[BUG] Fix DummyProbaRegressor._predict_var returning std dev instead of variance

skpro/regression/dummy.py

@@ -78,6 +78,7 @@ def _fit(self, X, y):
         self._y_columns = y.columns
         self._mu = np.mean(y.values)
         self._sigma = np.std(y.values)
+        self._var = np.var(y.values)
         if self.strategy == "empirical":
             self.distribution_ = Empirical(y)
         if self.strategy == "normal":
@@ -126,7 +127,7 @@ def _predict_var(self, X):
         X_ind = X.index
         X_n_rows = X.shape[0]
         y_pred = pd.DataFrame(
-            np.ones(X_n_rows) * self._sigma, index=X_ind, columns=self._y_columns
+            np.ones(X_n_rows) * self._var, index=X_ind, columns=self._y_columns
         )
         return y_pred
 

skpro/regression/tests/test_dummy.py

@@ -0,0 +1,123 @@
+"""Tests for DummyProbaRegressor."""
+# copyright: skpro developers, BSD-3-Clause License (see LICENSE file)
+
+import numpy as np
+import pandas as pd
+import pytest
+from sklearn.model_selection import train_test_split
+
+
+@pytest.fixture
+def diabetes_data():
+    """Return a small diabetes dataset split for train/test."""
+    from sklearn.datasets import load_diabetes
+
+    X, y = load_diabetes(return_X_y=True, as_frame=True)
+    X = X.iloc[:50]
+    y = pd.DataFrame(y.iloc[:50])
+    return train_test_split(X, y, random_state=42)
+
+
+class TestDummyProbaRegressorPredictVar:
+    """Tests that predict_var returns variance (sigma^2), not std dev (sigma)."""
+
+    def test_predict_var_equals_training_variance_empirical(self, diabetes_data):
+        """predict_var must equal np.var(y_train) for 'empirical' strategy."""
+        from skpro.regression.dummy import DummyProbaRegressor
+
+        X_train, X_test, y_train, _ = diabetes_data
+
+        reg = DummyProbaRegressor(strategy="empirical")
+        reg.fit(X_train, y_train)
+
+        y_pred_var = reg.predict_var(X_test)
+
+        expected_var = np.var(y_train.values)
+
+        # Every row of predict_var should equal the training variance
+        assert isinstance(y_pred_var, pd.DataFrame)
+        assert y_pred_var.shape[0] == X_test.shape[0]
+        np.testing.assert_allclose(
+            y_pred_var.values,
+            expected_var,
+            rtol=1e-5,
+            err_msg=(
+                "predict_var returned std dev instead of variance. "
+                f"Got {y_pred_var.values[0, 0]:.6f}, "
+                f"expected variance {expected_var:.6f}, "
+                f"std dev would be {np.std(y_train.values):.6f}."
+            ),
+        )
+
+    def test_predict_var_equals_training_variance_normal(self, diabetes_data):
+        """predict_var must equal np.var(y_train) for 'normal' strategy."""
+        from skpro.regression.dummy import DummyProbaRegressor
+
+        X_train, X_test, y_train, _ = diabetes_data
+
+        reg = DummyProbaRegressor(strategy="normal")
+        reg.fit(X_train, y_train)
+
+        y_pred_var = reg.predict_var(X_test)
+
+        expected_var = np.var(y_train.values)
+
+        assert isinstance(y_pred_var, pd.DataFrame)
+        np.testing.assert_allclose(
+            y_pred_var.values,
+            expected_var,
+            rtol=1e-5,
+            err_msg=(
+                "predict_var returned std dev instead of variance. "
+                f"Got {y_pred_var.values[0, 0]:.6f}, "
+                f"expected variance {expected_var:.6f}."
+            ),
+        )
+
+    def test_predict_var_not_equal_std(self, diabetes_data):
+        """Regression test: predict_var must NOT equal std dev of training labels."""
+        from skpro.regression.dummy import DummyProbaRegressor
+
+        X_train, X_test, y_train, _ = diabetes_data
+
+        reg = DummyProbaRegressor(strategy="empirical")
+        reg.fit(X_train, y_train)
+
+        y_pred_var = reg.predict_var(X_test)
+
+        std_dev = np.std(y_train.values)
+        variance = np.var(y_train.values)
+
+        # std dev and variance differ substantially; the prediction should match var
+        assert not np.allclose(y_pred_var.values, std_dev, rtol=1e-3), (
+            "predict_var is returning std dev instead of variance. "
+            f"std_dev={std_dev:.4f}, variance={variance:.4f}."
+        )
+
+    def test_predict_var_is_nonnegative(self, diabetes_data):
+        """Variance predictions must be non-negative."""
+        from skpro.regression.dummy import DummyProbaRegressor
+
+        X_train, X_test, y_train, _ = diabetes_data
+
+        for strategy in ["empirical", "normal"]:
+            reg = DummyProbaRegressor(strategy=strategy)
+            reg.fit(X_train, y_train)
+            y_pred_var = reg.predict_var(X_test)
+            assert (y_pred_var.values >= 0).all(), (
+                f"predict_var returned negative values for strategy='{strategy}'"
+            )
+
+    def test_predict_var_index_and_columns(self, diabetes_data):
+        """predict_var must have same index as X_test and same columns as y_train."""
+        from skpro.regression.dummy import DummyProbaRegressor
+
+        X_train, X_test, y_train, _ = diabetes_data
+
+        reg = DummyProbaRegressor(strategy="normal")
+        reg.fit(X_train, y_train)
+
+        y_pred_var = reg.predict_var(X_test)
+
+        assert (y_pred_var.index == X_test.index).all()
+        assert (y_pred_var.columns == y_train.columns).all()