fix: clear the stale type: ignore comments and gate warn_unused_ignores

33 line-level ignores across 10 modules no longer suppress anything, per
mypy --warn-unused-ignores in both the CausalPy and CausalPy-pymc6 envs
(identical unused set in both). Deleted them; narrowed the one partially
unused case in instrumental_variable.py to the code still needed
(call-arg, not union-attr).

warn_unused_ignores is now on in [tool.mypy] so a future fix that drops
the underlying error also has to drop its now-stale ignore.

Author: anevolbap

causalpy/checks/mccrary.py

@@ -91,7 +91,7 @@ def run(
         rd = experiment
         threshold = rd.treatment_threshold  # type: ignore[attr-defined]
         running_var = rd.running_variable_name  # type: ignore[attr-defined]
-        data = rd.data  # type: ignore[attr-defined]
+        data = rd.data
 
         x = data[running_var].values
         below = x[x < threshold]

causalpy/checks/placebo_in_time.py

@@ -528,7 +528,7 @@ def _compute_intervention_length(self, experiment: BaseExperiment) -> Any:
             return self.intervention_length
 
         treatment_time = experiment.treatment_time  # type: ignore[attr-defined]
-        data = experiment.data  # type: ignore[attr-defined]
+        data = experiment.data
 
         treatment_end = getattr(experiment, "treatment_end_time", None)
         if treatment_end is not None:
@@ -1042,13 +1042,9 @@ def _draw_expected_effect_samples(
             raise ValueError("expected_effect_prior is not set.")
         if hasattr(prior, "rvs"):
             if self.random_seed is None:
-                return np.asarray(prior.rvs(n))  # type: ignore[union-attr]
+                return np.asarray(prior.rvs(n))
             if self._rvs_accepts_random_state(prior):
-                return np.asarray(
-                    prior.rvs(  # type: ignore[union-attr]
-                        n, random_state=self._rng_for_stage(0)
-                    )
-                )
+                return np.asarray(prior.rvs(n, random_state=self._rng_for_stage(0)))
 
             prior_type = f"{type(prior).__module__}.{type(prior).__qualname__}"
             if unseeded_custom_priors is not None:
@@ -1065,7 +1061,7 @@ def _draw_expected_effect_samples(
                 "marks its type as unseeded.",
                 stacklevel=2,
             )
-            return np.asarray(prior.rvs(n))  # type: ignore[union-attr]
+            return np.asarray(prior.rvs(n))
         raise TypeError(
             f"expected_effect_prior must have an .rvs(n) method, got "
             f"{type(prior).__name__}."
@@ -1147,7 +1143,7 @@ def run(
         unseeded_custom_priors: list[dict[str, str]] = []
         factory = self._get_factory(context)
         treatment_time = experiment.treatment_time  # type: ignore[attr-defined]
-        data = experiment.data  # type: ignore[attr-defined]
+        data = experiment.data
         intervention_length = self._compute_intervention_length(experiment)
         required_pre_period_rows = self._get_intervention_window_observation_count(
             data, treatment_time, intervention_length

causalpy/experiments/interrupted_time_series.py

@@ -274,13 +274,13 @@ def input_validation(
             # Validate treatment_end_time > treatment_time
             # Type check: we've already validated both match the index type, so they're compatible
             # NOTE: Both treatment_time and treatment_end_time are INCLUSIVE (>=) in their respective periods
-            if treatment_end_time <= treatment_time:  # type: ignore[operator]
+            if treatment_end_time <= treatment_time:
                 raise ValueError(
                     f"treatment_end_time ({treatment_end_time}) must be greater than treatment_time ({treatment_time})"
                 )
             # Validate treatment_end_time is within data range
             # NOTE: treatment_end_time is INCLUSIVE, so it can equal data.index.max()
-            if treatment_end_time > data.index.max():  # type: ignore[operator]
+            if treatment_end_time > data.index.max():
                 raise ValueError(
                     f"treatment_end_time ({treatment_end_time}) is beyond the data range (max: {data.index.max()})"
                 )

causalpy/experiments/inverse_propensity_weighting.py

@@ -448,9 +448,9 @@ def _compute_ate_overlap(
             weighted_outcome_trt,
             n_ntrt,
             n_trt,
-        ) = self.make_overlap_adjustments(ps)  # type: ignore[assignment]
-        ntrt = np.sum(weighted_outcome_ntrt) / np.sum(n_ntrt)  # type: ignore[arg-type]
-        trt = np.sum(weighted_outcome_trt) / np.sum(n_trt)  # type: ignore[arg-type]
+        ) = self.make_overlap_adjustments(ps)
+        ntrt = np.sum(weighted_outcome_ntrt) / np.sum(n_ntrt)
+        trt = np.sum(weighted_outcome_trt) / np.sum(n_trt)
         ate = trt - ntrt
         return ate, trt, ntrt
 
@@ -478,7 +478,7 @@ def _compute_ate_doubly_robust(
             weighted_outcome_trt,
             _n_ntrt,
             _n_trt,
-        ) = self.make_doubly_robust_adjustment(ps)  # type: ignore[assignment]
+        ) = self.make_doubly_robust_adjustment(ps)
         trt = np.mean(weighted_outcome_trt)
         ntrt = np.mean(weighted_outcome_ntrt)
         ate = trt - ntrt

causalpy/experiments/panel_regression.py

@@ -854,11 +854,11 @@ def plot_trajectories(
         if units is not None:
             selected_units = units
         elif self.n_units <= n_sample:
-            selected_units = all_units  # type: ignore[assignment]
+            selected_units = all_units
         else:
             if select == "random":
                 rng = np.random.default_rng(42)
-                selected_units = rng.choice(all_units, size=n_sample, replace=False)  # type: ignore[assignment]
+                selected_units = rng.choice(all_units, size=n_sample, replace=False)
             elif select == "extreme":
                 # Select units with the largest and smallest mean outcomes
                 unit_means = self.data.groupby(self.unit_fe_variable, observed=True)[
@@ -953,7 +953,7 @@ def plot_trajectories(
                 # OLS: get fitted values for this unit
                 y_fitted = np.squeeze(self.model.predict(self.design["X"]))[
                     sorted_obs_indices
-                ]  # type: ignore[union-attr]
+                ]
                 ax.plot(
                     sorted_time_vals,
                     y_fitted,

causalpy/experiments/staggered_did.py

@@ -589,8 +589,8 @@ def _aggregate_effects_bayesian(
         ).groups
         att_gt_rows: list[dict] = []
         for key, idx in gt_groups.items():
-            g_val = key[0]  # type: ignore[index]
-            t_val = key[1]  # type: ignore[index]
+            g_val = key[0]
+            t_val = key[1]
             # Find positions in treated_indices
             positions = [np.where(treated_indices == i)[0][0] for i in idx]
             tau_gt = tau_draws_treated[:, :, positions].mean(axis=2)
@@ -1196,8 +1196,8 @@ def _get_group_time_placebo_data_bayesian(self) -> pd.DataFrame:
             ["G", self.time_variable_name], observed=True
         ).groups
         for key, idx in gt_groups.items():
-            g_val = key[0]  # type: ignore[index]
-            t_val = key[1]  # type: ignore[index]
+            g_val = key[0]
+            t_val = key[1]
             positions = [np.where(self.data.index == i)[0][0] for i in idx]
             tau_gt = tau_draws_all[:, :, positions].mean(axis=2)
             att_gt_rows.append(

causalpy/formula_utils.py

@@ -96,7 +96,7 @@ def _normalize_patsy_data(data: pd.DataFrame) -> pd.DataFrame:
         # ``isetitem`` accepts a Series at runtime; pandas-stubs omits it.
         normalized_data.isetitem(
             position,
-            pd.Series(values, index=data.index, dtype=object),  # type: ignore[arg-type]
+            pd.Series(values, index=data.index, dtype=object),
         )
     return normalized_data
 

causalpy/pymc_models.py

@@ -2119,7 +2119,7 @@ def build_model(
 
         # Add coeffs coordinate if we have exogenous variables
         if self._exog_var_names:
-            model_coords["coeffs"] = self._exog_var_names  # type: ignore[assignment]
+            model_coords["coeffs"] = self._exog_var_names
 
         with self:
             self.add_coords(model_coords)

causalpy/transforms.py

@@ -102,12 +102,12 @@ def memorize_chunk(
         if self._is_datetime_like(x):
             self._is_datetime = True
             x_dt = pd.to_datetime(x)
-            x_min: pd.Timestamp = pd.Timestamp(x_dt.min())  # type: ignore[assignment]
+            x_min: pd.Timestamp = pd.Timestamp(x_dt.min())
             if self._origin is None:
                 self._origin = x_min
             else:
                 # Handle chunked data - keep the overall minimum
-                self._origin = min(self._origin, x_min)  # type: ignore[assignment]
+                self._origin = min(self._origin, x_min)
 
     def memorize_finish(self) -> None:
         """Called after all chunks processed - finalize state."""
@@ -149,7 +149,7 @@ def transform(
             t_numeric = (threshold_dt - self._origin).total_seconds() / (24 * 3600)
         else:
             x_numeric = np.asarray(x, dtype=float)
-            t_numeric = float(threshold)  # type: ignore[arg-type]
+            t_numeric = float(threshold)
 
         return (x_numeric >= t_numeric).astype(float)
 
@@ -161,7 +161,7 @@ def _parse_threshold(
             return threshold
         else:
             # Assume it's something pandas can convert (str or numeric)
-            return pd.Timestamp(threshold)  # type: ignore[arg-type, return-value]
+            return pd.Timestamp(threshold)
 
 
 class RampTransform:
@@ -222,11 +222,11 @@ def memorize_chunk(
         if self._is_datetime_like(x):
             self._is_datetime = True
             x_dt = pd.to_datetime(x)
-            x_min: pd.Timestamp = pd.Timestamp(x_dt.min())  # type: ignore[assignment]
+            x_min: pd.Timestamp = pd.Timestamp(x_dt.min())
             if self._origin is None:
                 self._origin = x_min
             else:
-                self._origin = min(self._origin, x_min)  # type: ignore[assignment]
+                self._origin = min(self._origin, x_min)
 
     def memorize_finish(self) -> None:
         """Called after all chunks processed."""
@@ -269,7 +269,7 @@ def transform(
             t_numeric = (threshold_dt - self._origin).total_seconds() / (24 * 3600)
         else:
             x_numeric = np.asarray(x, dtype=float)
-            t_numeric = float(threshold)  # type: ignore[arg-type]
+            t_numeric = float(threshold)
 
         return np.maximum(0.0, x_numeric - t_numeric)
 
@@ -281,7 +281,7 @@ def _parse_threshold(
             return threshold
         else:
             # Assume it's something pandas can convert (str or numeric)
-            return pd.Timestamp(threshold)  # type: ignore[arg-type, return-value]
+            return pd.Timestamp(threshold)
 
 
 class ElapsedDaysTransform:
@@ -326,9 +326,9 @@ def transform(self, x: Any) -> np.ndarray:
 
 
 # Create callable stateful transforms for use in formulas
-step = patsy.stateful_transform(StepTransform)  # type: ignore[attr-defined]
-ramp = patsy.stateful_transform(RampTransform)  # type: ignore[attr-defined]
-elapsed = patsy.stateful_transform(ElapsedDaysTransform)  # type: ignore[attr-defined]
+step = patsy.stateful_transform(StepTransform)
+ramp = patsy.stateful_transform(RampTransform)
+elapsed = patsy.stateful_transform(ElapsedDaysTransform)
 
 __all__ = [
     "step",

pyproject.toml

@@ -235,7 +235,8 @@ ignore_missing_imports = true
 warn_unused_configs = true
 # The allowlist is empty: the package type-checks clean. What could not be expressed in the type system is a handful of line-level ``type: ignore`` comments, each with the reasoning next to it.
 #
-# ``warn_unused_ignores`` would make those comments self-cleaning, and is worth turning on, but it currently reports 33 pre-existing unused ignores across 10 modules. Clearing those is its own change, not a rider on this one.
+# ``warn_unused_ignores`` keeps those comments self-cleaning: a fix that removes the underlying error now fails the check unless the now-unneeded ignore is deleted with it.
+warn_unused_ignores = true
 
 [tool.marimo.runtime]
 watcher_on_save = "autorun"