CATE Notebook Error: econml.dml.DMLCateEstimator is not an existing causal estimator about dowhy HOT 6 CLOSED

FYI I got the same error as the previous user, but solved it in a different way.

I tried pip install econml I was getting the error:

ERROR: Failed building wheel for shap

This was solved by conda install -c conda-forge shap.

Then I was able to call model.estimate_effect() with "backdoor.econml" (and as an aside I could also import econml within the notebook as well).

from dowhy.

Hey @tszumowski thanks for trying this out. This error is likely due to an older version of sklearn/scikit-learn. Can you try updating to the latest version and check again?

from dowhy.

@amit-sharma. I was at version 0.22. When I ran pip install --upgrade scikit-learn, it stayed at 0.22. See attached pip-freeze for my full environment.
pip-freeze.txt

I believe the main error may be related to:

ImportError: econml.dml.DMLCateEstimator is not an existing causal estimator.

Not sure if it's related, but If I grep the checkout for DMLCateEstimator, I don't get any source results:

$ rg DMLCateEstimator
docs/source/example_notebooks/dowhy-conditional-treatment-effects.ipynb
259:     "evalue": "econml.dml.DMLCateEstimator is not an existing causal estimator.",
281:      "\u001b[0;31mImportError\u001b[0m: econml.dml.DMLCateEstimator is not an existing causal estimator."

from dowhy.

Ah, not sure why you are receiving this error. Can you try this Binder notebook and check if it works for you? https://mybinder.org/v2/gh/microsoft/dowhy/48a7a79a2836454f3f1b95ac0bd2eee4e504cb4f

Let me also try to fix the error message so that they are more informative. Currently the "is not an existing estimator" is a catch-all message that shows up on any exception. The root issue seems to be

~/.pyenv/versions/3.6.7/envs/aaa3/lib/python3.6/site-packages/econml-0.5-py3.6.egg/econml/utilities.py in <module>
     34 from collections.abc import Iterable
---> 35 from sklearn.model_selection._split import _CVIterableWrapper, CV_WARNING
     36 from sklearn.utils.multiclass import type_of_target

ImportError: cannot import name 'CV_WARNING'

which I thought may be due to sklearn. Not sure, let me know if the Binder link works and then we can debug

from dowhy.

@amit-sharma
I pulled down the latest from master this morning, started with a fresh virtual environment, and installed with

python setup.py install
pip install -r requirements.txt
pip install jupyter

Then I tried the notebook again. It worked.

Very possible I was in the wrong environment or something. Thanks for the help! I'll close this.

from dowhy.

FYI I got the same error as the previous user, but solved it in a different way.

I tried pip install econml I was getting the error:

ERROR: Failed building wheel for shap

This was solved by conda install -c conda-forge shap.

Then I was able to call model.estimate_effect() with "backdoor.econml" (and as an aside I could also import econml within the notebook as well).

Thanks , This worked for me

from dowhy.