Supported Algorithms

Hint

While not every parameter or combination of parameters from estimators is supported in the Extension for scikit-learn*, attempting to perform an unsupported operation will by default result in executing code from stock scikit-learn when no accelerated routine is available (referred to as a fallback throughout the documentation), thereby ensuring that any valid scikit-learn workflow will continue working under patching.

To verify whether accelerated routines are used for some operation or not, enable verbose mode. See Verbose Mode for details.

Note

Beyond some parameter combinations from estimators not being supported, some features from scikit-learn are not supported as a whole - see Unsupported scikit-learn features.

The Extension for scikit-learn* offers accelerated versions of the following classes and functions from scikit-learn, with some limitations described below:

on CPU

Classification

Algorithm	Parameters	Data formats	Other limitations
sklearn.svm.SVC	kernel must be one of ["linear", "rbf", "poly", "sigmoid"]	Negative weights are not supported.	probability=True is not supported with array API classes other than NumPy. It’s not possible to get accelerated predictions out of a multi-class classifier that was fitted by falling back to scikit-learn.
sklearn.svm.NuSVC	kernel must be one of ["linear", "rbf", "poly", "sigmoid"]	Negative weights are not supported.	probability=True is not supported with array API classes other than NumPy. It’s not possible to get accelerated predictions out of a multi-class classifier that was fitted by falling back to scikit-learn.
sklearn.ensemble.RandomForestClassifier	All parameters are supported except: warm_start = True ccp_alpha != 0 criterion != 'gini' n_estimators > 6024 bootstrap = True and/or max_samples != None are not supported when there are sample weights Non-integer max_samples larger than 1 or integer max_samples greater than number of rows, with bootstrap=True	Multi-output and sparse data are not supported. Missing values and infinite values are not supported.	Number of classes must be at least 2. Nothing will be printed if verbose > 0.
sklearn.ensemble.ExtraTreesClassifier	All parameters are supported except: warm_start = True ccp_alpha != 0 criterion != 'gini' n_estimators > 6024 bootstrap = True and/or max_samples != None are not supported when there are sample weights Non-integer max_samples larger than 1 or integer max_samples greater than number of rows, with bootstrap=True	Multi-output and sparse data are not supported. Missing values and infinite values are not supported.	Number of classes must be at least 2. Nothing will be printed if verbose > 0.
sklearn.neighbors.KNeighborsClassifier	For algorithm == 'kd_tree': all parameters except metric != 'euclidean' or 'minkowski' with p != 2 For algorithm == 'brute': all parameters except metric not in ['euclidean', 'manhattan', 'minkowski', 'chebyshev', 'cosine'] algorithm == 'ball_tree' is not supported.	Multi-output and sparse data are not supported.	Number of classes must be at least 2.
sklearn.linear_model.LogisticRegression	All parameters are supported except: solver not in ['lbfgs', 'newton-cg'] l1_ratio != 0 dual = True sample_weight != None class_weight != None Solver 'newton-cg' with fit_intercept = False is not supported	Sparse data is not supported.	Solver 'newton-cg' is only available in preview mode. Important: this estimator should not be used in parallel Python threads - for concurrent fits (e.g. from sklearn.model_selection.GridSearchCV), process-based parallelism should be used instead (default backend for joblib).
sklearn.linear_model.LogisticRegressionCV	All parameters are supported except: solver not in ['lbfgs', 'newton-cg'] l1_ratios not in [0, "warn"] dual = True sample_weight != None class_weight != None Solver 'newton-cg' with fit_intercept = False is not supported	Sparse data is not supported.	Estimator is only available in preview mode.

Regression

Algorithm	Parameters	Data formats	Other limitations
sklearn.svm.SVR	kernel must be one of ["linear", "rbf", "poly", "sigmoid"]		Negative weights are not supported.
sklearn.svm.NuSVR	kernel must be one of ["linear", "rbf", "poly", "sigmoid"]		Negative weights are not supported.
sklearn.ensemble.RandomForestRegressor	All parameters are supported except: warm_start = True ccp_alpha != 0 criterion != 'squared_error' n_estimators > 6024 bootstrap = True and/or max_samples != None are not supported when there are sample weights Non-integer max_samples larger than 1 or integer max_samples greater than number of rows, with bootstrap=True	Multi-output and sparse data are not supported. Missing values and infinite values are not supported.	Nothing will be printed if verbose > 0.
sklearn.ensemble.ExtraTreesRegressor	All parameters are supported except: warm_start = True ccp_alpha != 0 criterion != 'squared_error' n_estimators > 6024 bootstrap = True and/or max_samples != None are not supported when there are sample weights Non-integer max_samples larger than 1 or integer max_samples greater than number of rows, with bootstrap=True	Multi-output and sparse data are not supported. Missing values and infinite values are not supported.	Nothing will be printed if verbose > 0.
sklearn.neighbors.KNeighborsRegressor	For algorithm == 'kd_tree': all parameters except metric != 'euclidean' or 'minkowski' with p != 2 For algorithm == 'brute': all parameters except metric not in ['euclidean', 'manhattan', 'minkowski', 'chebyshev', 'cosine'] algorithm == 'ball_tree' is not supported.	Multi-output and sparse data are not supported
sklearn.linear_model.LinearRegression	All parameters are supported except: sample_weight != None positive = True (this is supported through the class sklearn.linear_model.ElasticNet)	Only dense data is supported.
sklearn.linear_model.Ridge	All parameters are supported except: solver != 'auto' sample_weight != None positive = True (this is supported through the class sklearn.linear_model.ElasticNet) alpha must be a scalar	Only dense data is supported.
sklearn.linear_model.ElasticNet	All parameters are supported except: sample_weight != None	Sparse data is not supported.	Estimator is only available in preview mode.
sklearn.linear_model.Lasso	All parameters are supported except: sample_weight != None	Sparse data is not supported.	Estimator is only available in preview mode.

Clustering

Algorithm	Parameters	Data formats
sklearn.cluster.KMeans	All parameters are supported except: algorithm != 'lloyd' (‘elkan’ falls back to ‘lloyd’) n_clusters = 1 sample_weight must be None, constant, or equal weights verbose = True will only print results from the last iteration, and will only print inertia numbers, not ‘convergence achieved’ messages.	No limitations
sklearn.cluster.DBSCAN	All parameters are supported except: metric != 'euclidean' or 'minkowski' with p != 2 algorithm not in ['brute', 'auto']	Only dense data is supported

Dimensionality Reduction

Algorithm	Parameters	Data formats	Other limitations
sklearn.decomposition.PCA	All parameters are supported except: svd_solver not in ['full', 'covariance_eigh']	Sparse data is not supported
sklearn.decomposition.IncrementalPCA		Sparse data is not supported	Estimator is only available in preview mode.
sklearn.manifold.TSNE	All parameters are supported except: metric != ‘euclidean’ or 'minkowski' with p != 2 n_components can only be 2 method != "barnes_hut"	Sparse data is not supported for the initialization and distance calculation stages.

Anomaly Detection

Algorithm	Parameters	Data formats
sklearn.neighbors.LocalOutlierFactor	For algorithm == 'kd_tree': all parameters except metric != 'euclidean' or 'minkowski' with p != 2 For algorithm == 'brute': all parameters except metric not in ['euclidean', 'manhattan', 'minkowski', 'chebyshev', 'cosine'] algorithm == 'ball_tree' is not supported.	Sparse data is not supported

Nearest Neighbors

Algorithm	Parameters	Data formats
sklearn.neighbors.NearestNeighbors	For algorithm == 'kd_tree': all parameters except metric != 'euclidean' or 'minkowski' with p != 2 For algorithm == 'brute': all parameters except metric not in ['euclidean', 'manhattan', 'minkowski', 'chebyshev', 'cosine'] algorithm == 'ball_tree' is not supported.	Sparse data is not supported

Other Tasks

Algorithm	Parameters	Data formats	Other limitations
sklearn.covariance.EmpiricalCovariance	All parameters are supported	Only dense data is supported	Estimator is only available in preview mode.
sklearnex.basic_statistics.BasicStatistics	All parameters are supported	Supported data formats: Dense data CSR sparse matrices	Sample weights are not supported for CSR data format
sklearn.model_selection.train_test_split	All parameters are supported	Supported data formats: Only dense data is supported Only integer and 32/64-bits floating point types are supported Data with more than 3 dimensions is not supported Only np.ndarray, pd.DataFrame and pd.Series inputs are supported.
sklearn.metrics.pairwise_distances	All parameters are supported except: metric not in ['cosine', 'correlation']	Supported data formats: Only dense data is supported Y must be None Input dtype must be np.float64
sklearn.metrics.roc_auc_score	All parameters are supported except: average != None sample_weight != None max_fpr != None multi_class != None	Only binary y_true is supported

on GPU

See also

GPU support

Classification

Algorithm	Parameters	Data formats	Other limitations
sklearn.svm.SVC	All parameters are supported except: kernel not in ["linear", "rbf"] class_weight != None probability != False (see limitations on this)	Only dense data is supported. Negative weights are not supported.	Only binary classification is supported. If passing probability=True, under target_offload, the initial model will be fitted on GPU, but subsequent calibration will be done on CPU through scikit-learn; whereas if passing array API inputs with probability=True, the whole procedure will fall back to scikit-learn on CPU.
sklearn.ensemble.RandomForestClassifier	All parameters are supported except: warm_start = True ccp_alpha != 0 criterion != 'gini' oob_score = True sample_weight != None n_estimators > 6024 bootstrap = True and/or max_samples != None are not supported when there are sample weights Non-integer max_samples larger than 1 or integer max_samples greater than number of rows, with bootstrap=True	Multi-output and sparse data are not supported. Missing values and infinite values are not supported.	Number of classes must be at least 2. Nothing will be printed if verbose > 0.
sklearn.ensemble.ExtraTreesClassifier	All parameters are supported except: warm_start = True ccp_alpha != 0 criterion != 'gini' oob_score = True sample_weight != None n_estimators > 6024 bootstrap = True and/or max_samples != None are not supported when there are sample weights Non-integer max_samples larger than 1 or integer max_samples greater than number of rows, with bootstrap=True	Multi-output and sparse data are not supported. Missing values and infinite values are not supported.	Number of classes must be at least 2. Nothing will be printed if verbose > 0.
sklearn.neighbors.KNeighborsClassifier	All parameters are supported except: algorithm != 'brute' weights = 'callable' metric not in ['euclidean', 'manhattan', 'minkowski', 'chebyshev', 'cosine']	Only dense data is supported.	Number of classes must be at least 2. The following methods are not accelerated by Extension for scikit-learn* and will fall back to scikit-learn on CPU, returning NumPy arrays when using array API inputs: radius_neighbors(), radius_neighbors_graph().
sklearn.linear_model.LogisticRegression	All parameters are supported except: solver != 'newton-cg' class_weight != None sample_weight != None dual = True intercept_scaling != 1 warm_start = True l1_ratio != 0	No limitations.	Only binary classification is supported.

Regression

Algorithm	Parameters	Data formats
sklearn.ensemble.RandomForestRegressor	All parameters are supported except: warm_start = True ccp_alpha != 0 criterion != 'squared_error' oob_score = True sample_weight != None n_estimators > 6024 bootstrap = True and/or max_samples != None are not supported when there are sample weights Non-integer max_samples larger than 1 or integer max_samples greater than number of rows, with bootstrap=True Nothing will be printed if verbose > 0	Multi-output and sparse data are not supported. Missing values and infinite values are not supported.
sklearn.ensemble.ExtraTreesRegressor	All parameters are supported except: warm_start = True ccp_alpha != 0 criterion != 'squared_error' oob_score = True sample_weight != None n_estimators > 6024 bootstrap = True and/or max_samples != None are not supported when there are sample weights Non-integer max_samples larger than 1 or integer max_samples greater than number of rows, with bootstrap=True Nothing will be printed if verbose > 0	Multi-output and sparse data are not supported. Missing values and infinite values are not supported.
sklearn.neighbors.KNeighborsRegressor	All parameters are supported except: algorithm != 'brute' weights = 'callable' metric not in ['euclidean', 'manhattan', 'minkowski', 'chebyshev', 'cosine']	Only dense data is supported. The following methods are not accelerated by Extension for scikit-learn* and will fall back to scikit-learn on CPU, returning NumPy arrays when using array API inputs: radius_neighbors(), radius_neighbors_graph().
sklearn.linear_model.Ridge	All parameters are supported except: solver != 'auto' sample_weight != None positive = True alpha must be a scalar	Only dense data is supported.
sklearn.linear_model.LinearRegression	All parameters are supported except: sample_weight != None positive = True	Only dense data is supported.

Clustering

Algorithm	Parameters	Data formats
sklearn.cluster.KMeans	All parameters are supported except: algorithm != 'lloyd' (‘elkan’ falls back to ‘lloyd’) n_clusters = 1 sample_weight must be None, constant, or equal weights init = 'k-means++' falls back to CPU verbose = True will only print results from the last iteration, and will only print inertia numbers, not ‘convergence achieved’ messages.	No limitations
sklearn.cluster.DBSCAN	All parameters are supported except: metric != 'euclidean' algorithm not in ['brute', 'auto']	Only dense data is supported

Dimensionality Reduction

Algorithm	Parameters	Data formats	Other limitations
sklearn.decomposition.PCA	All parameters are supported except: svd_solver not in ['full', 'covariance_eigh']	Sparse data is not supported
sklearn.decomposition.IncrementalPCA		Sparse data is not supported	Estimator is only available in preview mode. Additional parameter svd_solver (see Additional parameters in estimators and functions) is not supported on GPU.

Anomaly Detection

Algorithm	Parameters	Data formats	Other limitations
sklearn.neighbors.LocalOutlierFactor	All parameters are supported except: algorithm != 'brute' metric not in ['euclidean', 'manhattan', 'minkowski', 'chebyshev', 'cosine']	Only dense data is supported	If using target_offload, some computations outside of neighbor calculations (related to thresholds for outlierness) might happen on CPU.

Nearest Neighbors

Algorithm	Parameters	Data formats
sklearn.neighbors.NearestNeighbors	All parameters are supported except: algorithm != 'brute' metric not in ['euclidean', 'manhattan', 'minkowski', 'chebyshev', 'cosine']	Only dense data is supported. The following methods are not accelerated by Extension for scikit-learn* and will fall back to scikit-learn on CPU, returning NumPy arrays when using array API inputs: radius_neighbors(), radius_neighbors_graph().
sklearn.neighbors.LocalOutlierFactor	All parameters are supported except: algorithm != 'brute' metric not in ['euclidean', 'manhattan', 'minkowski', 'chebyshev', 'cosine']	Only dense data is supported.

Other Tasks

Algorithm	Parameters	Data formats	Other limitations
sklearn.covariance.EmpiricalCovariance	All parameters are supported	Only dense data is supported	Estimator is only available in preview mode.
sklearnex.basic_statistics.BasicStatistics	All parameters are supported	Supported data formats: Dense data CSR sparse matrices	Sample weights are not supported for CSR data format.

SPMD Support

See also

SPMD (multi-GPU distributed mode)

Classification

Algorithm	Parameters & Methods	Data formats	Other limitations
sklearn.ensemble.RandomForestClassifier	All parameters are supported except: warm_start = True ccp_alpha != 0 criterion != 'gini' oob_score = True sample_weight != None n_estimators > 6024 bootstrap = True and/or max_samples != None are not supported when there are sample weights Non-integer max_samples larger than 1 or integer max_samples greater than number of rows, with bootstrap=True	Multi-output and sparse data are not supported. Missing values and infinite values are not supported.	Number of classes must be at least 2. Nothing will be printed if verbose > 0.
sklearn.ensemble.ExtraTreesClassifier	All parameters are supported except: warm_start = True ccp_alpha != 0 criterion != 'gini' oob_score = True sample_weight != None n_estimators > 6024 bootstrap = True and/or max_samples != None are not supported when there are sample weights Non-integer max_samples larger than 1 or integer max_samples greater than number of rows, with bootstrap=True	Multi-output and sparse data are not supported. Missing values and infinite values are not supported.	Number of classes must be at least 2. Nothing will be printed if verbose > 0.
sklearn.neighbors.KNeighborsClassifier	All parameters are supported except: algorithm != 'brute' weights = 'callable' metric not in ['euclidean', 'manhattan', 'minkowski', 'chebyshev', 'cosine'] predict_proba method not supported	Only dense data is supported.	Number of classes must be at least 2.
sklearn.linear_model.LogisticRegression	All parameters are supported except: solver != 'newton-cg' class_weight != None sample_weight != None dual = True intercept_scaling != 1 warm_start = True l1_ratio != 0	Method score is not supported.	Only binary classification is supported

Regression

Algorithm	Parameters & Methods	Data formats
sklearn.ensemble.RandomForestRegressor	All parameters are supported except: warm_start = True ccp_alpha != 0 criterion != 'squared_error' oob_score = True sample_weight != None n_estimators > 6024 bootstrap = True and/or max_samples != None are not supported when there are sample weights Non-integer max_samples larger than 1 or integer max_samples greater than number of rows, with bootstrap=True Nothing will be printed if verbose > 0	Multi-output and sparse data are not supported. Missing values and infinite values are not supported.
sklearn.ensemble.ExtraTreesRegressor	All parameters are supported except: warm_start = True ccp_alpha != 0 criterion != 'squared_error' oob_score = True sample_weight != None n_estimators > 6024 bootstrap = True and/or max_samples != None are not supported when there are sample weights Non-integer max_samples larger than 1 or integer max_samples greater than number of rows, with bootstrap=True Nothing will be printed if verbose > 0	Multi-output and sparse data are not supported. Missing values and infinite values are not supported.
sklearn.neighbors.KNeighborsRegressor	All parameters are supported except: algorithm != 'brute' weights = 'callable' metric != 'euclidean' or 'minkowski' with p != 2	Only dense data is supported
sklearn.linear_model.LinearRegression	All parameters are supported except: sample_weight != None positive = True	Only dense data is supported.

Clustering

Algorithm	Parameters & Methods	Data formats
sklearn.cluster.KMeans	All parameters are supported except: algorithm != 'lloyd' (‘elkan’ falls back to ‘lloyd’) n_clusters = 1 sample_weight must be None, constant, or equal weights init = 'k-means++' falls back to CPU verbose = True will only print results from the last iteration, and will only print inertia numbers, not ‘convergence achieved’ messages.	No limitations
sklearn.cluster.DBSCAN	All parameters are supported except: metric != 'euclidean' algorithm not in ['brute', 'auto']	Only dense data is supported

Dimensionality Reduction

Algorithm	Parameters & Methods	Data formats	Other limitations
sklearn.decomposition.PCA	All parameters are supported except: svd_solver not in ['full', 'covariance_eigh']	Sparse data is not supported
sklearn.decomposition.IncrementalPCA	All parameters are supported except: svd_solver not in ['full', 'covariance_eigh']	Sparse data is not supported	Estimator is only available in preview mode.

Nearest Neighbors

Algorithm	Parameters	Data formats
sklearn.neighbors.NearestNeighbors	All parameters are supported except: algorithm != 'brute' metric not in ['euclidean', 'manhattan', 'minkowski', 'chebyshev', 'cosine']	Only dense data is supported

Other Tasks

Algorithm	Parameters	Data formats	Other limitations
sklearn.covariance.EmpiricalCovariance	All parameters are supported	Only dense data is supported	Estimator is only available in preview mode.
sklearnex.basic_statistics.BasicStatistics	All parameters are supported	Supported data formats: Dense data CSR sparse matrices	Sample weights not supported for CSR data format