Fix issue with MedianCDFQuantileScorer

Before, the scorer was not able to handle numpy object types directly. However, GCM often uses the object dtype to ensure support of mixing categorical and float values. This fixes the handling of object dtypes by explicitly converting them to floats first.

Signed-off-by: Patrick Bloebaum <bloebp@amazon.com>

dowhy/gcm/anomaly_scorers.py

@@ -43,13 +43,13 @@ def fit(self, X: np.ndarray) -> None:
         if (X.ndim == 2 and X.shape[1] > 1) or X.ndim > 2:
             raise ValueError("The MedianCDFQuantileScorer currently only supports one-dimensional data!")
 
-        self._distribution_samples = X.reshape(-1)
+        self._distribution_samples = X.reshape(-1).astype(float)
 
     def score(self, X: np.ndarray) -> np.ndarray:
         if self._distribution_samples is None:
             raise ValueError("Scorer has not been fitted!")
 
-        X = shape_into_2d(X)
+        X = shape_into_2d(X.astype(float))
 
         equal_samples = np.sum(np.isclose(X, self._distribution_samples, rtol=0, atol=0, equal_nan=True), axis=1)
         greater_samples = np.sum(X > self._distribution_samples, axis=1) + equal_samples / 2

tests/gcm/test_anomaly_scorers.py

@@ -29,3 +29,14 @@ def test_given_data_with_nans_when_using_median_quantile_scorer_with_nan_support
     assert scorer.score(np.array([1, 4, 8, np.nan])) == approx(
         [-np.log(2 * 0.5 / 10), -np.log(2 * 3.5 / 10), -np.log(2 * 0.5 / 10), -np.log(2 * 1 / 10)]
     )
+
+
+def test_given_numpy_arrays_with_object_type_when_using_median_quantile_scorer_then_does_not_raise_error():
+    training_data = np.array([1, 2, 3, 4, 5, 6, 7, 8, np.nan, np.nan], dtype=object)
+
+    scorer = RescaledMedianCDFQuantileScorer()
+    scorer.fit(training_data)
+
+    assert scorer.score(np.array([1, 4, 8, np.nan], dtype=object)) == approx(
+        [-np.log(2 * 0.5 / 10), -np.log(2 * 3.5 / 10), -np.log(2 * 0.5 / 10), -np.log(2 * 1 / 10)]
+    )