[Repo Assist] Monthly Activity 2026-04

[github-actions]

🤖 Repo Assist here — I'm an automated AI assistant for this repository.

Activity for April 2026

Suggested Actions for Maintainer

Comprehensive list of all pending actions requiring maintainer attention (excludes items already actioned and checked off).

• Review PR (new): test — add initial test coverage for gcm.fitting_sampling module (13 tests for fit, fit_causal_model_of_target, draw_samples) — Review
• Review PR [Repo Assist] fix: support multiple treatments in PlaceboTreatmentRefuter (closes #251) #1467: fix — support multiple treatments in PlaceboTreatmentRefuter (closes #251) — Review
• Review PR [Repo Assist] test: increase sample size in linear classification auto-assignment test #1466: test — increase sample size in linear classification auto-assignment test to fix flakiness — Review
• Review PR [Repo Assist] fix: correct exact_match_cols logic in DistanceMatchingEstimator (closes #814) #1465: fix — correct exact_match_cols logic in DistanceMatchingEstimator + 12 new tests (closes backdoor.distance_matching: Problem with exact matching #814) — Review
• Review PR [Repo Assist] feat: auto-select default estimation method when method_name is omitted #1464: feat — auto-select default estimation method when method_name is omitted (closes CausalModel.estimate_effect - UnboundLocalError: local variable 'identifier_name' referenced before assignment #365) — Review
• Review PR [Repo Assist] fix: update second-stage model estimand when pre-instantiated in TwoStageRegression #1462: fix — update second-stage model estimand when pre-instantiated in TwoStageRegression (closes Issue with Logistic Regression in Mediation Analysis #1335) — Review
• Review PR [Repo Assist] fix: handle list-form treatment_variable in estimate_effect_naive #1461: fix — handle list-form treatment_variable in estimate_effect_naive (closes Problems with multi-valued treatment and parameter 'evaluate_effect_strength' #416) — Review
• Review PR [Repo Assist] fix: correct categorical groupby KeyError and pandas FutureWarning in dummy_outcome_refuter #1460: fix — correct categorical groupby KeyError and pandas FutureWarning in dummy_outcome_refuter (closes Warnings using causal.model #1316) — Review
• Review PR Fix #1427: [Repo Assist] ci: pin KyleMayes/install-llvm-action to v2.... #1459: human contributor fix — pin KyleMayes/install-llvm-action to v2.0.5 (closes Repo Assist PR [Repo Assist] ci: pin KyleMayes/install-llvm-action to v2.0.5 across all workflows #1427) — Review
• Review PR [Repo Assist] fix: use string key for column lookup in conditional_MI (fixes KeyError with multi-char column names) #1455: fix — use string key in conditional_MI to fix KeyError with multi-char column names (closes Error handling column names in CIT as used by CausalModel.graph_refute method #949) — Review
• Review PR [Repo Assist] fix: implement asymptotic CI/SE via Delta method for LinearRegressionEstimator with effect modifiers #1432: fix — asymptotic CI/SE via Delta method for LinearRegressionEstimator with effect modifiers (closes Add asymptotic confidence intervals for average treatment effect for linear regression with effect modifiers #336) — Review
• Review PR [Repo Assist] perf: eliminate redundant set allocations and fix per-column kernel precision #1437: perf — eliminate redundant set allocations and fix per-column kernel precision — Review
• Review PR [Repo Assist] fix: pass encoded effect modifiers to EconML when categorical variables present #1435: fix — pass encoded effect modifiers to EconML when categorical variables present (closes Estimate Effect fails with Econml DML estimator #820) — Review
• Review PR [Repo Assist] fix: compute proper chi-squared p-value for CMI independence test in GraphRefuter #1431: fix — compute proper chi-squared p-value for CMI independence test in GraphRefuter (closes Conditional independence test on mutual information is not a statistical test of independence #413) — Review
• Review PR [Repo Assist] test: add tests for RandomCommonCause refuter #1413: test — add 6 tests for RandomCommonCause refuter — Review
• Review PR [Repo Assist] fix: preserve DataFrame index in one_hot_encode to fix data_subset_refuter bug with categorical columns #1408: fix — preserve DataFrame index in one_hot_encode (closes #1372) — Review
• Close PR [Repo Assist] ci: pin KyleMayes/install-llvm-action to v2.0.5 across all workflows #1427: superseded by human contributor PR Fix #1427: [Repo Assist] ci: pin KyleMayes/install-llvm-action to v2.... #1459 — View
• Close or merge PR fix: categorical encoding bug in Delta-method CI (review of #1432) #1444: categorical encoding fix incorporated into [Repo Assist] fix: implement asymptotic CI/SE via Delta method for LinearRegressionEstimator with effect modifiers #1432 directly — View
• Manual action needed for issue [Repo Assist] ci: remove LLVM/Clang install step (no longer needed with scipy ≥1.15) #1443: Repo Assist cannot push workflow files; human must apply LLVM-removal patch (instructions in issue body) — View
• Check comment on In gcm, allow building a model in presence of missing values if an appropriate estimator is selected #1300: Repo Assist explained current NaN-handling behaviour and provided implementation sketch for missing-value support in gcm — View
• Check comment on Placebo refuter with multiple treatments #251: Repo Assist fixed multi-treatment placebo refuter bug — PR submitted — View
• Check comment on effect_modifiers #767: Repo Assist explained num_quantiles_to_discretize_cont_cols must go in method_params["init_params"] — View
• Check comment on Estimate refutation throws error: PicklingError: logger cannot be pickled #953: Repo Assist diagnosed PicklingError root cause (Python < 3.12 loggers) and workarounds — View
• Check comment on Problems with multi-valued treatment and parameter 'evaluate_effect_strength' #416: Repo Assist linked to PR [Repo Assist] fix: handle list-form treatment_variable in estimate_effect_naive #1461 which fixes the multi-valued treatment bug — View
• Check comment on backdoor.distance_matching: Problem with exact matching #814: Repo Assist identified 3 bugs in exact_match_cols, fix PR submitted — View
• Check comment on CausalModel.estimate_effect - UnboundLocalError: local variable 'identifier_name' referenced before assignment #365: Repo Assist implemented the feature design from Jan 2022 — PR [Repo Assist] feat: auto-select default estimation method when method_name is omitted #1464 created — View
• Check comment on 'fit_estimator' argument in 'estimate_effect' function of 'CausalModel' object #414: Repo Assist confirmed fit_estimator is now supported; advise users to upgrade — View
• Check comment on Setting the seed for estimate effect methods #418: Repo Assist explained propensity_score_model with random_state=42 workaround — View
• Check comment on Natural Direct Effect estimation does not support multiple mediators #966: Repo Assist noted that PR fix: identify all valid mediators in NIE/NDE estimation #1418 by @kvr06-ai addresses the multiple-mediator NDE/NIE bug — View
• Check comment on estimate_effect() got an unexpected keyword argument 'num_quantiles_to_discretize_cont_cols' #1035: Repo Assist clarified num_quantiles_to_discretize_cont_cols must be inside method_params — View
• Check comment on Error in Conditional Effect Estimation in backdoor.linear_regression_estimator #401: Repo Assist explained that the pd.get_dummies bug is fixed via sklearn OneHotEncoder — View
• Check comment on Replication of DoWhy backdoor.linear_regression with multiple treatments and effect modifiers #1367: Repo Assist showed how to access the statsmodels model via estimate.estimator.model — View
• Check comment on Mediation analysis and (likely) IV bug #946: Repo Assist traced NDE bug and linked fix PR — View
• Check comment on p-value NaN in some pathological cases with non-bootstrap method #807: Repo Assist explained NaN p-value root cause; fix PR [Repo Assist] fix: return p_value=1 when refuter simulations have zero variance (closes #807) #1454 has been merged — View
• Check comment on model.estimate_effect and model.refute_astimate throws 'A column-vector y was passed ...' error #1212: Repo Assist explained root cause (EconML array-valued CATE vs scalar bootstrap test) — View
• Check comment on Adding model validation after user specifies model #412: Repo Assist summarised current graph validation state — View
• Check comment on Add asymptotic confidence intervals for average treatment effect for linear regression with effect modifiers #336: Repo Assist welcomed contributor and summarised implementation approach — View
• Check comment on Remove graph-learner API from DoWhy and point to causal-learn / dodiscover instead #1039: Repo Assist summarised rationale for deprecating graph_learners — View
• Check comment on Add video/screenshot guide for creating graph in DAGitty and importing into DoWhy #529: Repo Assist outlined DAGitty → DoWhy guide structure — View
• Check comment on Error handling column names in CIT as used by CausalModel.graph_refute method #949: Repo Assist traced KeyError root cause and linked fix PR [Repo Assist] fix: use string key for column lookup in conditional_MI (fixes KeyError with multi-char column names) #1455 — View
• Check comment on The causal structure returned by causal-learn is used in DoWhy. #1365: Repo Assist showed how to convert causal-learn PC graph to NetworkX DiGraph — View

Future Work for Repo Assist

• Deprecation PR for dowhy.graph_learners (once maintainers confirm direction on Remove graph-learner API from DoWhy and point to causal-learn / dodiscover instead #1039)
• Refactor DistanceMatchingEstimator.estimate_effect to reduce cyclomatic complexity (currently 22, limit is 10)
• Add __getstate__/__setstate__ to CausalEstimator to fix logger pickling (Estimate refutation throws error: PicklingError: logger cannot be pickled #953) — a proper fix for Python < 3.12 users
• Implement missing-value support in gcm fitting (In gcm, allow building a model in presence of missing values if an appropriate estimator is selected #1300): skip NaN rows in feature columns only when the underlying sklearn model does not support allow_nan

Run History

2026-04-20 01:12 UTC — Run

• 🧪 Created PR (new): test — add initial test coverage for gcm.fitting_sampling module — 13 tests for fit, fit_causal_model_of_target, draw_samples (Task 9)
• 💬 Commented on In gcm, allow building a model in presence of missing values if an appropriate estimator is selected #1300: explained NaN handling in fitting_sampling.py and provided implementation sketch for sklearn allow_nan detection (Task 2)

2026-04-19 13:04 UTC — Run

• 🔧 Created PR: fix — support multiple treatments in PlaceboTreatmentRefuter + 2 new tests (closes #251) (Task 10)
• 💬 Commented on effect_modifiers #767: explained num_quantiles_to_discretize_cont_cols must go in method_params["init_params"] (Task 2)
• 💬 Commented on Estimate refutation throws error: PicklingError: logger cannot be pickled #953: diagnosed PicklingError root cause (Python < 3.12 loggers) and workarounds (Task 2)
• 💬 Commented on Placebo refuter with multiple treatments #251: described the fix and linked to draft PR

2026-04-19 01:11 UTC — Run

• 🔧 Created PR: test — increase _generate_linear_classification_data() from 100→500 samples to fix flaky auto-assignment test (Task 4: Engineering Investments)
• 💬 Commented on Problems with multi-valued treatment and parameter 'evaluate_effect_strength' #416: linked to PR [Repo Assist] fix: handle list-form treatment_variable in estimate_effect_naive #1461 which fixes the multi-valued treatment bug (Task 2)

2026-04-18 13:04 UTC — Run

• 🔧 Created PR: fix — correct exact_match_cols logic in DistanceMatchingEstimator (closes #814) — 3 bugs fixed
• 🧪 Added tests/causal_estimators/test_distance_matching_estimator.py — 12 new tests (Task 9: Testing Improvements)
• 💬 Commented on backdoor.distance_matching: Problem with exact matching #814: described root causes and fix

2026-04-18 01:07 UTC — Run

• 🔧 Created PR [Repo Assist] feat: auto-select default estimation method when method_name is omitted #1464: feat — auto-select default estimation method when method_name is omitted (closes #365)
• 💬 Commented on CausalModel.estimate_effect - UnboundLocalError: local variable 'identifier_name' referenced before assignment #365, 'fit_estimator' argument in 'estimate_effect' function of 'CausalModel' object #414, Setting the seed for estimate effect methods #418

2026-04-17 13:06 UTC — Run

• 🔧 Created PR [Repo Assist] fix: update second-stage model estimand when pre-instantiated in TwoStageRegression #1462: fix — update _target_estimand when a pre-instantiated second_stage_model is passed to TwoStageRegressionEstimator (closes #1335)
• 🔧 Created PR [Repo Assist] refactor: remove deprecated copy=False from pandas astype() calls #1463 (merged): refactor — remove deprecated copy=False from pandas astype() calls

2026-04-17 UTC — Run

• 🔧 Created PR [Repo Assist] fix: handle list-form treatment_variable in estimate_effect_naive #1461: fix — handle list-form treatment_variable in estimate_effect_naive (closes #416)
• 💬 Commented on Natural Direct Effect estimation does not support multiple mediators #966, estimate_effect() got an unexpected keyword argument 'num_quantiles_to_discretize_cont_cols' #1035, Error in Conditional Effect Estimation in backdoor.linear_regression_estimator #401

2026-04-17 UTC — Run

• 🔧 Created PR [Repo Assist] fix: correct categorical groupby KeyError and pandas FutureWarning in dummy_outcome_refuter #1460: fix — correct categorical groupby KeyError + add observed=True in dummy_outcome_refuter (closes #1316)
• 💬 Commented on The causal structure returned by causal-learn is used in DoWhy. #1365

2026-04-16 01:13 UTC — Run

• 💬 Commented on Replication of DoWhy backdoor.linear_regression with multiple treatments and effect modifiers #1367
• 🔧 Attempted bundled CI update PR (blocked by protected workflow files); issue [Repo Assist] ci: bundle GitHub Actions updates (github-script v9, cache v5) #1458 created (now resolved — Dependabot Bump actions/github-script from 8.0.0 to 9.0.0 #1450 and Bump actions/cache from 4 to 5 #1451 merged)

2026-04-15 13:06 UTC — Run

• 🔧 Created PR [Repo Assist] fix: correct isinstance check in placebo_treatment_refuter for IV estimators (closes #1180) #1457 (merged): fix — correct isinstance check in placebo_treatment_refuter for IV estimators (closes #1180)

2026-04-15 01:11 UTC — Run

• 🔧 Created PR [Repo Assist] fix: correct NDE estimand setup in TwoStageRegressionEstimator (closes #946) #1456 (merged): fix — correct NDE estimand setup in TwoStageRegressionEstimator.__init__ (closes #946)

2026-04-14 13:07 UTC — Run

• 🏷️ Labelled plotting the graph in dowhy_example_effect_of_memberrewards_program.ipynb #493, How to install the GCM module of DoWHY  #513, Are there alternatives to backdoor.linear_regression for multiple continuous treatment estimation? #517, [Refute] Placebo Test #526, Multiclass outcome, binary treatment, and mixed (binary and continuous) confounders #927 with question; Error handling column names in CIT as used by CausalModel.graph_refute method #949 with bug; [MAINT] Refactor any dependencies on independence tests to use optional dependency pywhy-stats #1033 with refactor
• 🔧 Created PR [Repo Assist] fix: use string key for column lookup in conditional_MI (fixes KeyError with multi-char column names) #1455: fix — use string key in conditional_MI (closes #949)

2026-04-14 07:34 UTC — Run

• 🔧 Created PR [Repo Assist] fix: return p_value=1 when refuter simulations have zero variance (closes #807) #1454 (merged): fix — return p_value=1 when std of simulations is 0 (closes #807)
• 🔧 Created PR [Repo Assist] fix: remove invalid [0] subscript on np.std() scalar in kappa auto-inference #1420 (merged): fix — remove invalid [0] subscript on np.std() scalar in kappa auto-inference

2026-04-01 11:15 UTC — Run

• 🔧 Created PRs [Repo Assist] fix: implement asymptotic CI/SE via Delta method for LinearRegressionEstimator with effect modifiers #1432, [Repo Assist] fix: compute proper chi-squared p-value for CMI independence test in GraphRefuter #1431, [Repo Assist] fix: pass encoded effect modifiers to EconML when categorical variables present #1435, [Repo Assist] perf: eliminate redundant set allocations and fix per-column kernel precision #1437, [Repo Assist] fix: correct categorical groupby KeyError and pandas FutureWarning in dummy_outcome_refuter #1460, [Repo Assist] fix: handle list-form treatment_variable in estimate_effect_naive #1461, [Repo Assist] fix: update second-stage model estimand when pre-instantiated in TwoStageRegression #1462, [Repo Assist] refactor: remove deprecated copy=False from pandas astype() calls #1463, [Repo Assist] fix: preserve DataFrame index in one_hot_encode to fix data_subset_refuter bug with categorical columns #1408, [Repo Assist] test: add tests for RandomCommonCause refuter #1413
• 💬 Commented on Add asymptotic confidence intervals for average treatment effect for linear regression with effect modifiers #336, Add progress bar to refutation operations since they take a long time to run #368, Estimate Effect fails with Econml DML estimator #820, p-value NaN in some pathological cases with non-bootstrap method #807, Inconsistency in the placebo_treatment_refuter when using estimate_effect of IV #1180, p-value NaN in some pathological cases with non-bootstrap method #807
• 📝 Created this April 2026 monthly summary [Repo Assist] Monthly Activity 2026-04 #1433

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

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