Add non-finite outcome validation in EfficientDiD.fit()

Reject NaN/Inf values in the outcome column before they can propagate
into Omega*/EIF calculations. Adds a regression test for the validation.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: igerber

diff_diff/efficient_did.py

@@ -217,6 +217,15 @@ def fit(
                 "panel where every unit is observed in every time period."
             )
 
+        # Reject non-finite outcomes (NaN/Inf corrupt Omega*/EIF calculations)
+        non_finite_mask = ~np.isfinite(df[outcome])
+        if non_finite_mask.any():
+            n_bad = int(non_finite_mask.sum())
+            raise ValueError(
+                f"Found {n_bad} non-finite value(s) in outcome column '{outcome}'. "
+                "EfficientDiD requires finite outcomes for all unit-period observations."
+            )
+
         # Reject duplicate (unit, time) rows
         dup_mask = df.duplicated(subset=[unit, time], keep=False)
         if dup_mask.any():

tests/test_efficient_did.py

@@ -321,6 +321,13 @@ def test_pt_post_no_never_treated_raises(self):
         with pytest.raises(ValueError, match="never-treated"):
             EfficientDiD(pt_assumption="post").fit(df, "y", "unit", "time", "first_treat")
 
+    def test_nan_outcome_raises(self):
+        """Non-finite outcomes in a balanced panel should be rejected."""
+        df = _make_simple_panel()
+        df.loc[df.index[0], "y"] = np.nan
+        with pytest.raises(ValueError, match="non-finite"):
+            EfficientDiD().fit(df, "y", "unit", "time", "first_treat")
+
     def test_duplicate_unit_time_raises(self):
         """Duplicate (unit, time) rows should be rejected."""
         df = _make_simple_panel()