scikit-learn
diff --git a/‎sklearn/ensemble/_hist_gradient_boosting/tests/test_gradient_boosting.py
Lines changed: 36 additions & 0 deletions b/‎sklearn/ensemble/_hist_gradient_boosting/tests/test_gradient_boosting.py
Lines changed: 36 additions & 0 deletions
@@ -7,6 +7,7 @@
     HalfMultinomialLoss,
     HalfPoissonLoss,
     HalfSquaredError,
+    PinballLoss,
 )
 from sklearn.datasets import make_classification, make_regression
 from sklearn.datasets import make_low_rank_matrix
@@ -35,6 +36,7 @@
     "squared_error": HalfSquaredError,
     "absolute_error&quo
10000
t;: AbsoluteError,
     "poisson": HalfPoissonLoss,
+    "quantile": PinballLoss,
     "binary_crossentropy": HalfBinomialLoss,
     "categorical_crossentropy": HalfMultinomialLoss,
 }
@@ -249,6 +251,40 @@ def test_absolute_error_sample_weight():
     gbdt.fit(X, y, sample_weight=sample_weight)
 
 
+@pytest.mark.parametrize("quantile", [0.2, 0.5, 0.8])
+def test_asymmetric_error(quantile):
+    """Test quantile regression for asymmetric distributed targets."""
+    n_samples = 10_000
+    rng = np.random.RandomState(42)
+    # take care that X @ coef + intercept > 0
+    X = np.concatenate(
+        (
+            np.abs(rng.randn(n_samples)[:, None]),
+            -rng.randint(2, size=(n_samples, 1)),
+        ),
+        axis=1,
+    )
+    intercept = 1.23
+    coef = np.array([0.5, -2])
+    # For an exponential distribution with rate lambda, e.g. exp(-lambda * x),
+    # the quantile at level q is:
+    #   quantile(q) = - log(1 - q) / lambda
+    #   scale = 1/lambda = -quantile(q) / log(1-q)
+    y = rng.exponential(
+        scale=-(X @ coef + intercept) / np.log(1 - quantile), size=n_samples
+    )
+    model = HistGradientBoostingRegressor(
+        loss="quantile",
+        loss_param=quantile,
+    ).fit(X, y)
+    assert_allclose(np.mean(model.predict(X) > y), quantile, rtol=1e-2)
+
+    loss_true_quantile = model._loss(y, X @ coef + intercept)
+    loss_pred_quantile = model._loss(y, model.predict(X))
+    # we are overfitting
+    assert loss_pred_quantile <= loss_true_quantile
+
+
 @pytest.mark.parametrize("y", [([1.0, -2.0, 0.0]), ([0.0, 0.0, 0.0])])
 def test_poisson_y_positive(y):
     # Test that ValueError is raised if either one y_i < 0 or sum(y_i) <= 0.