D-Wave scikit-learn Plugin

This package provides a scikit-learn transformer for feature selection using a quantum-classical hybrid solver.

This plugin makes use of a Leap™ quantum-classical hybrid solver. Developers can get started by signing up for the Leap quantum cloud service for free. Those seeking a more collaborative approach and assistance with building a production application can reach out to D-Wave directly and also explore the feature selection offering in AWS Marketplace.

The package's main class, SelectFromQuadraticModel, can be used in any existing sklearn pipeline. For an introduction to hybrid methods for feature selection, see the Feature Selection for CQM.

Examples

Basic Usage

A minimal example of using the plugin to select 20 of 30 features of an sklearn dataset:

>>> from sklearn.datasets import load_breast_cancer
>>> from dwave.plugins.sklearn import SelectFromQuadraticModel
... 
>>> X, y = load_breast_cancer(return_X_y=True)
>>> X.shape
(569, 30)
>>> # solver can also be equal to "cqm"
>>> X_new = SelectFromQuadraticModel(num_features=20, solver="nl").fit_transform(X, y)
>>> X_new.shape
(569, 20)

For large problems, the default runtime may be insufficient. You can use the CQM solver's time_limit or Nonlinear (NL) solver's time_limit method to find the minimum accepted runtime for your problem; alternatively, simply submit as above and check the returned error message for the required runtime.

The feature selector can be re-instantiated with a longer time limit.

>>> # solver can also be equal to "nl"
>>> X_new = SelectFromQuadraticModel(num_features=20, time_limit=200, solver="cqm").fit_transform(X, y)

Tuning

You can use SelectFromQuadraticModel with scikit-learn's hyper-parameter optimizers.

For example, the number of features can be tuned using a grid search. Please note that this will submit many problems to the hybrid solver.

>>> import numpy as np
...
>>> from sklearn.datasets import load_breast_cancer
>>> from sklearn.ensemble import RandomForestClassifier
>>> from sklearn.model_selection import GridSearchCV
>>> from sklearn.pipeline import Pipeline
>>> from dwave.plugins.sklearn import SelectFromQuadraticModel
...
>>> X, y = load_breast_cancer(return_X_y=True)
...
>>> num_features = X.shape[1]
>>> searchspace = np.linspace(1, num_features, num=5, dtype=int, endpoint=True)
...
>>> # solver can also be equal to "cqm"
>>> pipe = Pipeline([
>>>   ('feature_selection', SelectFromQuadraticModel(solver="nl")),
>>>   ('classification', RandomForestClassifier())
>>> ])
...
>>> clf = GridSearchCV(pipe, param_grid=dict(feature_selection__num_features=searchspace))
>>> search = clf.fit(X, y)
>>> print(search.best_params_)
{'feature_selection__num_features': 22}

Installation

To install the core package:

pip install dwave-scikit-learn-plugin

License

Released under the Apache License 2.0

Contributing

Ocean's contributing guide has guidelines for contributing to Ocean packages.

Release Notes

dwave-scikit-learn-plugin makes use of reno to manage its release notes.

When making a contribution to dwave-scikit-learn-plugin that will affect users, create a new release note file by running

reno new your-short-descriptor-here

You can then edit the file created under releasenotes/notes/. Remove any sections not relevant to your changes. Commit the file along with your changes.