Extension for Scikit-learn* 2025.9.0

Extension for Scikit-learn* is happy to introduce 2025.9.0 release!

🚨 What's New

• Introduced new Extension for Scikit-learn* functionality:
  • Enabled SPMD API support in kNN search
  • Enabled array API support in the PCA, EmpiricialCovariance and their incremental variants

🪲 Bug Fixes

• Fixed error on decision function with GPU arrays
• Fixed re-usage of non-reusable daal4py objects
• Prevented Logistic Regression from turning to inf in predictions

Acknowledgements

Thanks to everyone who helped us make 2025.9.0 release possible!

@Alexsandruss, @Alexandr-Solovev, @Vika-F, @david-cortes-intel, @icfaust, @napetrov, @maria-Petrova, @homksei, @ahuber21, @ethanglaser, @razdoburdin, @avolkov-intel, @KateBlueSky, @yuejiaointel, @DDJHB, @kjackiew, @richardnorth3

Full Changelog: 2025.8.0...2025.9.0

Extension for Scikit-learn* 2025.8.0

Extension for Scikit-learn* is happy to introduce 2025.8.0 release!

🚨 What's New

• Introduced new Extension for Scikit-learn* functionality:
  • Enabled array API support in DBSCAN
  • Enabled array API support in BasicStatistics, LinearRegression, Ridge algorithms and their incremental variants
  • Added parameters for covariance and PCA controlling batched vs. non-batched route
  • Enabled common verbosity arguments for pytest

🪲 Bug Fixes

• Fixed incorrect type castings and mismatched operators
• Fixed double division of logistic model regularization by sum of weights
• Fixed incorrect processing of logistic regularization when passed to sklearn
• GPU version of Logistic Regression now returns a correct shape of probabilities

Acknowledgements

Thanks to everyone who helped us make 2025.8.0 release possible!

@Alexsandruss, @Alexandr-Solovev, @Vika-F, @david-cortes-intel, @icfaust, @napetrov, @maria-Petrova, @homksei, @ahuber21, @ethanglaser, @razdoburdin, @avolkov-intel, @KateBlueSky, @yuejiaointel, @DDJHB, @kjackiew, @richardnorth3

New Contributors

• @KateBlueSky made their first contribution in #2640

Full Changelog: 2025.7.0...2025.8.0

Extension for Scikit-learn* 2025.7.0

Extension for Scikit-learn* is happy to introduce 2025.7.0 release!

🚨 What's New

• Introduced new Extension for Scikit-learn* functionality:
  • Added support sklearn 1.6 conformance testing
  • Added grain_size hyperparameter into EmpiricalCovariance and PCA algorithms
  • Enabled model conversion from TreeLite

🪲 Bug Fixes

• Minor basic stats quality fixes
• Fixed policy changes for spmd
• Fixed Forest dpctl predict queue misalignment
• Fixed logic to check for existence of different oneDAL libraries

Acknowledgements

Thanks to everyone who helped us make 2025.7.0 release possible!

@Alexsandruss, @Alexandr-Solovev, @Vika-F, @david-cortes-intel, @icfaust, @napetrov, @maria-Petrova, @homksei, @ahuber21, @ethanglaser, @razdoburdin, @avolkov-intel, @yuejiaointel

Full Changelog: 2025.6.1...2025.7.0

Extension for Scikit-learn* 2025.6.1

Extension for Scikit-learn* is happy to introduce 2025.6.1 release!

🚨 What's New

• Introduced new Extension for Scikit-learn* functionality:
  • Model builders can now work with XGBoost regression models that involve link functions
  • Improved XGBoost compatibility for object modeling
  • Added a new class with .predict() for logistic regression model builder

🪲 Bug Fixes

• Bug fixes for decision trees
• Fixed forcing D4P compiler to ICX when building with DPC
• Fixes for sklearn 1.7 pre-release support

Acknowledgements

Thanks to everyone who helped us make 2025.6.1 release possible!

@Alexsandruss, @Alexandr-Solovev, @Vika-F, @david-cortes-intel, @icfaust, @napetrov, @maria-Petrova, @homksei, @ahuber21, @ethanglaser, @razdoburdin, @avolkov-intel, @yuejiaointel

Full Changelog: 2025.5.0...2025.6.1

Extension for Scikit-learn* 2025.5.0

Extension for Scikit-learn* is happy to introduce 2025.5.0 release!

🚨 What's New

• Introduced new Extension for Scikit-learn* functionality:
  • Add new parameters for linear regression
  • Accelerated array_api inputs for sklearnex's validate_data and _check_sample_weight
  • Model builders can now work with XGBoost regression models that involve link functions
  • XGBoost model objects don't get invalidated after converting them to daal4py
  • There's now a class with .predict() for logistic regression model builder

🪲 Bug Fixes

• Normalized Decision Tree .values attribute to match sklearn
• Fixed incorrect scale of base_score being used for XGB regression objectives
• Fixed csr k-Means Init offloading when SYCL CPU device is unavailable

Acknowledgements

Thanks to everyone who helped us make 2025.5.0 release possible!

@Alexsandruss, @Alexandr-Solovev, @Vika-F, @david-cortes-intel, @icfaust, @napetrov, @maria-Petrova, @homksei, @ahuber21, @ethanglaser, @razdoburdin, @avolkov-intel, @yuejiaointel

New Contributors

• @tanannie22 made their first contribution in #2343

Full Changelog: 2025.4.0...2025.5.0

Extension for Scikit-learn* 2025.4.0

Extension for Scikit-learn* is happy to introduce 2025.4.0 release!

🚨 What's New

• Introduced new Extension for Scikit-learn* functionality:
  • Improved the help documentation by adding clickable links
  • Added support for Python 3.13 support
  • Added support for Sklearn 1.6

Acknowledgements

Thanks to everyone who helped us make 2025.4.0 release possible!

@Alexsandruss, @Alexandr-Solovev, @Vika-F, @david-cortes-intel, @icfaust, @napetrov, @maria-Petrova, @homksei, @ahuber21, @ethanglaser, @razdoburdin, @avolkov-intel, @yuejiaointel,

Full Changelog: 2025.2.0...2025.4.0

Extension for Scikit-learn* 2025.2.0

Extension for Scikit-learn* is happy to introduce 2025.2.0 release!

🚨 What's New

• Introduced new Intel® Extension for Scikit-learn* functionality:
  • Enabled linear regression on GPU with non-PSD systems
  • Added serialization for IncrementalBasicStatistics, IncrementalEmpiricalCovariance, IncrementalPCA
  • Moved Ridge Regression out of preview
  • Disabled patching for k-Means(n_clusters=1)
  • Added sklearnex version of validate_data, _check_sample_weight
  • Upgraded IncrementalLinearRegression for underdetermined systems
  • Enabled new RNG engines support

🪲 Bug Fixes

• Enabled proper GPU offloading with fp64 support when dpctl unavailable
• Fixed to_table for a non-array input when a low-precision queue is used
• Fixed incremental pca example patching logic

Acknowledgements

Thanks to everyone who helped us make 2025.2.0 release possible!

@Alexsandruss, @Alexandr-Solovev, @Vika-F, @david-cortes-intel, @icfaust, @napetrov, @maria-Petrova, @homksei, @ahuber21, @ethanglaser, @razdoburdin, @avolkov-intel, @yuejiaointel,

New Contributors

• @yuejiaointel made their first contribution in #2229

Full Changelog: 2025.1.0...2025.2.0

Intel® Extension for Scikit-learn* 2025.1.0

Intel® Extension for Scikit-learn* is happy to introduce 2025.1.0 release!

🚨 What's New

• Introduced new Intel® Extension for Scikit-learn* functionality:
  • Enabled accelerated Linear Regression for overdetermined systems
  • Enabled hyperparameter support for Random Forest classifier inference
  • Enabled serialization in daal4py algorithm classes

🪲 Bug Fixes

• Fixed int overflow in FTI model convertor
• Updated BasicStatistics and IncrementalBasicStatistics to follow additional sklearn conventions
• Fixed n_jobs support coverage to indirectly-supported oneDAL methods
• Fixed KMeans score check in _onedal_*_supported and n_jobs support for score
• Corrected skips in design rule checks (test_common.py) caused by fragile whitelist_to_blacklist
• Fixed test_estimators[LogisticRegression()-check_estimators_unfitted] conformance for gpu support
• Updated functional support fallback logic for a DPNP/DPCTL ndarray inputs
• Fixed an issue in aliased _onedal_cpu_supported and _onedal_gpu_supported in fit_check_before_support_check
• Fixed logic of k-NN algos kneighbors() call when algorithm='brute' and fit with GPU

🔨 Library Engineering

• Added Python 3.13 support for Intel® Extension for Scikit-learn* packages
• Added Sklearn 1.6 support for Intel® Extension for Scikit-learn* packages

Acknowledgements

Thanks to everyone who helped us make 2025.1.0 release possible!

@Alexsandruss, @Alexandr-Solovev, @Vika-F, @david-cortes-intel, @icfaust, @napetrov, @maria-Petrova, @homksei, @ahuber21, @ethanglaser, @samir-nasibli, @olegkkruglov, @razdoburdin, @avolkov-intel, @md-shafiul-alam

Full Changelog: 2025.0.0...2025.1.0

Intel® Extension for Scikit-learn* 2025.0.0

Intel® Extension for Scikit-learn* is happy to introduce 2025.0.0 release!

🚨 What's New

• Introduced new Intel® Extension for Scikit-learn* functionality:
  • Enabled functional support for Array API
  • k-Means algorithm is moved out of preview namespace
  • SHAP value support for XGBoost's binary classification models
  • SPMD interfaces support: IncrementalLinearRegression, IncrementalPCA, IncrementalEmpiricalCovariance

🪲 Bug Fixes

• Fix issues with sklearn conformance for preview Ridge for 2024.6.0
• Fix on preview ridge tests having too little error tolerance for coefficients assertions
• Fix for Logistic Regression loss scaling
• Fix to prevent support_usm_ndarray from changing queue if explicitly provided
• Fix Multivariate Ridge Regression coefficients
• Fix circular import in daal4py/sklearnex device_offloading
• Align sklearnex BasicStatistics._onedal_fit with other algos

❌ Deprecation Notice

• Removed Python 3.8 support

Acknowledgements

Thanks to everyone who helped us make 2025.0.0 release possible!

@Alexsandruss, @Alexandr-Solovev, @Vika-F, @icfaust, @napetrov, @maria-Petrova, @ahuber21, @ethanglaser, @samir-nasibli, @aepanchi, @emmwalsh, @olegkkruglov, @razdoburdin, @avolkov-intel, @md-shafiul-alam, @david-cortes-intel

Full Changelog: 2024.7.0...2025.0.0

Intel® Extension for Scikit-learn* 2024.7.0

Intel® Extension for Scikit-learn* is happy to introduce 2024.7.0 release!

🚨 What's New

• Introduced new Intel® Extension for Scikit-learn* functionality:
  • Sparse data support for LogisticRegression
  • Basic Statistic API improvement
  • Added random_state warning to SVM probability estimates
  • Unified daal4py and sklearnex builds

🪲 Bug Fixes

• Fix issues with sklearn conformance for preview Ridge for 2024.6.0
• Fix on preview ridge tests having too little error tolerance for coefficients assertions
• Fix for Logistic Regression loss scaling
• Fix to prevent support_usm_ndarray from changing queue if explicitly provided
• Fix Multivariate Ridge Regression coefficients
• Fix circular import in daal4py/sklearnex device_offloading
• Align sklearnex BasicStatistics._onedal_fit with other algos

❌ Deprecation Notice

• Removed Python 3.8 support

Acknowledgements

Thanks to everyone who helped us make 2024.7.0 release possible!

@Alexsandruss, @Alexandr-Solovev, @Vika-F, @icfaust, @napetrov, @maria-Petrova, @ahuber21, @ethanglaser, @samir-nasibli, @aepanchi, @emmwalsh, @olegkkruglov, @razdoburdin, @avolkov-intel, @md-shafiul-alam

Full Changelog: 2024.6.0...2024.7.0