LPLR model #365
LPLR model #365
Conversation
Added test set-up
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
SvenKlaassen
left a comment
SvenKlaassen
left a comment
There was a problem hiding this comment.
Thanks. Looks already quite great.
Can you further add two test files test_model_defaults.py and test_return_types.py to the plm submodule similar to (https://github.com/DoubleML/doubleml-for-py/blob/main/doubleml/did/tests/test_return_types.py and https://github.com/DoubleML/doubleml-for-py/blob/main/doubleml/did/tests/test_model_defaults.py).
Only add the corresponding tests for DoubleMLLPLR and I can move the tests from the other classes from the tests directory later.
doubleml/plm/lplr.py
Outdated
| def _nuisance_tuning( | ||
| self, smpls, param_grids, scoring_methods, n_folds_tune, n_jobs_cv, search_mode, n_iter_randomized_search | ||
| ): | ||
| if self._i_rep is None: |
There was a problem hiding this comment.
Especially since we would like to remove fold-specific hyperparameters.
I've added them. I had to change the check function a little bit. |
SvenKlaassen
left a comment
SvenKlaassen
left a comment
There was a problem hiding this comment.
Thanks. I added skips for the docstring tests and evaluate_learners() to the tests (this needed some update to skip evaluation if targets are not available (for ml_t).
Could you add in comments in the _nuisance_tuning if some approximations are done to allow simple on fold tuning?
Else i only had one other single comment.
| res["preds"][smpls_single_split[1]] += model.predict_proba(x[smpls_single_split[1]])[:, 1] | ||
| else: | ||
| res["preds"][smpls_single_split[1]] += model.predict(x[smpls_single_split[1]]) | ||
| res["preds"] /= len(smpls) |
Addition of logistic partially linear model.
Created PR to check for merging and review changes. Unit tests in work.