search

DoubleML / doubleml-docs

Documentation and User Guide for DoubleML - Double Machine Learning in Python & R
https://docs.doubleml.org
BSD 3-Clause "New" or "Revised" License
21 stars 10 forks source link

Deploy issue in notebook py_double_ml_pension #45

Open MalteKurz opened 3 years ago

MalteKurz commented 3 years ago

Seems to be caused by the new scikit-learn in version 1.0.0, see https://github.com/DoubleML/doubleml-docs/runs/3742278774?check_suite_focus=true.

MalteKurz commented 3 years ago

This is really a bit tricky. Seems to be related to this new check https://github.com/scikit-learn/scikit-learn/blob/6d5774f6895aa84a8ec74762bcb29fc7e5173d41/sklearn/linear_model/_base.py#L728-L782 on Gram matrices introduced here https://github.com/scikit-learn/scikit-learn/pull/19004. As far as I can see np.dot() does not always deliver "numerically" the same result if either called with two 2d-arrays (specific element extracted) or with two 1d-arrays corresponding to that specific element. Unfortunately, I am not (yet) able to extract a minimal working example in order to file a issue to scikit-learn.

AlejandroTL commented 1 year ago

Which is the current state of this issue? I have just suffered it using versions: scikit-learn 1.2.2 DoubleML 0.6.1

Thanks a lot!