Which DML model to use for a binary treatment and a binary outcome?

[ctrivino1]

Hi I am trying to perform causal analysis with an outcome and treatment that are both binary. Here is my code:

'''
obj_dml_data = dml.DoubleMLData(data,y_col='binary_outcome',d_cols='binary_treatment')

dml_mlplr = dml.DoubleMLPLR(obj_dml_data,
ml_l = RandomForestClassifier(),
ml_m = RandomForestClassifier(),
n_folds = 5)

summary = dml_mlplr.fit(store_predictions=True).summary

summary
"""

Outcome:
data = {
'oef': ['binary_treatment'],
'std err': [0.509206],
't': [44.761216],
'P>|t|': [0.0],
'2.5 %': [0.48691],
'97.5 %': [0.531503]
}

I do get an output but I see this warning and would like to get some clarification on what it means:
'Learner provided for ml_l is probably invalid: RandomForestClassifier() is (probably) no regressor.'

What exactly does this mean? do I need to you a different dml model other than DoubleMLPLR?
Thanks!

[SvenKlaassen]

Currently, we assume that the learner for ml_l ist a regressor see DoubleMLPLR as the predict() method is used to obtain predictions for $\mathbb{E}[Y|X]$. This can lead to some misleading results if a classifier is used (here we would have to rely on predict_proba().

You should use a regressor or external predictions for ml_l, but I think it would be a good idea to extend the model to correctly use classifiers.

[ctrivino1]

That would be a great feature to add, If you guys implement it then please let me know! Thanks