libuplift.meta.nested#
Nested models where control outcome predictions are used in an uplift model.
Classes#
Nested regression model. |
|
Dependent Data Representation metamodel. It is a double model |
Module Contents#
- class libuplift.meta.nested.NestedMeanUpliftRegressor(base_estimator=LinearRegression())[source]#
Bases:
libuplift.base.UpliftRegressorMixin,libuplift.meta.base.UpliftMetaModelBaseNested regression model.
First builds a model on controls, then subtracts its training predictions from target. An uplift model is then build on the new target.
Only available for regression models.
- class libuplift.meta.nested.DDRUpliftClassifier(base_estimator=LogisticRegression(), feature_prediction_method='predict_proba', direction='C->T')[source]#
Bases:
libuplift.base.UpliftClassifierMixin,libuplift.meta.base.UpliftMetaModelBaseDependent Data Representation metamodel. It is a double model where control predictions are added as a variable in the treatment model.
The model was proposed in A. Betlei, E. Diemert, and M.-R. Amini Uplift Prediction with Dependent Feature Representation in Imbalanced Treatment and Control Conditions, ICONIP, 2018.
- directionstring, default=”C->T” “C->T” means control
predictions are used as an additional predictor for the treatment model, “T->C” means the reverse: predictions of all treatment models are used (jointly) as predictors for the control model.