scikit-learn-compatible estimators from Civis Analytics

Installation with pip is recommended:

$ pip install civisml-extensions

For development, a few additional dependencies are needed:

$ pip install -r dev-requirements.txt

This package contains scikit-learn-compatible estimators for stacking ( StackedClassifier, StackedRegressor), non-negative linear regression ( NonNegativeLinearRegression), preprocessing pandas DataFrames ( DataFrameETL), and using Hyperband for cross-validating hyperparameters ( HyperbandSearchCV).

Usage of these estimators follows the standard sklearn conventions. Here is an example of using the StackedClassifier:

>>> from sklearn.linear_model import LogisticRegression
>>> from sklearn.ensemble import RandomForestClassifier
>>> from civismlext.stacking import StackedClassifier
>>>
>>> # Define some Train data and labels
>>> Xtrain, ytrain = <train_features>, <train_labels>
>>>
>>> # Note that the final estimator 'metalr' is the meta-estimator
>>> estlist = [('rf', RandomForestClassifier()),
>>>            ('lr', LogisticRegression()),
>>>            ('metalr', LogisticRegression())]
>>>
>>> mysm = StackedClassifier(estlist)
>>> # Set some parameters, if you didn't set them at instantiation
>>> mysm.set_params(rf__random_state=7, lr__random_state=8,
>>>                 metalr__random_state=9, metalr__C=10**7)
>>>
>>> # Fit
>>> mysm.fit(Xtrain, ytrain)
>>>
>>> # Predict!
>>> ypred = mysm.predict_proba(Xtest)

You can learn more about stacking and see an example use of the StackedRegressor and NonNegativeLinearRegression estimators in a talk presented at PyData NYC in November, 2017.

See the doc strings of the various estimators for more information.

Please see CONTRIBUTING.md for information about contributing to this project.

BSD-3

See LICENSE.md for details.