Dowhy CausalModel does not have 'causal_estimator' attribute about causal-inference-and-discovery-in-python HOT 2 CLOSED

Hi @Lyutenant - thank you for raising this up.

You are correct, there were some changes in the DoWhy's API since version 0.9+ and .causal_estimator object has been removed from the CausalModel class.

There are two ways of generating predictions on new data using the new API (examples follow the Chapter 09 notebook, S-Learner section):

1. Accessing the fitted estimator from the estimate object:

estimate._estimator_object.effect(earnings_interaction_test.drop(['true_effect', 'took_a_course'], axis=1))

2. Passing new data to the target_units parameter while setting fit_estimator to False:

model.estimate_effect(
    identified_estimand=estimand,
    method_name='backdoor.econml.metalearners.SLearner',
    fit_estimator=False,
    target_units=earnings_interaction_test.drop(['true_effect', 'took_a_course'], axis=1),
).cate_estimates

I personally find the second way somehow less intuitive and less clear than the first one.

I also want to share with you that I encountered some unexpected behaviors in version 0.10.1 when using these methods. I raised an issue with the DoWhy team here: py-why/dowhy#1038

If you asked my opinion, I'd recommend that you stick with DoWhy 0.8 at least for now.

I hope this reply will be helpful to you.

Wish you a joyful journey with the book and happy causal learning!

from causal-inference-and-discovery-in-python.

Hi @AlxndrMlk , thanks for your prompt reply.
I switched back to dowhy 0.8 and the issue is gone. I am taking your advice to stick to 0.8 for now, at least until the issue you have raised with the DoWhy team is solved.

p.s. I do enjoy your book a lot!

from causal-inference-and-discovery-in-python.