Intel® Extension for Scikit-learn

2026.0.0 verified Mon Apr 27 auth: no python

Intel® Extension for Scikit-learn (formerly daal4py) accelerates scikit-learn algorithms on Intel hardware via patching. Version 2026.0.0 supports Python >=3.7 and scikit-learn 1.6-1.8. Monthly releases.

pip install scikit-learn-intelex

Common errors

error ModuleNotFoundError: No module named 'sklearnex' ↓

cause Package name is 'scikit-learn-intelex', not 'sklearnex'. Installation missing.

fix

pip install scikit-learn-intelex

error AttributeError: module 'sklearnex' has no attribute 'patch' ↓

cause Function name is patch_sklearn, not patch.

fix

from sklearnex import patch_sklearn; patch_sklearn()

error ValueError: The parameter 'verbose' is not recognized. ↓

cause Some internal parameters changed between scikit-learn versions. Example with LogisticRegression in sklearn 1.8.

fix

Upgrade to scikit-learn-intelex >=2025.10.0 and ensure matching sklearn version.

Warnings

gotcha Call patch_sklearn() after importing scikit-learn, but before creating any estimator instances. Patch order matters. ↓

fix Ensure patch_sklearn() is called early, typically at module level.

deprecated Internal dpctl tensor handling functionality is deprecated and will be removed in a future release. ↓

fix Avoid relying on dpctl tensor internals; use standard NumPy/PyTorch arrays or dpctl.tensor public API.

breaking scikit-learn-intelex does not support all scikit-learn algorithms (e.g., some ensemble methods still run on original scikit-learn). Check the list of supported algorithms. ↓

fix Verify algorithm support in the official documentation before relying on acceleration.

gotcha If you import from sklearnex estimators directly (e.g., from sklearnex.linear_model import Ridge), it may not be patched correctly and can conflict with patched scikit-learn. ↓

fix Use patch_sklearn() and then import from sklearn.* instead of sklearnex.* for consistency.

Install

conda install -c conda-forge scikit-learn-intelex

Imports

patch_sklearn

wrong

from sklearnex import patch

correct

from sklearnex import patch_sklearn

Function name is patch_sklearn, not patch.

unpatch_sklearn
```
from sklearnex import unpatch_sklearn
```

Quickstart

Patch scikit-learn to use Intel-optimized backends, then use scikit-learn API normally.

from sklearnex import patch_sklearn
patch_sklearn()

from sklearn.linear_model import Ridge
ridge = Ridge(alpha=1.0)
ridge.fit([[0, 0], [1, 1]], [0, 1])
print(ridge.predict([[0.5, 0.5]]))