LPLR model

[jer2ig]

Addition of logistic partially linear model.

Created PR to check for merging and review changes. Unit tests in work.

To fix the issue, update the signature of set_sample_splitting in DoubleMLLPLR to match that of the method in the base class, SampleSplittingMixin.set_sample_splitting. Since we do not have the exact signature from the base class, but know it requires at least two arguments (likely self, sample_splitting or something similar), the best and most robust fix is to accept at least the same positional arguments and possibly any optional/keyword arguments the base method supports. Typically, when unsure and aiming for compatibility, accepting *args, **kwargs covers any additional arguments expected.

Implement this change in doubleml/plm/lplr.py at the definition of set_sample_splitting. The new method should mirror the old behavior of raising NotImplementedError, but with the correct expanded signature.

---

To fix the error, you must ensure that the code does not use the return value of a function that does not return anything (i.e., a procedure), as this leads to confusing or broken behaviour and runtime errors. The single best solution is to either:

• Implement solve_score to return two values as expected, or
• Remove any expectation of return values, depending on whether such output is intended.

Given the context, the line thetas[i_rep], ses[i_rep] = solve_score(M_hat_list, t_hat_list, m_hat_list) suggests solve_score is expected to return two values (likely a theta and a standard error). Therefore, the best solution is to implement a placeholder solve_score that returns two dummy values (for example, 0.0, 0.0), or, if possible, return appropriate calculated results. Since fit_selection relies on these values for further calculation, returning two float values from solve_score is necessary for functionality.

All code changes are limited to the shown code block, so only edit the definition of solve_score in doubleml/plm/tests/_utils_lplr_manual.py.

---

To fix a genuinely unused local variable, the assignment should be safely removed. In this case, n_obs = len(y) can be deleted from line 96 in fit_nuisance_selection, as it does not impact the function’s logic, outputs, or structure. No further code changes or imports are necessary since no other reference to n_obs exists within the shown scope.

To fix the issue of the unused local variable ml_M in the fit_nuisance_selection function, simply remove the assignment ml_M = clone(learner_M) on line 97. This will clean up the code by eliminating the unused variable without altering any necessary functionality. There are no side effects in the right-hand side (clone(learner_M)) that need to be preserved, since the result is not used elsewhere. Only the assignment to ml_M should be deleted; leave the corresponding assignments for ml_t and ml_m unless they are also unused (this fix only concerns ml_M as requested).

To resolve this issue, the best course of action is to remove the assignment to ml_t entirely from fit_nuisance_selection. This can be safely done since the right-hand side, clone(learner_t), as with scikit-learn's standard behavior, does not have side effects; it simply returns a new estimator object. No further code changes are needed elsewhere in the function, as ml_t is not referenced. The removal should be confined only to the relevant line in fit_nuisance_selection to preserve existing functionality.

To fix this problem without affecting existing functionality, simply remove the unused assignment: M_best_params = [xx.best_params_ for xx in M_tune_res]. As recommended, if the right-hand side has side effects, keep them; in this case, it's a simple list comprehension extracting attributes—no side effects are expected. Verify that deleting this line does not impact later code (within the shown snippet, it does not appear to), so simply delete the line from doubleml/plm/tests/_utils_lplr_manual.py.

To fix the problem, we should remove the assignment to the unused variable t_best_params on line 214. Because its right-hand side (the list comprehension [xx.best_params_ for xx in t_tune_res]) is not kept for a side effect, it is safe to remove the entire line. No other changes are required, as this will not affect the existing functionality of the method. Be careful to only remove this specific line, and not any unrelated code.

---

To fix this issue, we should remove the assignment to m_best_params since it serves no purpose in the current function. Removing it will make the code cleaner and eliminate the static analysis warning. The modification involves simply deleting line 215 (m_best_params = [xx.best_params_ for xx in m_tune_res]) in the tune_nuisance function in doubleml/plm/tests/_utils_lplr_manual.py without impacting any other aspect of program functionality.

---

The best fix is to remove line 217:

t_tune_res = tune_grid_search(t_targets, x, ml_t, smpls, param_grid_t, n_folds_tune)

This eliminates the unused assignment. It is safe since the right-hand side function call doesn't have required side effects—the earlier similar code calls the same function where the result is used, and here it is not. Only remove the assignment, not any functional use elsewhere.

No additional imports, definitions, or method changes are needed. Only this line in doubleml/plm/tests/_utils_lplr_manual.py needs to be deleted.

[JanTeichertKluge]

Please refer to the change in PR #338 , specifically the new mixin. This replaces the class-specific implementation of all sample-splitting methods. I would recommend adapting this to the current main version and using the mixins.