Overview

Extension for scikit-learn is a free software AI accelerator designed to deliver up to 100X acceleration to existing workflows from scikit-learn (e.g. logistic regression models, random forests, support vector machines, among others), the most widely-used Python library for machine learning on tabular data.

This software acceleration is achieved through SIMD instructions, multi-threading, hardware-specific cache optimizations, and algorithmic optimizations.

Features:

• Average speed up of 8.5x on both training and inference over scikit-learn, with equivalent mathematical accuracy.
• Support for different hardware configurations, including GPUs and multi-GPU setups.
• Easy to integrate into existing scikit-learn workflows without code modifications through monkeypatching, or as an API-compatible module sklearnex.
• Can produce much faster predictions from decision-tree models generated by other libraries such as XGBoost or LightGBM.

Acceleration

Benchmarks code

Optimizations

Easiest way to benefit from accelerations from the extension is by patching scikit-learn with it:

• Enable CPU optimizations

  import numpy as np
  from sklearnex import patch_sklearn
  patch_sklearn()
  
  from sklearn.cluster import DBSCAN
  
  X = np.array([[1., 2.], [2., 2.], [2., 3.],
                [8., 7.], [8., 8.], [25., 80.]], dtype=np.float32)
  clustering = DBSCAN(eps=3, min_samples=2).fit(X)
  

• Enable GPU optimizations

  Note: executing on GPU has additional system software requirements - see details.

  import numpy as np
  from sklearnex import patch_sklearn, config_context
  patch_sklearn()
  
  from sklearn.cluster import DBSCAN
  
  X = np.array([[1., 2.], [2., 2.], [2., 3.],
                [8., 7.], [8., 8.], [25., 80.]], dtype=np.float32)
  with config_context(target_offload="gpu:0"):
      clustering = DBSCAN(eps=3, min_samples=2).fit(X)
  

:eyes: Read about other ways to patch scikit-learn.

:eyes: Check out available notebooks for more examples.

Usage without patching

Alternatively, all functionalities are also available under a separate module which can be imported directly, without involving any patching.

• To run on CPU:

  import numpy as np
  from sklearnex.cluster import DBSCAN
  
  X = np.array([[1., 2.], [2., 2.], [2., 3.],
                [8., 7.], [8., 8.], [25., 80.]], dtype=np.float32)
  clustering = DBSCAN(eps=3, min_samples=2).fit(X)
  

• To run on GPU:

  import numpy as np
  from sklearnex import config_context
  from sklearnex.cluster import DBSCAN
  
  X = np.array([[1., 2.], [2., 2.], [2., 3.],
                [8., 7.], [8., 8.], [25., 80.]], dtype=np.float32)
  with config_context(target_offload="gpu:0"):
      clustering = DBSCAN(eps=3, min_samples=2).fit(X)
  

Installation

To install Extension for scikit-learn, run:

pip install scikit-learn-intelex

Package is also offered through other channels such as conda-forge. See all installation instructions in the Installation Guide.

Documentation

• Technical documentation
• Release Notes
• Medium Blog Posts

Extension and oneDAL

Acceleration in patched scikit-learn classes is achieved by replacing calls to scikit-learn with calls to oneDAL (oneAPI Data Analytics Library) behind the scenes:

• oneAPI Data Analytics Library

How to Contribute

We welcome community contributions, check our Contributing Guidelines to learn more.

---

* The Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.