scikit-learn · OmarManzoor · Jul 8, 2024 · Jun 14, 2024 · Jun 14, 2024 · Jun 17, 2024
diff --git a/sklearn/metrics/_regression.py b/sklearn/metrics/_regression.py
@@ -1301,12 +1301,14 @@ def _mean_tweedie_deviance(y_true, y_pred, sample_weight, power):
    """Mean Tweedie deviance regression loss."""
    xp, _ = get_namespace(y_true, y_pred)
    p = power
+    zero = xp.asarray(0, dtype=y_true.dtype)
    if p < 0:
        # 'Extreme stable', y any real number, y_pred > 0
        dev = 2 * (
-            xp.pow(xp.where(y_true > 0, y_true, 0), 2 - p) / ((1 - p) * (2 - p))
-            - y_true * xp.pow(y_pred, 1 - p) / (1 - p)
-            + xp.pow(y_pred, 2 - p) / (2 - p)
+            xp.pow(xp.where(y_true > 0, y_true, zero), xp.asarray(2 - p))
+            / ((1 - p) * (2 - p))
+            - y_true * xp.pow(y_pred, xp.asarray(1 - p)) / (1 - p)
+            + xp.pow(y_pred, xp.asarray(2 - p)) / (2 - p)
        )
    elif p == 0:
        # Normal distribution, y and y_pred any real number
@@ -1319,9 +1321,9 @@ def _mean_tweedie_deviance(y_true, y_pred, sample_weight, power):
        dev = 2 * (xp.log(y_pred / y_true) + y_true / y_pred - 1)
    else:
        dev = 2 * (
-            xp.pow(y_true, 2 - p) / ((1 - p) * (2 - p))
-            - y_true * xp.pow(y_pred, 1 - p) / (1 - p)
-            + xp.pow(y_pred, 2 - p) / (2 - p)
+            xp.pow(y_true, xp.asarray(2 - p)) / ((1 - p) * (2 - p))
+            - y_true * xp.pow(y_pred, xp.asarray(1 - p)) / (1 - p)
+            + xp.pow(y_pred, xp.asarray(2 - p)) / (2 - p)
        )
    return float(_average(dev, weights=sample_weight))

@@ -1401,14 +1403,14 @@ def mean_tweedie_deviance(y_true, y_pred, *, sample_weight=None, power=0):
    message = f"Mean Tweedie deviance error with power={power} can only be used on "
    if power < 0:
        # 'Extreme stable', y any real number, y_pred > 0
-        if (y_pred <= 0).any():
+        if xp.any(y_pred <= 0):
            raise ValueError(message + "strictly positive y_pred.")
    elif power == 0:
        # Normal, y and y_pred can be any real number
        pass
    elif 1 <= power < 2:
        # Poisson and compound Poisson distribution, y >= 0, y_pred > 0
-        if (y_true < 0).any() or (y_pred <= 0).any():
+        if xp.any(y_true < 0) or xp.any(y_pred <= 0):
            raise ValueError(message + "non-negative y and strictly positive y_pred.")
    elif power >= 2:
        # Gamma and Extreme stable distribution, y and y_pred > 0

diff --git a/sklearn/metrics/tests/test_common.py b/sklearn/metrics/tests/test_common.py
@@ -1942,6 +1942,8 @@ def check_array_api_metric_pairwise(metric, array_namespace, device, dtype_name)
        check_array_api_multiclass_classification_metric,
    ],
    mean_tweedie_deviance: [check_array_api_regression_metric],
+    partial(mean_tweedie_deviance, power=-0.5): [check_array_api_regression_metric],
+    partial(mean_tweedie_deviance, power=1.5): [check_array_api_regression_metric],
    r2_score: [
        check_array_api_regression_metric,
        check_array_api_regression_metric_multioutput,