GridSearchCV & RandomizedSearchCV GPU support #2406

Activity

1. napetrov

   added

   enhancementNew feature or requestNew feature or request

   on Apr 6, 2025
2. napetrov

   added the

   Feature

   issue type on Apr 6, 2025
3. david-cortes-intel commented on Aug 8, 2025

   

   david-cortes-intel

   on Aug 8, 2025

   Contributor

   More actions

   This is now solved on the dpctl side: IntelPython/dpctl#2128

   Should be possible to use the classes from sklearn with sklearnex estimators and dpctl/dpnp arrays after version 0.21 of dpctl.
4. david-cortes-intel commented on Oct 28, 2025

   

   david-cortes-intel

   on Oct 28, 2025

   Contributor

   More actions

   This is now possible to do with array API classes.

   Example with DPNP (>= 0.19):

   import os
   os.environ["SKLEARNEX_VERBOSE"] = "INFO"
   os.environ["SCIPY_ARRAY_API"] = "1"
   import numpy as np
   import dpnp
   from sklearnex import config_context
   from sklearnex.linear_model import Ridge
   from sklearn.datasets import make_regression
   from sklearn.model_selection import GridSearchCV, KFold
   
   X_cpu, y_cpu = make_regression(random_state=123)
   X_gpu = dpnp.array(X_cpu, dtype=np.float32, device="gpu")
   y_gpu = dpnp.array(y_cpu, dtype=np.float32, device="gpu")
   
   cv = GridSearchCV(
       estimator=Ridge(),
       param_grid={"alpha": [1e-1, 1, 1e1]},
       n_jobs=1,
       refit=True,
       cv=KFold(
           n_splits=5,
           shuffle=True,
           random_state=123,
       ),
   )
   with config_context(array_api_dispatch=True):
       cv.fit(X_gpu, y_gpu)

   Example with PyTorch:

   import os
   os.environ["SKLEARNEX_VERBOSE"] = "INFO"
   os.environ["SCIPY_ARRAY_API"] = "1"
   import torch
   from sklearnex import config_context
   from sklearnex.linear_model import Ridge
   from sklearn.datasets import make_regression
   from sklearn.model_selection import GridSearchCV, KFold
   
   X_cpu, y_cpu = make_regression(random_state=123)
   X_gpu = torch.tensor(X_cpu, dtype=torch.float32, device="xpu")
   y_gpu = torch.tensor(y_cpu, dtype=torch.float32, device="xpu")
   
   cv = GridSearchCV(
       estimator=Ridge(),
       param_grid={"alpha": [1e-1, 1, 1e1]},
       n_jobs=1,
       refit=True,
       cv=KFold(
           n_splits=5,
           shuffle=True,
           random_state=123,
       ),
   )
   with config_context(array_api_dispatch=True):
       cv.fit(X_gpu, y_gpu)
5. david-cortes-intel

   closed this as completedon Oct 28, 2025