[Repo Assist] fix: use string key for column lookup in conditional_MI (fixes KeyError with multi-char column names)

dowhy/utils/cit.py

@@ -136,11 +136,13 @@ def conditional_MI(data=None, x=None, y=None, z=None):
                 = H(X,Z) - H(Z) - H(X,Y,Z) + H(Y,Z)
                 = H(X,Z) + H(Y,Z) - H(X,Y,Z) - H(Z)
     :param data : dataset
-    :param x,y,z : column names from dataset
+    :param x : column name (str) in dataset
+    :param y : column name (str) in dataset
+    :param z : list of column names in dataset
     :returns : conditional mutual information between X and Y given Z
     """
-    X = data[list(x)].astype(int)
-    Y = data[list(y)].astype(int)
+    X = data[x].astype(int)
+    Y = data[y].astype(int)
     t = list(z)
     Z = data[t].astype(int)
     Z = Z.values.tolist()

tests/causal_refuters/test_graph_refuter.py

@@ -0,0 +1,58 @@
+import numpy as np
+import pandas as pd
+
+from dowhy.causal_refuters.graph_refuter import GraphRefuter
+
+
+class TestGraphRefuter:
+    """Tests for the GraphRefuter class, including regression tests for reported bugs."""
+
+    def test_conditional_mi_multi_char_column_names(self):
+        """Regression test for #949: KeyError when column names have multiple characters.
+
+        Previously, `conditional_MI` called `list(x)` on a string column name,
+        iterating over individual characters instead of treating the name as a key.
+        """
+        rng = np.random.default_rng(42)
+        n = 200
+        data = pd.DataFrame(
+            {
+                "Foo": rng.integers(0, 3, size=n),
+                "Bar": rng.integers(0, 3, size=n),
+                "treatment": rng.integers(0, 2, size=n),
+                "outcome": rng.integers(0, 2, size=n),
+            }
+        )
+        refuter = GraphRefuter(data=data)
+        # Should not raise KeyError
+        refuter.conditional_mutual_information(x="Foo", y="Bar", z=["treatment"])
+
+    def test_conditional_mi_single_char_column_names(self):
+        """Single-character column names should still work correctly."""
+        rng = np.random.default_rng(42)
+        n = 200
+        data = pd.DataFrame(
+            {
+                "A": rng.integers(0, 3, size=n),
+                "B": rng.integers(0, 3, size=n),
+                "C": rng.integers(0, 2, size=n),
+            }
+        )
+        refuter = GraphRefuter(data=data)
+        refuter.conditional_mutual_information(x="A", y="B", z=["C"])
+
+    def test_graph_refuter_with_multi_char_columns(self):
+        """End-to-end test: refute_model should work with multi-character column names."""
+        rng = np.random.default_rng(0)
+        n = 300
+        foo = rng.integers(0, 3, size=n)
+        bar = rng.integers(0, 3, size=n)
+        treatment = (foo + rng.integers(0, 2, size=n)) % 2
+        outcome = (bar + treatment + rng.integers(0, 2, size=n)) % 2
+
+        data = pd.DataFrame({"Foo": foo, "Bar": bar, "treatment": treatment, "outcome": outcome})
+        refuter = GraphRefuter(data=data)
+        # Foo _|_ treatment | []  (no conditioning set)
+        independence_constraints = [("Foo", "treatment", [])]
+        result = refuter.refute_model(independence_constraints=independence_constraints)
+        assert result is not None