libuplift.meta.multi_model

Uplift models based on multiple classification/regression models.

Classes

MultimodelUpliftRegressor	Multimodel uplift regressor.
TLearnerUpliftRegressor	Multimodel uplift regressor.
MultimodelUpliftClassifier	Multimodel uplift classifier.
TLearnerUpliftClassifier	Multimodel uplift classifier.
MultimodelUpliftLinearRegressor	Uplift regressor with coef_ and intercept_ fields.
MultimodelUpliftLinearRegressorJamesSeparate	Multimodel uplift regressor.
MultimodelUpliftLinearRegressorJamesU	Multimodel uplift regressor.
MultimodelUpliftLinearRegressorMSESeparate	Multimodel uplift regressor.
MultimodelUpliftLinearRegressorMSEU	Multimodel uplift regressor.

Module Contents

class libuplift.meta.multi_model.MultimodelUpliftRegressor(base_estimator=LinearRegression(), ignore_control=False)

Bases: libuplift.base.UpliftRegressorMixin, _MultimodelUpliftModel

Multimodel uplift regressor.

Build separate models for control and all treatments, subtract control predictions from treatment predictions.

Parameters:

base_estimatora sklearn regressor or list of (name, regressor)

tuples. If a list is provided the first model is used for control, successive one for treatments. If different parameters are to be used per each model, the list version must be given. If a single estimator is given it will be cloned for every treatment.

predict(X)

class libuplift.meta.multi_model.TLearnerUpliftRegressor(base_estimator=LinearRegression(), ignore_control=False)

Bases: MultimodelUpliftRegressor

Multimodel uplift regressor.

Build separate models for control and all treatments, subtract control predictions from treatment predictions.

Parameters:

base_estimatora sklearn regressor or list of (name, regressor)

tuples. If a list is provided the first model is used for control, successive one for treatments. If different parameters are to be used per each model, the list version must be given. If a single estimator is given it will be cloned for every treatment.

class libuplift.meta.multi_model.MultimodelUpliftClassifier(base_estimator=LogisticRegression())

Bases: _MultimodelUpliftClassifierBase

Multimodel uplift classifier.

Build separate models for control and all treatments, subtract control predicted probs from treatment predicted probs.

Parameters:

base_estimatora sklearn classifier supporting predict_proba or

list of (name, regressor) tuples. If a list is provided the first model is used for control, successive one for treatments. If different parameters are to be used per each model, the list version must be given. If a single estimator is given it will be cloned for every treatment.

class libuplift.meta.multi_model.TLearnerUpliftClassifier(base_estimator=LogisticRegression())

Bases: MultimodelUpliftClassifier

Multimodel uplift classifier.

Build separate models for control and all treatments, subtract control predicted probs from treatment predicted probs.

Parameters:

base_estimatora sklearn classifier supporting predict_proba or

list of (name, regressor) tuples. If a list is provided the first model is used for control, successive one for treatments. If different parameters are to be used per each model, the list version must be given. If a single estimator is given it will be cloned for every treatment.

class libuplift.meta.multi_model.MultimodelUpliftLinearRegressor(base_estimator=LinearRegression(), ignore_control=False)

Bases: MultimodelUpliftRegressor, sklearn.linear_model._base.LinearModel

Uplift regressor with coef_ and intercept_ fields.

fit(*args, **kwargs)

predict(X)

class libuplift.meta.multi_model.MultimodelUpliftLinearRegressorJamesSeparate(base_estimator=LinearRegression(), ignore_control=False)

Bases: MultimodelUpliftRegressor, sklearn.linear_model._base.LinearModel

Multimodel uplift regressor.

Build separate models for control and all treatments, subtract control predictions from treatment predictions.

Parameters:

base_estimatora sklearn regressor or list of (name, regressor)

tuples. If a list is provided the first model is used for control, successive one for treatments. If different parameters are to be used per each model, the list version must be given. If a single estimator is given it will be cloned for every treatment.

fit(*args, **kwargs)

predict(X)

class libuplift.meta.multi_model.MultimodelUpliftLinearRegressorJamesU(base_estimator=LinearRegression(), ignore_control=False)

Bases: MultimodelUpliftRegressor, sklearn.linear_model._base.LinearModel

Multimodel uplift regressor.

Build separate models for control and all treatments, subtract control predictions from treatment predictions.

Parameters:

base_estimatora sklearn regressor or list of (name, regressor)

tuples. If a list is provided the first model is used for control, successive one for treatments. If different parameters are to be used per each model, the list version must be given. If a single estimator is given it will be cloned for every treatment.

fit(*args, **kwargs)

predict(X)

class libuplift.meta.multi_model.MultimodelUpliftLinearRegressorMSESeparate(base_estimator=LinearRegression(), ignore_control=False)

Bases: MultimodelUpliftRegressor, sklearn.linear_model._base.LinearModel

Multimodel uplift regressor.

Build separate models for control and all treatments, subtract control predictions from treatment predictions.

Parameters:

base_estimatora sklearn regressor or list of (name, regressor)

tuples. If a list is provided the first model is used for control, successive one for treatments. If different parameters are to be used per each model, the list version must be given. If a single estimator is given it will be cloned for every treatment.

fit(*args, **kwargs)

predict(X)

class libuplift.meta.multi_model.MultimodelUpliftLinearRegressorMSEU(base_estimator=LinearRegression(), ignore_control=False)

Bases: MultimodelUpliftRegressor, sklearn.linear_model._base.LinearModel

Multimodel uplift regressor.

Build separate models for control and all treatments, subtract control predictions from treatment predictions.

Parameters:

base_estimatora sklearn regressor or list of (name, regressor)

tuples. If a list is provided the first model is used for control, successive one for treatments. If different parameters are to be used per each model, the list version must be given. If a single estimator is given it will be cloned for every treatment.

fit(*args, **kwargs)

predict(X)