[Repo Assist] test(estimator): add regression test for categorical confounders in conditional effects (#401)

[github-actions]

🤖 This PR was created by Repo Assist, an automated AI assistant.

Summary

Adds test_categorical_common_cause_consistent_encoding to tests/causal_estimators/test_conditional_effects.py as a regression test for #401.

Background

Issue #401 reported that backdoor.linear_regression with categorical common causes (confounders) and numeric effect modifiers raised a shape mismatch:

ValueError: shapes (N, K_subset) and (K_full,) not aligned
```

The root cause: when `_estimate_conditional_effects` groups data by a numeric effect modifier, some strata may not contain all categorical levels of a confounder. The old `pd.get_dummies`-based encoding produced different column counts per stratum, mismatching the model trained on the full dataset.

The fix (already in main via the `Encoders` class in `dowhy/utils/encoding.py`) fits a `sklearn.preprocessing.OneHotEncoder` once on the full dataset and reuses it (`encoder.transform()`) for every stratum, always producing the same number of columns.

## What this PR adds

A single new test `test_categorical_common_cause_consistent_encoding`:

- Creates a 2,000-row dataset with a **3-level categorical confounder** (A/B/C) where level C is deliberately rare (4% of rows) — so it is absent from many effect-modifier strata.
- Runs `backdoor.linear_regression` with a numeric effect modifier, triggering `_estimate_conditional_effects`.
- Asserts:
  - No exception is raised (the old bug would have raised `ValueError` on subset encoding).
  - `conditional_estimates` is a non-empty `pd.Series` with all-finite values.
  - The ATE is close to the true effect (2.0 ± 0.5 tolerance).
  - The caller's DataFrame has not been mutated (no `__categorical__` temporary columns left behind).

## Test Status

```
tests/causal_estimators/test_conditional_effects.py::test_conditional_effects_single_effect_modifier PASSED
tests/causal_estimators/test_conditional_effects.py::test_conditional_effects_multiple_effect_modifiers PASSED
tests/causal_estimators/test_conditional_effects.py::test_conditional_effects_constant_multiple_effect_modifiers PASSED
tests/causal_estimators/test_conditional_effects.py::test_conditional_effects_all_missing_multiple_effect_modifiers_returns_empty_multiindex PASSED
tests/causal_estimators/test_conditional_effects.py::test_categorical_common_cause_consistent_encoding PASSED

5 passed in 3.03s (Python 3.12, pandas 3.0.5, scikit-learn 1.9)

✅ black --check passes
✅ isort --check passes
✅ flake8 — no new errors introduced

Generated by 🌈 Repo Assist, see workflow run. Learn more.

To install this agentic workflow, run

gh aw add githubnext/agentics/workflows/repo-assist.md@11c9a2c442e519ff2b427bf58679f5a525353f76

[Copilot]

Warning

Copilot couldn't run its full agentic review because it didn't start before the timeout. Make sure your repository has a runner available, or add a copilot-code-review.yml file specifying one with the runs-on attribute. See the docs for more details.

Pull request overview

Adds a regression test to ensure conditional effect estimation remains stable when categorical confounder levels are missing from some effect-modifier strata (issue #401), and updates the module docstring to reflect the expanded regression coverage.

Changes:

• Expanded the module docstring to document the categorical-confounder encoding regression context.
• Added test_categorical_common_cause_consistent_encoding to validate consistent one-hot feature dimensions across strata and lack of input DataFrame mutation.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Made this deterministic in 33f4b6c: rows with level C are confined to the top of the effect-modifier range so they fall into a single quantile bin, and added a pre-check assertion that at least one stratum lacks level C before calling estimate_effect.

[Copilot]

Updated in 33f4b6c to assert that no column starts with CausalEstimator.TEMP_CAT_COLUMN_PREFIX and that the original column set is unchanged, instead of the exact ordered list.