly bound updated compression flags (#6713) @robertmaynard
• Fix type and support for PCA.noise_variance_ (#6693) @jcrist
• Update dask RF regressor partial_inference (#6691) @TomAugspurger
• Properly handle degenerate trees in FIL (#6673) @hcho3
• Fix strict marker in scikit-learn test xfail list. (#6661) @csadorf
• DOC Repair docstring formatting (#6659) @betatim
• Fix the field ids in the pr/issue status automation (#6656) @csadorf
• Adjust scikit-learn test failure handling (#6646) @csadorf
• Remove arbitrary score threshold for the test_qn test with softmax (#6636) @csadorf
• Do not set n_features_in_ attribute before fit (#6624) @betatim
• Map Barnes-Hut to FFT for T-SNE in cuml.accel. (#6619) @csadorf
• Mark sklearn test test_cross_val_predict[coo_array] as flaky. (#6610) @csadorf
• Do not run the scikit-learn test suite in parallel. (#6609) @csadorf
• Mark one kmeans and one t_sne sklearn test as flaky (#6598) @csadorf
• Relax threshold for failing ARIMA pytest (#6579) @divyegala
• Fix race condition in _block_gemv kernel (#6578) @divyegala
• Fix launch parameters for kernel shap (#6577) @divyegala
• Update SVC tests for CCCL update (#6569) @viclafargue
• Update FIL Model Loading Parameter in FIL demo notebook (#6562) @csadorf
• Localise output‑dtype; remove global set_api_output_dtype calls (#6561) @Ofek-Haim
• Lower logistic regression digits test threshold to 0.9. (#6552) @csadorf
• FIX Propagate random state to numpy rng in make_classification (#6518) @betatim
• Fix cudf.DataFrame construction with dict like inputs and ensure CumlArray.to_output('cudf') does not convert NaN to NA (#6517) @mroeschke
• Calling gc in SVC tests (#6514) @viclafargue
• Fix logger macros (#6511) @vyasr
• Simplicial set functions fixes and cleanups (#6493) @viclafargue
• Inspect KNN graph during smooth KNN generation and fail if necessary (#6491) @viclafargue
• More permanently disable BATCH_TREE_REORG (#6374) @wphicks

📖 Documentation

• Move FIL docs to user docs (#6762) @dantegd
• Document cuml.accel doesn't support set_output (#6717) @jcrist
• DOC Update zero-code-change documentation (#6697) @csadorf
• Document WSL2 limitation re UVM (#6663) @csadorf
• DOC Make MultinomialNB dask doctest more lax (#6482) @betatim
• Future-proof SVC documentation (#6481) @viclafargue
• DOC Add developer documentation about deprecation policy (#6476) @betatim

🚀 New Features

• Multi-gpu KNN build for UMAP using all-neighbors API (#6654) @jinsolp
• Add caching and object size metrics to conda CI builds (#6573) @divyegala
• Use Gram kernel from cuVS instead of Raft (#6538) @lowener

🛠️ Improvements

• Migrate cuml.experimental.fil -> cuml.fil (#6799) @hcho3
• Mark test_uniform_grid as flaky in sklearn tests (#6791) @jcrist
• Deprecate a few missed cuml-only non-keyword-only kwargs (#6789) @jcrist
• Delay importing of dask (#6788) @jcrist
• Port DBSCAN to ProxyBase/InteropMixin (#6786) @jcrist
• Deprecate device selection (#6784) @jcrist
• use 'rapids-init-pip' in wheel CI, other CI changes (#6779) @jameslamb
• Support SVC and SVR in cuml.accel (#6778) @jcrist
• chore: reduce ceiling of cuml wheel size (#6771) @gforsyth
• Port KMeans to use ProxyBase/InteropMixin (#6770) @jcrist
• Port sklearn.manifold wrappers to use ProxyBase (#6761) @jcrist
• Fix dask warning in FIL demo notebook (#6758) @jcrist
• Remove deprecations for 25.06 (#6754) @jcrist
• Check that as_sklearn'ed models can be fitted (#6751) @betatim
• Implement sklearn_tags on mixin using vendored compat module. (#6746) @csadorf
• Test that proxied method signatures match their CPU versions (#6740) @jcrist
• Remove upper bound constraint on xgboost. (#6739) @csadorf
• Deprecate non-keyword-only cuml-only kwargs (#6738) @jcrist
• Fix DistributedDataHandler.calculate_parts_to_sizes (#6737) @TomAugspurger
• Finish CUDA 12.9 migration and use branch-25.06 workflows (#6733) @bdice
• Remove use of legacy FIL from Python cuML (#6728) @hcho3
• Update Dask RandomForest classifier (#6726) @TomAugspurger
• Update to clang 20 (#6724) @bdice
• Use matplotlib-base in docs/notebooks environments. (#6723) @bdice
• FIX Allow prereleases when evaluating xfail conditions (#6722) @betatim
• Quote head_rev in conda recipes (#6701) @bdice
• Don't setup managed memory if rmm isn't using default settings (#6700) @jcrist
• Expand sklearn matrix used in tests (#6698) @jcrist
• Port sklearn.decomposition to new proxy estimator structure (#6695) @jcrist
• fix(rattler): add librmm to host for libcuml to fix overlinking error (#6686) @gforsyth
• CUDA 12.9 use updated compression flags (#6684) @robertmaynard
• Increase scikit-learn compatibility to 1.5+ (#6666) @csadorf
• Revert "CI: Increase nightly CI check window to 14 days. (#6628)" (#6660) @csadorf
• Remove cuml-cpu and CPU <> GPU docs (#6651) @jcrist
• More dask==2025.4.1 compatibility (#6645) @TomAugspurger
• Use RBC from cuVS (#6644) @divyegala
• Exclude librmm.so from auditwheel (#6632) @bdice
• Enhance accelerator parity testing framework (#6629) @csadorf
• CI: Increase nightly CI check window to 14 days. (#6628) @csadorf
• Add support for Python 3.13 (#6616) @gforsyth
• Compatibility with Dask 2025.4.1 (#6614) @TomAugspurger
• New Estimator Proxy architecture (#6613) @jcrist
• Using cuVS NN Descent in UMAP (#6611) @jinsolp
• Make scikit-learn a required dependency (#6608) @betatim
• Speedup dask LogisticRegression tests (#6607) @jcrist
• Add GitHub Actions workflow for PR and issue project field automation (#6602) @csadorf
• Use nogil for libcuml calls (#6601) @jcrist
• Remove cuml.common.cuda (#6600) @jcrist
• Remove most of import_utils.py (#6599) @jcrist
• Remove guarded imports of scipy (#6596) @jcrist
• Remove remainder of code branching on cuml-cpu support (#6593) @jcrist
• Add timeout to scikit-learn test suite (#6591) @jcrist
• Remove safe_imports.py (#6588) @jcrist
• Remove usage of cudf.core.buffer (#6586) @mroeschke
• Split wheel tests into two jobs (#6585) @jcrist
• Only run one cudf.pandas enabled build per PR (#6584) @jcrist
• Don't run memleak tests in CI (#6583) @jcrist
• Documentation and Testing Infrastructure Updates (#6580) @csadorf
• Move Random Forest n_streams UserWarning out of execution (#6574) @ncclementi
• Remove conditional compilation blocks leftover from cuml-cpu (#6572) @jcrist
• Download build artifacts from Github for CI (#6566) @VenkateshJaya
• Don't exclude sklearn tests from acceleration (#6565) @jcrist
• Add .git-blame-ignore-revs (#6564) @csadorf
• Add cuml.accel.is_proxy (#6559) @jcrist
• Update KMeans test to improve stability by setting n_init to 2 (#6553) (#6555) @csadorf
• Support weights="distance" for KNeighbors* in cuml.accel (#6554) @jcrist
• Improve detection of scikit-learn parity regressions (#6553) @csadorf
• Revise cuml.accel module patching and import mechanism (#6550) @jcrist
• Add isort to pre-commit hooks (#6549) @csadorf
• Refactor dataset generation and testing utilities (#6548) @csadorf
• Move cuml.accel tests out into own directory (#6547) @jcrist
• refactor(rattler): enable strict channel priority for builds (#6546) @gforsyth
• Remove unnecessary CudfBuffer and vendor cuDF min_signed_type function (#6543) @mroeschke
• Remove _experimental_dispatching attribute (#6542) @jcrist
• Avoid private attributes on cuDF objects (#6537) @mroeschke
• Reduce CI test combinations for scikit-learn acceleration tests (#6536) @csadorf
• Vendor RAPIDS.cmake (#6531) @bdice
• Disable building and uploading cuml-cpu in CI (#6529) @dantegd
• Remove CUDA whole compilation ODR violations (#6525) @divyegala
• SLHC to Migrate to cuVS Agglomerative from RAFT (#6507) @tarang-jain
• Solve forward merger conflicts of 6435 (#6501) @csadorf
• Add minimal import check to CI (#6484) @jcrist
• Moving wheel builds to specified location and uploading build artifacts to Github (#6472) @VenkateshJaya

v25.04.00

🚨 Breaking Changes

• Promote experimental FIL to stable (#6464) @wphicks
• Support non-trivial classes_ in LogisticRegression (#6346) @jcrist
• Use new rapids-logger library (#6289) @vyasr

🐛 Bug Fixes

• Unblock CI for 25.04 (#6519) @csadorf
• Skip test_rf_classification_seed for cudf.pandas tests. (#6500) @csadorf
• Remove dask to sparse workarounds (#6489) @TomAugspurger
• Fix accidental required sklearn import (#6483) @jcrist
• FIX Pass instances to the constructor of the meta-estimator in SVC (#6471) @betatim
• Fix compiler dependency on ARM (#6456) @bdice
• Revert "Temporarily increase max_days_without_success (#6390)" (#6455) @divyegala
• Enhance argument handling in UniversalBase to accept NoneType (#6453) @csadorf
• Fix UMAP transform (#6449) @viclafargue
• Fallback to CPU for sparse inputs for KMeans (#6448) @csadorf
• Limited support for array-like inputs (#6442) @csadorf
• Fix test_accuracy_score test on cudf.pandas build (#6439) @jcrist
• Support positional args in cuml.accel estimators (#6423) @jcrist
• Fix HDBSCAN python docs related to metric (#6422) @divyegala
• declare runtime dependency on 'packaging', update scikit-learn and hdbscan for cuml-cpu to match cuml (#6420) @jameslamb
• Fix forwarding of initial_alpha and learning_rate in UMAP (#6417) @jcrist
• Implement Ridge .solver_ estimated attribute (#6415) @csadorf
• Fix multi-target prediction in linear models (#6414) @csadorf
• Ensure output_type="pandas" returns user-facing pandas (#6407) @jcrist
• Ensure LinearSVC supports all input types (#6404) @jcrist
• Make UMAP callback pickleable (#6402) @jcrist
• Correctly align trees in experimental FIL (#6397) @wphicks
• A few log level handling cleanups (#6393) @jcrist
• Fix cupy error in KernelDensity with epanechnikov (#6388) @jcrist
• Check if both ground truth and result are NaN in MSEObjectiveTest for RF (#6387) @wphicks
• Support non-native endian inputs to LabelEncoder (#6384) @jcrist
• Fix hypothesis test in test_kernel_ridge.py (#6382) @jcrist
• Remove debug logging from nearest neighbors test (#6376) @csadorf
• Ensure FIL CPU can be run without an available GPU (#6373) @wphicks
• Correctly pass through unhashable objects during hyperparam lookup (#6369) @wphicks
• Correct translation of RandomForest criterion hyperparameter (#6363) @wphicks
• UMAP fallback improvements (#6358) @viclafargue
• A few GPU<->CPU interop fixes (#6355) @jcrist
• Fix output type of KernelRidge.predict (#6354) @jcrist
• Skip test_extract_partitions_shape test. (#6338) @csadorf
• Require sphinx<8.2.0 (#6336) @csadorf
• Fix call to NearestNeighbors.kneighbors() without args. (#6333) @csadorf
• Commenting out debug print statements in UMAP (#6332) @viclafargue
• Correct UMAP dispatch trigger condition (#6330) @viclafargue
• Fix issue caused by None arg provided by sklearn pipeline (#6326) @viclafargue
• Improvements for interoperability of Random Forest (#6320) @dantegd
• Improvements for PCA CPU interoperability (#6319) @dantegd
• Plumb metric and metric_kwds through to UMAP with nn_descent (#6304) @jcrist
• Use L2Expanded/L2SqrtExpanded instead of the *Unexpanded variants (#6303) @jcrist
• Fix UMAP issues with large inputs (#6245) @viclafargue
• Fix LogisticRegression.decision_function output shape (#6235) @jcrist
• Compute noise_variance_ in PCA implementation (#6234) @jcrist
• Multiple CPU interop fixes for serialization and cloning (#6223) @dantegd

📖 Documentation

• Clarify support in docs (#6452) @bdice
• Provide docs for cuml.accel command line feature (#6322) @wphicks

🚀 New Features

• Promote experimental FIL to stable (#6464) @wphicks
• UMAP 32bits dispatch mechanism (#6314) @viclafargue
• Add aarch64 conda environments (#6226) @jcrist
• SVM cpu/gpu interop (#6089) @viclafargue

🛠️ Improvements

• Don't use private APIs in FIL notebook (#6488) @jcrist
• Only use released versions of dask (#6485) @jcrist
• Move SVM working set log to debug (#6477) @jcrist
• C++ SG_RF_TEST OOMs in 11.4 nightly (#6474) @divyegala
• Revert "ci: pre-filter 11.4 jobs before they are enabled in shared workflows (#6447)" (#6470) @divyegala
• Support __len__ in ensemble estimators (#6468) @jcrist
• Add deprecation warning to cuml-cpu builds (#6466) @jcrist
• Fix template argument of DeviceSegmentedReduce kernels (#6459) @miscco
• Add scikit-learn acceleration tests to nightly CI tests (#6457) @csadorf
• Fix benchmark builds in devcontainers (#6450) @bdice
• ci: pre-filter 11.4 jobs before they are enabled in shared workflows (#6447) @gforsyth
• Reenable more UMAP tests on ARM (#6446) @jcrist
• Clean up after removing KMeans deprecation warning (#6445) @betatim
• Port all conda recipes to rattler-build (#6440) @gforsyth
• Remove the deprecation warning and switch to new default (#6433) @betatim
• Remove old deprecations (#6427) @jcrist
• Fix oob_score=True handling with accelerator (#6425) @betatim
• remove unused imports (#6424) @jameslamb
• Disable codecov comments (#6419) @jcrist
• Require hypothesis tests have explicit examples (#6409) @aamijar
• Improve accuracy_score compatibility with sklearn. (#6406) @jcrist
• Drop cub{Min, Max} (#6405) @miscco
• Move accel implementation fully to cuml.accel (#6401) @jcrist
• cuml-cpu: fix import issues, enable conda import tests (#6400) @jameslamb
• Drop numba_utils (#6396) @jakirkham
• Simplify cpu_to_gpu and gpu_to_cpu (#6395) @jcrist
• Adds get_level() (#6392) @Intron7
• Temporarily increase max_days_without_success (#6390) @jcrist
• Drop deprecated CCCL functors (#6389) @miscco
• Solve forward merger conflicts of #6313 (#6385) @dantegd
• CI: Run extended accel mode tests (#6377) @csadorf
• Use conda-build instead of conda-mambabuild (#6368) @bdice
• Don't log coverage output to terminal in CI (#6367) @jcrist
• Remove KernelRidge from cuml.accel (#6361) @jcrist
• Enable UVM in cuml.accel (#6359) @wphicks
• Drop deprecated CUB iterators in favor of Thrust iterators (#6356) @miscco
• Adjust accel verbosity (#6353) @csadorf
• Avoid importing umap with import cuml (#6352) @jcrist
• Make proxy module patching skip patching if patched library is not installed (#6348) @dantegd
• Add new cuml.accel module (#6347) @jcrist
• Support non-trivial classes_ in LogisticRegression (#6346) @jcrist
• Make r2_score compatible with Scikit-Learn (#6337) @jcrist
• Remove left-over code related to strict mode. (#6335) @csadorf
• Make sure specifying default value or nothing behaves the same in KMeans (#6331) @betatim
• Make KernelRidge inherit from UniversalBase (#6327) @betatim
• Reduce peak memory in UMAP.fit/UMAP.fit_transform (#6323) @jcrist
• Consolidate more Conda solves in CI (#6321) @KyleFromNVIDIA
• Require CMake 3.30.4 (#6318) @robertmaynard
• Copy attributes from the original class to the proxy (#6306) @betatim
• Use shared-workflows branch-25.04 (#6298) @bdice
• Add build_type input field for test.yaml (#6296) @gforsyth
• Do not use LEGACY_PTX_ARCH argument for BlockReduce (#6290) @miscco
• Use new rapids-logger library (#6289) @vyasr
• disallow fallback to Make in wheel builds (#6286) @jameslamb
• Forward-merge branch-25.02 to branch-25.04 (#6283) @bdice
• Migrate to NVKS for amd64 CI runners (#6280) @bdice
• Add verify-codeowners hook (#6275) @KyleFromNVIDIA
• Speed up FIL and TSNE pytests in CI (#6270) @dantegd
• Bracnh 25.04 merge branch 25.02 (#6268) @vyasr
• Add shellcheck to pre-commit and fix warnings (#6246) @gforsyth
• Prepare for n_init=auto in KMeans (#6142) @betatim