Extension for scikit-learn#
In addition to the Python API, AOCL-DA offers an extension to enable existing users of scikit-learn to extract better performance while making minimal changes to their code.
To use the extension, you must patch your existing code to replace the scikit-learn symbols with AOCL-DA symbols. This can be done by inserting the following lines prior to your scikit-learn import statement.
from aoclda.sklearn import skpatch, undo_skpatch
skpatch()
You can switch back to standard scikit-learn using
undo_skpatch()
Note that after calling undo_skpatch, you must reimport scikit-learn.
The skpatch and undo_skpatch functions can also be called with string or list arguments, specifying which scikit-learn package should be patched, for example:
skpatch("PCA")
skpatch(["LinearRegression", "Ridge"])
Alternatively, you may wish to use the aoclda.sklearn module from the command line, without
making any changes to your own code:
python -m aoclda.sklearn your_python_script.py
python -m aoclda.sklearn -m your_python_module
The following scikit-learn classes are currently available in the AOCL-DA extension.
scikit-learn class |
Notes |
|---|---|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Note that only a subset of the AOCL-DA functionality is available in this manner, and if, after
patching, you attempt to call class member functions which have not been implemented by AOCL-DA,
then a RuntimeError will be thrown. It is recommended that for the full benefit of using AOCL-DA
you use the Python APIs described on the subsequent pages of this manual.
The scikit-learn dispatcher#
Since the AOCL-DA scikit-learn extension is not yet feature-complete, we also offer a dispatcher which can automatically select functions from Intel’s Extension for scikit-learn where they are known to perform well.
To enable this dispatcher, set the environment variable USE_INTEL_SKLEARNEX. The AOCL-DA scikit-learn extension will then detect your Intel installation and use it where appropriate.