Issue on page /21-Meta-Learners.html

[kevin-mercurio]

The implementation of the final estimator is slightly off relative to the publication:

x_cate_train = (ps_predict(train,0)mx0.predict(train[X]) + ps_predict(train,1)mx1.predict(train[X]))

x_cate_test = test.assign(cate=(ps_predict(test,0)mx0.predict(test[X]) + ps_predict(test,1)mx1.predict(test[X])))

The propensity score should actually be swapped, such that your estimators are:

def ps_predict(df, t): return g.predict_proba(df[X])[:, t]

x_cate_train = (ps_predict(train,1)mx0.predict(train[X]) + ps_predict(train,0)mx1.predict(train[X]))

x_cate_test = test.assign(cate=(ps_predict(test,1)mx0.predict(test[X]) + ps_predict(test,0)mx1.predict(test[X])))

[matheusfacure]

Solved in another issue