From 4481be6c15d4e4a2444c6d9ca62b42fc8cfc7597 Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Thu, 2 Nov 2023 19:27:01 +0100
Subject: [PATCH 01/29] converted mae to array api

---
 sklearn/metrics/_regression.py        | 11 +++---
 sklearn/metrics/tests/test_common.py  | 19 +++++++++++
 sklearn/utils/_array_api.py           | 49 +++++++++++++++++++++++++++
 sklearn/utils/tests/test_array_api.py | 32 +++++++++++++++++
 4 files changed, 107 insertions(+), 4 deletions(-)

diff --git a/sklearn/metrics/_regression.py b/sklearn/metrics/_regression.py
index 0259a3f41620c..80a456f2e1103 100644
--- a/sklearn/metrics/_regression.py
+++ b/sklearn/metrics/_regression.py
@@ -34,6 +34,7 @@
 from scipy.special import xlogy
 
 from ..exceptions import UndefinedMetricWarning
+from ..utils._array_api import _average, get_namespace
 from ..utils._param_validation import Hidden, Interval, StrOptions, validate_params
 from ..utils.stats import _weighted_percentile
 from ..utils.validation import (
@@ -99,15 +100,16 @@ def _check_reg_targets(y_true, y_pred, multioutput, dtype="numeric"):
         just the corresponding argument if ``multioutput`` is a
         correct keyword.
     """
+    xp, _ = get_namespace(y_true, y_pred)
     check_consistent_length(y_true, y_pred)
     y_true = check_array(y_true, ensure_2d=False, dtype=dtype)
     y_pred = check_array(y_pred, ensure_2d=False, dtype=dtype)
 
     if y_true.ndim == 1:
-        y_true = y_true.reshape((-1, 1))
+        y_true = xp.reshape(y_true, (-1, 1))
 
     if y_pred.ndim == 1:
-        y_pred = y_pred.reshape((-1, 1))
+        y_pred = xp.reshape(y_pred, (-1, 1))
 
     if y_true.shape[1] != y_pred.shape[1]:
         raise ValueError(
@@ -204,11 +206,12 @@ def mean_absolute_error(
     >>> mean_absolute_error(y_true, y_pred, multioutput=[0.3, 0.7])
     0.85...
     """
+    xp, _ = get_namespace(y_true, y_pred)
     y_type, y_true, y_pred, multioutput = _check_reg_targets(
         y_true, y_pred, multioutput
     )
     check_consistent_length(y_true, y_pred, sample_weight)
-    output_errors = np.average(np.abs(y_pred - y_true), weights=sample_weight, axis=0)
+    output_errors = _average(xp.abs(y_pred - y_true), weights=sample_weight, axis=0)
     if isinstance(multioutput, str):
         if multioutput == "raw_values":
             return output_errors
@@ -216,7 +219,7 @@ def mean_absolute_error(
             # pass None as weights to np.average: uniform mean
             multioutput = None
 
-    return np.average(output_errors, weights=multioutput)
+    return _average(output_errors, weights=multioutput)
 
 
 @validate_params(
diff --git a/sklearn/metrics/tests/test_common.py b/sklearn/metrics/tests/test_common.py
index af652d1c90b41..a9a10fb1faed5 100644
--- a/sklearn/metrics/tests/test_common.py
+++ b/sklearn/metrics/tests/test_common.py
@@ -1793,6 +1793,24 @@ def check_array_api_multiclass_classification_metric(
         )
 
 
+def check_array_api_regression_metric(metric, array_namespace, device, dtype):
+    y_true_np = np.array([3, -0.5, 2, 7])
+    y_pred_np = np.array([2.5, 0.0, 2, 8])
+    check_array_api_metric(
+        metric, array_namespace, device, dtype, y_true_np=y_true_np, y_pred_np=y_pred_np
+    )
+    if "sample_weight" in signature(metric).parameters:
+        check_array_api_metric(
+            metric,
+            array_namespace,
+            device,
+            dtype,
+            y_true_np=y_true_np,
+            y_pred_np=y_pred_np,
+            sample_weight=np.array([0.0, 0.1, 2.0, 1.0]),
+        )
+
+
 metric_checkers = {
     accuracy_score: [
         check_array_api_binary_classification_metric,
@@ -1802,6 +1820,7 @@ def check_array_api_multiclass_classification_metric(
         check_array_api_binary_classification_metric,
         check_array_api_multiclass_classification_metric,
     ],
+    mean_absolute_error: [check_array_api_regression_metric],
 }
 
 
diff --git a/sklearn/utils/_array_api.py b/sklearn/utils/_array_api.py
index 24534faa931e8..64a61fd3cdd5f 100644
--- a/sklearn/utils/_array_api.py
+++ b/sklearn/utils/_array_api.py
@@ -595,3 +595,52 @@ def _estimator_with_converted_arrays(estimator, converter):
 def _atol_for_type(dtype):
     """Return the absolute tolerance for a given dtype."""
     return numpy.finfo(dtype).eps * 100
+
+
+def _average(X, axis=None, weights=None):
+    """Compute the weighted average along the specified axis.
+
+    This function is a port of numpy.average that supports the Array API.
+
+    Please see the original documentation for more details:
+    https://numpy.org/doc/stable/reference/generated/numpy.average.html
+
+    Parameters
+    ----------
+    X : array object
+        Array containing data to be averaged.
+    axis : None or int or tuple of ints, optional
+        Axis or axes along which to average `X`.
+    weights : array_like, optional
+        An array of weights associated with the values in `X`. Each value in
+        `X` contributes to the average according to its associated weight.
+
+    Returns
+    -------
+    retval : array object
+        Return the average along the specified axis.
+    """
+    xp, _ = get_namespace(X)
+
+    if _is_numpy_namespace(xp):
+        return xp.asarray(numpy.average(X, axis=axis, weights=weights))
+
+    if weights is None:
+        return xp.mean(X, axis=axis)
+
+    weights = xp.asarray(weights, device=device(X))
+    if X.shape != weights.shape:
+        if axis is None:
+            raise TypeError(
+                "Axis must be specified when shapes of a and weights differ."
+            )
+        if weights.ndim != 1:
+            raise TypeError("1D weights expected when shapes of a and weights differ.")
+        if weights.shape[0] != X.shape[axis]:
+            raise ValueError("Length of weights not compatible with specified axis.")
+        weights = xp.broadcast_to(weights, (X.ndim - 1) * (1,) + weights.shape)
+        weights = xp.swapaxes(weights, -1, axis)
+    weights_sum = xp.sum(weights, axis=axis)
+    if xp.any(weights_sum == 0):
+        raise ZeroDivisionError("Weights sum to zero, can't be normalized")
+    return xp.sum(xp.multiply(X, weights), axis=axis) / xp.sum(weights, axis=axis)
diff --git a/sklearn/utils/tests/test_array_api.py b/sklearn/utils/tests/test_array_api.py
index 866fd0e1d56f3..161b2786dafdc 100644
--- a/sklearn/utils/tests/test_array_api.py
+++ b/sklearn/utils/tests/test_array_api.py
@@ -10,6 +10,7 @@
     _ArrayAPIWrapper,
     _asarray_with_order,
     _atol_for_type,
+    _average,
     _convert_to_numpy,
     _estimator_with_converted_arrays,
     _nanmax,
@@ -371,3 +372,34 @@ def test_get_namespace_array_api_isdtype(wrapper):
 
     with pytest.raises(ValueError, match="Unrecognized data type"):
         assert xp.isdtype(xp.int16, "unknown")
+
+
+@skip_if_array_api_compat_not_configured
+@pytest.mark.parametrize(
+    "library", ["numpy", "numpy.array_api", "cupy", "cupy.array_api", "torch"]
+)
+@pytest.mark.parametrize(
+    "X,weights,axis,expected",
+    [
+        [[0.0, 1.0, 2.0, 3.0], None, 0, 1.5],
+        [[0.0, 1.0, 2.0, 3.0], [0.0, 1.0, 1.0, 2.0], 0, 2.25],
+        [[[0.0, 1.0, 2.0, 3.0], [4.0, 5.0, 6.0, 7.0]], None, None, 3.5],
+        [[[0.0, 1.0, 2.0, 3.0], [4.0, 5.0, 6.0, 7.0]], None, 1, [1.5, 5.5]],
+        [
+            [[0.0, 1.0, 2.0, 3.0], [4.0, 5.0, 6.0, 7.0]],
+            [0.0, 1.0, 1.0, 2.0],
+            1,
+            [2.25, 6.25],
+        ],
+    ],
+)
+def test__average(library, X, axis, weights, expected):
+    xp = pytest.importorskip(library)
+
+    if isinstance(expected, list):
+        expected = xp.asarray(expected)
+
+    with config_context(array_api_dispatch=True):
+        result = _average(xp.asarray(X), axis=axis, weights=weights)
+
+    assert_allclose(result, expected)

From 8b3dc8dba7826d582edf5ebb2ab882c89cd29616 Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Mon, 6 Nov 2023 23:27:46 +0100
Subject: [PATCH 02/29] fixes for MPS device

---
 sklearn/metrics/tests/test_common.py | 11 ++++++++---
 sklearn/utils/_array_api.py          |  8 +++++---
 2 files changed, 13 insertions(+), 6 deletions(-)

diff --git a/sklearn/metrics/tests/test_common.py b/sklearn/metrics/tests/test_common.py
index a9a10fb1faed5..0daf890dd2f65 100644
--- a/sklearn/metrics/tests/test_common.py
+++ b/sklearn/metrics/tests/test_common.py
@@ -1743,9 +1743,14 @@ def check_array_api_metric(
 
     with config_context(array_api_dispatch=True):
         if sample_weight is not None:
-            sample_weight = xp.asarray(sample_weight, device=device)
+            sample_weight = xp.asarray(sample_weight.astype(dtype), device=device)
         metric_xp = metric(y_true_xp, y_pred_xp, sample_weight=sample_weight)
 
+        if not isinstance(metric_xp, float):
+            # if the result is not a scalar, the array has to be in the CPU
+            # before transforming it to numpy
+            metric_xp = xp.asarray(metric_xp, device="cpu")
+
         assert_allclose(
             metric_xp,
             metric_np,
@@ -1794,8 +1799,8 @@ def check_array_api_multiclass_classification_metric(
 
 
 def check_array_api_regression_metric(metric, array_namespace, device, dtype):
-    y_true_np = np.array([3, -0.5, 2, 7])
-    y_pred_np = np.array([2.5, 0.0, 2, 8])
+    y_true_np = np.array([3, -0.5, 2, 7], dtype="float32")
+    y_pred_np = np.array([2.5, 0.0, 2, 8], dtype="float32")
     check_array_api_metric(
         metric, array_namespace, device, dtype, y_true_np=y_true_np, y_pred_np=y_pred_np
     )
diff --git a/sklearn/utils/_array_api.py b/sklearn/utils/_array_api.py
index 64a61fd3cdd5f..73333c2158dc4 100644
--- a/sklearn/utils/_array_api.py
+++ b/sklearn/utils/_array_api.py
@@ -462,13 +462,15 @@ def _weighted_sum(sample_score, sample_weight, normalize=False, xp=None):
             sample_weight_np = None
         return float(numpy.average(sample_score_np, weights=sample_weight_np))
 
+    # We move to cpu device ahead of time since certain devices may not support
+    # float64, but we want the same precision for all devices and namespaces.
     if not xp.isdtype(sample_score.dtype, "real floating"):
-        # We move to cpu device ahead of time since certain devices may not support
-        # float64, but we want the same precision for all devices and namespaces.
         sample_score = xp.astype(xp.asarray(sample_score, device="cpu"), xp.float64)
 
     if sample_weight is not None:
-        sample_weight = xp.asarray(sample_weight, dtype=sample_score.dtype)
+        sample_weight = xp.asarray(
+            xp.asarray(sample_weight, device="cpu"), dtype=sample_score.dtype
+        )
         if not xp.isdtype(sample_weight.dtype, "real floating"):
             sample_weight = xp.astype(sample_weight, xp.float64)
 

From 0f6ea74edd71a5a0cbfa8b6c687e0c4a2d0a4357 Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Mon, 6 Nov 2023 23:39:40 +0100
Subject: [PATCH 03/29] updated docs

---
 doc/modules/array_api.rst |  1 +
 doc/whats_new/v1.4.rst    | 11 ++++++++---
 2 files changed, 9 insertions(+), 3 deletions(-)

diff --git a/doc/modules/array_api.rst b/doc/modules/array_api.rst
index 777f0d1e2f17c..0a860533f13a3 100644
--- a/doc/modules/array_api.rst
+++ b/doc/modules/array_api.rst
@@ -104,6 +104,7 @@ Metrics
 -------
 
 - :func:`sklearn.metrics.accuracy_score`
+- :func:`sklearn.metrics.mean_absolute_error`
 - :func:`sklearn.metrics.zero_one_loss`
 
 Tools
diff --git a/doc/whats_new/v1.4.rst b/doc/whats_new/v1.4.rst
index 00539260ffed6..acae24b6c8e2b 100644
--- a/doc/whats_new/v1.4.rst
+++ b/doc/whats_new/v1.4.rst
@@ -386,9 +386,14 @@ Changelog
 - |Enhancement| Added `neg_root_mean_squared_log_error_scorer` as scorer
   :pr:`26734` by :user:`Alejandro Martin Gil <101AlexMartin>`.
 
-- |Enhancement| :func:`sklearn.metrics.accuracy_score` and
-  :func:`sklearn.metrics.zero_one_loss` now support Array API compatible inputs.
-  :pr:`27137` by :user:`Edoardo Abati <EdAbati>`.
+- |MajorFeature| The following metrics now support the
+  `Array API <https://data-apis.org/array-api/latest/>`_. See
+  :ref:`array_api` for more details.
+
+  - :func:`sklearn.metrics.accuracy_score` :pr:`27137`` by :user:`Edoardo Abati <EdAbati>`
+  - :func:`sklearn.metrics.mean_absolute_error` :pr:`27736` by :user:`Edoardo Abati <EdAbati>`
+  - :func:`sklearn.metrics.zero_one_loss` :pr:`27137` by :user:`Edoardo Abati <EdAbati>`
+
 
 - |API| Deprecated `needs_threshold` and `needs_proba` from :func:`metrics.make_scorer`.
   These parameters will be removed in version 1.6. Instead, use `response_method` that

From 746d8eca1b9c7ad29c2aef1de9c2eb04fe7b877b Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Wed, 8 Nov 2023 08:40:56 +0100
Subject: [PATCH 04/29] returning float when scalar

---
 sklearn/metrics/_regression.py | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/sklearn/metrics/_regression.py b/sklearn/metrics/_regression.py
index 1dcb3e9a2e443..840edfae58c4d 100644
--- a/sklearn/metrics/_regression.py
+++ b/sklearn/metrics/_regression.py
@@ -182,7 +182,7 @@ def mean_absolute_error(
 
     Returns
     -------
-    loss : float or ndarray of floats
+    loss : float or array of floats
         If multioutput is 'raw_values', then mean absolute error is returned
         for each output separately.
         If multioutput is 'uniform_average' or an ndarray of weights, then the
@@ -219,7 +219,10 @@ def mean_absolute_error(
             # pass None as weights to np.average: uniform mean
             multioutput = None
 
-    return _average(output_errors, weights=multioutput)
+    mean_absolute_error = _average(output_errors, weights=multioutput)
+    if mean_absolute_error.shape == ():
+        return float(mean_absolute_error)
+    return mean_absolute_error
 
 
 @validate_params(

From 4b2604e8cf3fda4d79ac4e4f61519cdd476e275b Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Tue, 14 Nov 2023 19:56:59 +0100
Subject: [PATCH 05/29] fixed comment

Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>
---
 sklearn/metrics/tests/test_common.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/sklearn/metrics/tests/test_common.py b/sklearn/metrics/tests/test_common.py
index 0daf890dd2f65..6f80f8763d68a 100644
--- a/sklearn/metrics/tests/test_common.py
+++ b/sklearn/metrics/tests/test_common.py
@@ -1747,8 +1747,8 @@ def check_array_api_metric(
         metric_xp = metric(y_true_xp, y_pred_xp, sample_weight=sample_weight)
 
         if not isinstance(metric_xp, float):
-            # if the result is not a scalar, the array has to be in the CPU
-            # before transforming it to numpy
+            # If the result is not a scalar, the array has to be in the CPU
+            # before transforming it to a numpy array.
             metric_xp = xp.asarray(metric_xp, device="cpu")
 
         assert_allclose(

From bc9e6f86809cb3180f8129ca8fb4f7a41e53917b Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Tue, 14 Nov 2023 19:57:36 +0100
Subject: [PATCH 06/29] added float32 comment

Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>
---
 sklearn/metrics/tests/test_common.py | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/sklearn/metrics/tests/test_common.py b/sklearn/metrics/tests/test_common.py
index 6f80f8763d68a..1ee74374c1965 100644
--- a/sklearn/metrics/tests/test_common.py
+++ b/sklearn/metrics/tests/test_common.py
@@ -1799,6 +1799,8 @@ def check_array_api_multiclass_classification_metric(
 
 
 def check_array_api_regression_metric(metric, array_namespace, device, dtype):
+    # Not all Array API / device combinations support `float64` values, hence
+    # limit this test to the `float32` case for now.
     y_true_np = np.array([3, -0.5, 2, 7], dtype="float32")
     y_pred_np = np.array([2.5, 0.0, 2, 8], dtype="float32")
     check_array_api_metric(

From cb3613dd0b65d0e8f61babb6edac9d8bc856f2a8 Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Tue, 14 Nov 2023 20:05:22 +0100
Subject: [PATCH 07/29] improved error message

---
 sklearn/utils/_array_api.py | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/sklearn/utils/_array_api.py b/sklearn/utils/_array_api.py
index 73333c2158dc4..9571e2a2bac8a 100644
--- a/sklearn/utils/_array_api.py
+++ b/sklearn/utils/_array_api.py
@@ -634,10 +634,14 @@ def _average(X, axis=None, weights=None):
     if X.shape != weights.shape:
         if axis is None:
             raise TypeError(
-                "Axis must be specified when shapes of a and weights differ."
+                "Axis must be specified when shapes of X and weights differ. "
+                f"Got {X.shape = } and {weights.shape = }."
             )
         if weights.ndim != 1:
-            raise TypeError("1D weights expected when shapes of a and weights differ.")
+            raise TypeError(
+                "1D weights expected when shapes of X and weights differ. "
+                f"Got {X.shape = } and {weights.shape = }."
+            )
         if weights.shape[0] != X.shape[axis]:
             raise ValueError("Length of weights not compatible with specified axis.")
         weights = xp.broadcast_to(weights, (X.ndim - 1) * (1,) + weights.shape)

From 0b58f4bbad7d5569318a47579de87a63f1428f16 Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Tue, 14 Nov 2023 20:14:35 +0100
Subject: [PATCH 08/29] added test with axis=0

---
 sklearn/utils/tests/test_array_api.py | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/sklearn/utils/tests/test_array_api.py b/sklearn/utils/tests/test_array_api.py
index 161b2786dafdc..d94190780ea4e 100644
--- a/sklearn/utils/tests/test_array_api.py
+++ b/sklearn/utils/tests/test_array_api.py
@@ -385,12 +385,19 @@ def test_get_namespace_array_api_isdtype(wrapper):
         [[0.0, 1.0, 2.0, 3.0], [0.0, 1.0, 1.0, 2.0], 0, 2.25],
         [[[0.0, 1.0, 2.0, 3.0], [4.0, 5.0, 6.0, 7.0]], None, None, 3.5],
         [[[0.0, 1.0, 2.0, 3.0], [4.0, 5.0, 6.0, 7.0]], None, 1, [1.5, 5.5]],
+        [[[0.0, 1.0, 2.0, 3.0], [4.0, 5.0, 6.0, 7.0]], None, 0, [2.0, 3.0, 4.0, 5.0]],
         [
             [[0.0, 1.0, 2.0, 3.0], [4.0, 5.0, 6.0, 7.0]],
             [0.0, 1.0, 1.0, 2.0],
             1,
             [2.25, 6.25],
         ],
+        [
+            [[0.0, 1.0, 2.0, 3.0], [4.0, 5.0, 6.0, 7.0]],
+            [0.5, 2.0],
+            0,
+            [3.2, 4.2, 5.2, 6.2],
+        ],
     ],
 )
 def test__average(library, X, axis, weights, expected):

From 002cacf047dfc7ea4bf57f116fffbcf16cedcaeb Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Tue, 14 Nov 2023 20:39:32 +0100
Subject: [PATCH 09/29] added error tests

---
 sklearn/utils/tests/test_array_api.py | 36 +++++++++++++++++++++++++++
 1 file changed, 36 insertions(+)

diff --git a/sklearn/utils/tests/test_array_api.py b/sklearn/utils/tests/test_array_api.py
index d94190780ea4e..47fe9c9fdd68b 100644
--- a/sklearn/utils/tests/test_array_api.py
+++ b/sklearn/utils/tests/test_array_api.py
@@ -410,3 +410,39 @@ def test__average(library, X, axis, weights, expected):
         result = _average(xp.asarray(X), axis=axis, weights=weights)
 
     assert_allclose(result, expected)
+
+
+@skip_if_array_api_compat_not_configured
+@pytest.mark.parametrize("library", ["cupy", "cupy.array_api", "torch"])
+@pytest.mark.parametrize(
+    "axis,weights,expected_error,expected_msg",
+    [
+        (
+            None,
+            [0.5, 2.0],
+            TypeError,
+            "Axis must be specified when shapes of X and weights differ.",
+        ),
+        (
+            0,
+            [[0.5], [2.0]],
+            TypeError,
+            "1D weights expected when shapes of X and weights differ.",
+        ),
+        (
+            1,
+            [0.5, 2.0],
+            ValueError,
+            "Length of weights not compatible with specified axis.",
+        ),
+        (0, [0.5, -0.5], ZeroDivisionError, "Weights sum to zero, can't be normalized"),
+    ],
+)
+def test__average_error(library, axis, weights, expected_error, expected_msg):
+    xp = pytest.importorskip(library)
+
+    X = xp.asarray([[1.0, 2.0, 3.0], [1.0, 2.0, 3.0]], dtype=xp.float32)
+
+    with config_context(array_api_dispatch=True):
+        with pytest.raises(expected_error, match=expected_msg):
+            _average(X, axis=axis, weights=weights)

From 6f452a67a7649172fbd1add4a18f6b39157643b7 Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Tue, 14 Nov 2023 20:53:53 +0100
Subject: [PATCH 10/29] fix to dtype=float32

---
 sklearn/metrics/tests/test_common.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/sklearn/metrics/tests/test_common.py b/sklearn/metrics/tests/test_common.py
index 1ee74374c1965..1f9d76a9a8da4 100644
--- a/sklearn/metrics/tests/test_common.py
+++ b/sklearn/metrics/tests/test_common.py
@@ -1814,7 +1814,7 @@ def check_array_api_regression_metric(metric, array_namespace, device, dtype):
             dtype,
             y_true_np=y_true_np,
             y_pred_np=y_pred_np,
-            sample_weight=np.array([0.0, 0.1, 2.0, 1.0]),
+            sample_weight=np.array([0.0, 0.1, 2.0, 1.0], dtype="float32"),
         )
 
 

From 48bd567fd5403ccbec65f69189906240df16bca4 Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Wed, 15 Nov 2023 00:01:51 +0100
Subject: [PATCH 11/29] test multioutput

---
 sklearn/metrics/tests/test_common.py | 30 +++++++++++++++++++++++++++-
 1 file changed, 29 insertions(+), 1 deletion(-)

diff --git a/sklearn/metrics/tests/test_common.py b/sklearn/metrics/tests/test_common.py
index 1f9d76a9a8da4..411c9217616e0 100644
--- a/sklearn/metrics/tests/test_common.py
+++ b/sklearn/metrics/tests/test_common.py
@@ -1818,6 +1818,31 @@ def check_array_api_regression_metric(metric, array_namespace, device, dtype):
         )
 
 
+def check_array_api_multioutput_regression_metric(
+    metric, array_namespace, device, dtype
+):
+    # Not all Array API / device combinations support `float64` values, hence
+    # limit this test to the `float32` case for now.
+    y_true_np = np.array([[0.5, 1], [-1, 1], [7, -6]], dtype="float32")
+    y_pred_np = np.array([[0, 2], [-1, 2], [8, -5]], dtype="float32")
+
+    metric = partial(metric, multioutput="raw_values")
+
+    check_array_api_metric(
+        metric, array_namespace, device, dtype, y_true_np=y_true_np, y_pred_np=y_pred_np
+    )
+    if "sample_weight" in signature(metric).parameters:
+        check_array_api_metric(
+            metric,
+            array_namespace,
+            device,
+            dtype,
+            y_true_np=y_true_np,
+            y_pred_np=y_pred_np,
+            sample_weight=np.array([0.0, 0.1, 2.0], dtype="float32"),
+        )
+
+
 metric_checkers = {
     accuracy_score: [
         check_array_api_binary_classification_metric,
@@ -1827,7 +1852,10 @@ def check_array_api_regression_metric(metric, array_namespace, device, dtype):
         check_array_api_binary_classification_metric,
         check_array_api_multiclass_classification_metric,
     ],
-    mean_absolute_error: [check_array_api_regression_metric],
+    mean_absolute_error: [
+        check_array_api_regression_metric,
+        check_array_api_multioutput_regression_metric,
+    ],
 }
 
 

From 855aad03a6044f3ab091f71f771731d46ca90adb Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Wed, 15 Nov 2023 00:15:26 +0100
Subject: [PATCH 12/29] Update sklearn/metrics/_regression.py

Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>
---
 sklearn/metrics/_regression.py | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

diff --git a/sklearn/metrics/_regression.py b/sklearn/metrics/_regression.py
index 840edfae58c4d..3c97ae9a809b7 100644
--- a/sklearn/metrics/_regression.py
+++ b/sklearn/metrics/_regression.py
@@ -219,10 +219,15 @@ def mean_absolute_error(
             # pass None as weights to np.average: uniform mean
             multioutput = None
 
+    # Average across the outputs (if needed).
     mean_absolute_error = _average(output_errors, weights=multioutput)
-    if mean_absolute_error.shape == ():
-        return float(mean_absolute_error)
-    return mean_absolute_error
+    
+    # Since `y_pred.ndim <= 2` and `y_true.ndim <= 2`, the second call to _average
+    # should always return a scalar array that we convert to a Python float to
+    # consistently return the same eager evaluated value, irrespective of the
+    # Array API implementation.
+    assert mean_absolute_error.shape == ()
+    return float(mean_absolute_error)
 
 
 @validate_params(

From d3378d5ecd1a3763832db821337e1a0d311f72a4 Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Wed, 15 Nov 2023 08:52:33 +0100
Subject: [PATCH 13/29] fix linting

---
 sklearn/metrics/_regression.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/sklearn/metrics/_regression.py b/sklearn/metrics/_regression.py
index 3c97ae9a809b7..a7e6b7b1b18eb 100644
--- a/sklearn/metrics/_regression.py
+++ b/sklearn/metrics/_regression.py
@@ -221,7 +221,7 @@ def mean_absolute_error(
 
     # Average across the outputs (if needed).
     mean_absolute_error = _average(output_errors, weights=multioutput)
-    
+
     # Since `y_pred.ndim <= 2` and `y_true.ndim <= 2`, the second call to _average
     # should always return a scalar array that we convert to a Python float to
     # consistently return the same eager evaluated value, irrespective of the

From 64471851945c23df79c93b20e0fc7b119b5c0ef5 Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Wed, 15 Nov 2023 23:23:31 +0100
Subject: [PATCH 14/29] added new error message

---
 sklearn/utils/_array_api.py           | 5 ++++-
 sklearn/utils/tests/test_array_api.py | 7 +------
 2 files changed, 5 insertions(+), 7 deletions(-)

diff --git a/sklearn/utils/_array_api.py b/sklearn/utils/_array_api.py
index 9571e2a2bac8a..c48a9e80e43ff 100644
--- a/sklearn/utils/_array_api.py
+++ b/sklearn/utils/_array_api.py
@@ -643,7 +643,10 @@ def _average(X, axis=None, weights=None):
                 f"Got {X.shape = } and {weights.shape = }."
             )
         if weights.shape[0] != X.shape[axis]:
-            raise ValueError("Length of weights not compatible with specified axis.")
+            raise ValueError(
+                f"Length of weights ({weights.shape=}) not compatible with "
+                f"{X.shape=} and {axis=}."
+            )
         weights = xp.broadcast_to(weights, (X.ndim - 1) * (1,) + weights.shape)
         weights = xp.swapaxes(weights, -1, axis)
     weights_sum = xp.sum(weights, axis=axis)
diff --git a/sklearn/utils/tests/test_array_api.py b/sklearn/utils/tests/test_array_api.py
index 47fe9c9fdd68b..e67828202d60e 100644
--- a/sklearn/utils/tests/test_array_api.py
+++ b/sklearn/utils/tests/test_array_api.py
@@ -429,12 +429,7 @@ def test__average(library, X, axis, weights, expected):
             TypeError,
             "1D weights expected when shapes of X and weights differ.",
         ),
-        (
-            1,
-            [0.5, 2.0],
-            ValueError,
-            "Length of weights not compatible with specified axis.",
-        ),
+        (1, [0.5, 2.0], ValueError, "Length of weights"),
         (0, [0.5, -0.5], ZeroDivisionError, "Weights sum to zero, can't be normalized"),
     ],
 )

From b390da32cd7963fd8e24262de300aa863795e459 Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Mon, 27 Nov 2023 22:45:13 +0100
Subject: [PATCH 15/29] using _convert_to_numpy in tests

---
 sklearn/metrics/tests/test_common.py | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/sklearn/metrics/tests/test_common.py b/sklearn/metrics/tests/test_common.py
index 411c9217616e0..84284da2be3a9 100644
--- a/sklearn/metrics/tests/test_common.py
+++ b/sklearn/metrics/tests/test_common.py
@@ -55,6 +55,7 @@
 from sklearn.utils import shuffle
 from sklearn.utils._array_api import (
     _atol_for_type,
+    _convert_to_numpy,
     yield_namespace_device_dtype_combinations,
 )
 from sklearn.utils._testing import (
@@ -1747,9 +1748,7 @@ def check_array_api_metric(
         metric_xp = metric(y_true_xp, y_pred_xp, sample_weight=sample_weight)
 
         if not isinstance(metric_xp, float):
-            # If the result is not a scalar, the array has to be in the CPU
-            # before transforming it to a numpy array.
-            metric_xp = xp.asarray(metric_xp, device="cpu")
+            metric_xp = _convert_to_numpy(metric_xp, xp)
 
         assert_allclose(
             metric_xp,

From b8fe7c496de2b07712c1d00854bb3c73b4cdbb66 Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Mon, 27 Nov 2023 23:39:06 +0100
Subject: [PATCH 16/29] adding xfail for cupy.array_api

---
 sklearn/metrics/tests/test_common.py | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/sklearn/metrics/tests/test_common.py b/sklearn/metrics/tests/test_common.py
index 84284da2be3a9..6d420641f1d04 100644
--- a/sklearn/metrics/tests/test_common.py
+++ b/sklearn/metrics/tests/test_common.py
@@ -1869,4 +1869,10 @@ def yield_metric_checker_combinations(metric_checkers=metric_checkers):
 )
 @pytest.mark.parametrize("metric, check_func", yield_metric_checker_combinations())
 def test_array_api_compliance(metric, array_namespace, device, dtype, check_func):
+    if (
+        metric == mean_absolute_error
+        and check_func == check_array_api_regression_metric
+        and array_namespace == "cupy.array_api"
+    ):
+        pytest.xfail(reason="module 'cupy.array_api' has no attribute 'swapaxes'")
     check_func(metric, array_namespace, device, dtype)

From d3cf189cdd1344839648de935bdbd138d18b43d5 Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Tue, 12 Dec 2023 11:12:01 +0100
Subject: [PATCH 17/29] Update sklearn/utils/_array_api.py

Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>
---
 sklearn/utils/_array_api.py | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/sklearn/utils/_array_api.py b/sklearn/utils/_array_api.py
index c48a9e80e43ff..543aca6c5f266 100644
--- a/sklearn/utils/_array_api.py
+++ b/sklearn/utils/_array_api.py
@@ -469,6 +469,9 @@ def _weighted_sum(sample_score, sample_weight, normalize=False, xp=None):
 
     if sample_weight is not None:
         sample_weight = xp.asarray(
+            # TODO: remove the successive xp.asarray calls once
+            # https://github.com/data-apis/array-api/issues/647 is widely
+            # implemented by all supported Array API libraries.
             xp.asarray(sample_weight, device="cpu"), dtype=sample_score.dtype
         )
         if not xp.isdtype(sample_weight.dtype, "real floating"):

From fcd20784663e37748efc3b600ee1cd98d2766a59 Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Tue, 12 Mar 2024 17:46:29 +0100
Subject: [PATCH 18/29] remove redundant test

---
 sklearn/utils/tests/test_array_api.py | 69 ---------------------------
 1 file changed, 69 deletions(-)

diff --git a/sklearn/utils/tests/test_array_api.py b/sklearn/utils/tests/test_array_api.py
index c79abf806e7a0..6b5f600fb01c2 100644
--- a/sklearn/utils/tests/test_array_api.py
+++ b/sklearn/utils/tests/test_array_api.py
@@ -473,72 +473,3 @@ def test_get_namespace_array_api_isdtype(wrapper):
 
     with pytest.raises(ValueError, match="Unrecognized data type"):
         assert xp.isdtype(xp.int16, "unknown")
-
-
-@skip_if_array_api_compat_not_configured
-@pytest.mark.parametrize(
-    "library", ["numpy", "numpy.array_api", "cupy", "cupy.array_api", "torch"]
-)
-@pytest.mark.parametrize(
-    "X,weights,axis,expected",
-    [
-        [[0.0, 1.0, 2.0, 3.0], None, 0, 1.5],
-        [[0.0, 1.0, 2.0, 3.0], [0.0, 1.0, 1.0, 2.0], 0, 2.25],
-        [[[0.0, 1.0, 2.0, 3.0], [4.0, 5.0, 6.0, 7.0]], None, None, 3.5],
-        [[[0.0, 1.0, 2.0, 3.0], [4.0, 5.0, 6.0, 7.0]], None, 1, [1.5, 5.5]],
-        [[[0.0, 1.0, 2.0, 3.0], [4.0, 5.0, 6.0, 7.0]], None, 0, [2.0, 3.0, 4.0, 5.0]],
-        [
-            [[0.0, 1.0, 2.0, 3.0], [4.0, 5.0, 6.0, 7.0]],
-            [0.0, 1.0, 1.0, 2.0],
-            1,
-            [2.25, 6.25],
-        ],
-        [
-            [[0.0, 1.0, 2.0, 3.0], [4.0, 5.0, 6.0, 7.0]],
-            [0.5, 2.0],
-            0,
-            [3.2, 4.2, 5.2, 6.2],
-        ],
-    ],
-)
-def test__average(library, X, axis, weights, expected):
-    xp = pytest.importorskip(library)
-
-    if isinstance(expected, list):
-        expected = xp.asarray(expected)
-
-    with config_context(array_api_dispatch=True):
-        result = _average(xp.asarray(X), axis=axis, weights=weights)
-
-    assert_allclose(result, expected)
-
-
-@skip_if_array_api_compat_not_configured
-@pytest.mark.parametrize("library", ["cupy", "cupy.array_api", "torch"])
-@pytest.mark.parametrize(
-    "axis,weights,expected_error,expected_msg",
-    [
-        (
-            None,
-            [0.5, 2.0],
-            TypeError,
-            "Axis must be specified when shapes of X and weights differ.",
-        ),
-        (
-            0,
-            [[0.5], [2.0]],
-            TypeError,
-            "1D weights expected when shapes of X and weights differ.",
-        ),
-        (1, [0.5, 2.0], ValueError, "Length of weights"),
-        (0, [0.5, -0.5], ZeroDivisionError, "Weights sum to zero, can't be normalized"),
-    ],
-)
-def test__average_error(library, axis, weights, expected_error, expected_msg):
-    xp = pytest.importorskip(library)
-
-    X = xp.asarray([[1.0, 2.0, 3.0], [1.0, 2.0, 3.0]], dtype=xp.float32)
-
-    with config_context(array_api_dispatch=True):
-        with pytest.raises(expected_error, match=expected_msg):
-            _average(X, axis=axis, weights=weights)

From dab04a026326e52a925a3a2c2a0116650b3ddf9b Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Tue, 12 Mar 2024 17:52:52 +0100
Subject: [PATCH 19/29] updated relevant release note

---
 doc/whats_new/v1.4.rst | 4 +---
 doc/whats_new/v1.5.rst | 3 ++-
 2 files changed, 3 insertions(+), 4 deletions(-)

diff --git a/doc/whats_new/v1.4.rst b/doc/whats_new/v1.4.rst
index 790ebda9d9ee9..fdea14b41d056 100644
--- a/doc/whats_new/v1.4.rst
+++ b/doc/whats_new/v1.4.rst
@@ -434,9 +434,7 @@ See :ref:`array_api` for more details.
   :pr:`27137` by :user:`Edoardo Abati <EdAbati>`;
 - :func:`sklearn.model_selection.train_test_split` in :pr:`26855` by `Tim Head`_;
 - :func:`~utils.multiclass.is_multilabel` in :pr:`27601` by
-  :user:`Yaroslav Korobko <Tialo>`;
-- :func:`sklearn.metrics.mean_absolute_error` :pr:`27736` by :user:`Edoardo Abati <EdAbati>`.
-
+  :user:`Yaroslav Korobko <Tialo>`.
 
 **Classes:**
 
diff --git a/doc/whats_new/v1.5.rst b/doc/whats_new/v1.5.rst
index aef2d6d8af2a9..53f8812fa2de7 100644
--- a/doc/whats_new/v1.5.rst
+++ b/doc/whats_new/v1.5.rst
@@ -34,7 +34,8 @@ See :ref:`array_api` for more details.
 
 - :func:`sklearn.metrics.r2_score` now supports Array API compliant inputs.
   :pr:`27904` by :user:`Eric Lindgren <elindgren>`, `Franck Charras <fcharras>`,
-  `Olivier Grisel <ogrisel>` and `Tim Head <betatim>`.
+  `Olivier Grisel <ogrisel>` and `Tim Head <betatim>`;
+- :func:`sklearn.metrics.mean_absolute_error` :pr:`27736` by :user:`Edoardo Abati <EdAbati>`.
 
 **Classes:**
 

From ce388726e5c36885c17a6d1bc7d71422ada68bf0 Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Wed, 13 Mar 2024 08:21:57 +0100
Subject: [PATCH 20/29] re enabled
 check_array_api_multioutput_regression_metric

---
 sklearn/metrics/tests/test_common.py | 56 ++++------------------------
 1 file changed, 8 insertions(+), 48 deletions(-)

diff --git a/sklearn/metrics/tests/test_common.py b/sklearn/metrics/tests/test_common.py
index 9dc9c74cfc223..95c84db660342 100644
--- a/sklearn/metrics/tests/test_common.py
+++ b/sklearn/metrics/tests/test_common.py
@@ -1817,53 +1817,6 @@ def check_array_api_multiclass_classification_metric(
     )
 
 
-# def check_array_api_regression_metric(metric, array_namespace, device, dtype):
-#     # Not all Array API / device combinations support `float64` values, hence
-#     # limit this test to the `float32` case for now.
-#     y_true_np = np.array([3, -0.5, 2, 7], dtype="float32")
-#     y_pred_np = np.array([2.5, 0.0, 2, 8], dtype="float32")
-#     check_array_api_metric(
-#         metric, array_namespace, device, dtype, y_true_np=y_true_np,
-#         y_pred_np=y_pred_np
-#     )
-#     if "sample_weight" in signature(metric).parameters:
-#         check_array_api_metric(
-#             metric,
-#             array_namespace,
-#             device,
-#             dtype,
-#             y_true_np=y_true_np,
-#             y_pred_np=y_pred_np,
-#             sample_weight=np.array([0.0, 0.1, 2.0, 1.0], dtype="float32"),
-#         )
-
-
-# def check_array_api_multioutput_regression_metric(
-#     metric, array_namespace, device, dtype
-# ):
-#     # Not all Array API / device combinations support `float64` values, hence
-#     # limit this test to the `float32` case for now.
-#     y_true_np = np.array([[0.5, 1], [-1, 1], [7, -6]], dtype="float32")
-#     y_pred_np = np.array([[0, 2], [-1, 2], [8, -5]], dtype="float32")
-
-#     metric = partial(metric, multioutput="raw_values")
-
-#     check_array_api_metric(
-#         metric, array_namespace, device, dtype, y_true_np=y_true_np,
-#         y_pred_np=y_pred_np
-#     )
-#     if "sample_weight" in signature(metric).parameters:
-#         check_array_api_metric(
-#             metric,
-#             array_namespace,
-#             device,
-#             dtype,
-#             y_true_np=y_true_np,
-#             y_pred_np=y_pred_np,
-#             sample_weight=np.array([0.0, 0.1, 2.0], dtype="float32"),
-#         )
-
-
 def check_array_api_regression_metric(metric, array_namespace, device, dtype_name):
     y_true_np = np.array([[1, 3], [1, 2]], dtype=dtype_name)
     y_pred_np = np.array([[1, 4], [1, 1]], dtype=dtype_name)
@@ -1891,6 +1844,13 @@ def check_array_api_regression_metric(metric, array_namespace, device, dtype_nam
     )
 
 
+def check_array_api_multioutput_regression_metric(
+    metric, array_namespace, device, dtype_name
+):
+    metric = partial(metric, multioutput="raw_values")
+    check_array_api_regression_metric(metric, array_namespace, device, dtype_name)
+
+
 array_api_metric_checkers = {
     accuracy_score: [
         check_array_api_binary_classification_metric,
@@ -1902,7 +1862,7 @@ def check_array_api_regression_metric(metric, array_namespace, device, dtype_nam
     ],
     mean_absolute_error: [
         check_array_api_regression_metric,
-        # check_array_api_multioutput_regression_metric,
+        check_array_api_multioutput_regression_metric,
     ],
     r2_score: [check_array_api_regression_metric],
 }

From fcf5ee4a1ecc142c1b4453279440db2a3b0ac88b Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Wed, 13 Mar 2024 08:28:48 +0100
Subject: [PATCH 21/29] removed the cast to float

---
 sklearn/metrics/_regression.py | 9 +--------
 1 file changed, 1 insertion(+), 8 deletions(-)

diff --git a/sklearn/metrics/_regression.py b/sklearn/metrics/_regression.py
index 4811903808731..e16b3fab95584 100644
--- a/sklearn/metrics/_regression.py
+++ b/sklearn/metrics/_regression.py
@@ -226,14 +226,7 @@ def mean_absolute_error(
             multioutput = None
 
     # Average across the outputs (if needed).
-    mean_absolute_error = _average(output_errors, weights=multioutput)
-
-    # Since `y_pred.ndim <= 2` and `y_true.ndim <= 2`, the second call to _average
-    # should always return a scalar array that we convert to a Python float to
-    # consistently return the same eager evaluated value, irrespective of the
-    # Array API implementation.
-    assert mean_absolute_error.shape == ()
-    return float(mean_absolute_error)
+    return _average(output_errors, weights=multioutput)
 
 
 @validate_params(

From 58b799eba31a4b05898329779ed20359d5c44064 Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Wed, 13 Mar 2024 18:08:53 +0100
Subject: [PATCH 22/29] readded cast to float

---
 sklearn/metrics/_regression.py | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/sklearn/metrics/_regression.py b/sklearn/metrics/_regression.py
index e16b3fab95584..4811903808731 100644
--- a/sklearn/metrics/_regression.py
+++ b/sklearn/metrics/_regression.py
@@ -226,7 +226,14 @@ def mean_absolute_error(
             multioutput = None
 
     # Average across the outputs (if needed).
-    return _average(output_errors, weights=multioutput)
+    mean_absolute_error = _average(output_errors, weights=multioutput)
+
+    # Since `y_pred.ndim <= 2` and `y_true.ndim <= 2`, the second call to _average
+    # should always return a scalar array that we convert to a Python float to
+    # consistently return the same eager evaluated value, irrespective of the
+    # Array API implementation.
+    assert mean_absolute_error.shape == ()
+    return float(mean_absolute_error)
 
 
 @validate_params(

From 0e7ed2a91f192742d585f70e28ca599ce0d1cc7c Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Wed, 13 Mar 2024 18:36:34 +0100
Subject: [PATCH 23/29] match r2_score changes

---
 sklearn/metrics/_regression.py | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)

diff --git a/sklearn/metrics/_regression.py b/sklearn/metrics/_regression.py
index 4811903808731..1205ee8e48f75 100644
--- a/sklearn/metrics/_regression.py
+++ b/sklearn/metrics/_regression.py
@@ -212,12 +212,19 @@ def mean_absolute_error(
     >>> mean_absolute_error(y_true, y_pred, multioutput=[0.3, 0.7])
     0.85...
     """
-    xp, _ = get_namespace(y_true, y_pred)
-    y_type, y_true, y_pred, multioutput = _check_reg_targets(
-        y_true, y_pred, multioutput
+    input_arrays = [y_true, y_pred, sample_weight, multioutput]
+    xp, _ = get_namespace(*input_arrays)
+
+    dtype = _find_matching_floating_dtype(y_true, y_pred, sample_weight, xp=xp)
+
+    _, y_true, y_pred, multioutput = _check_reg_targets(
+        y_true, y_pred, multioutput, dtype=dtype, xp=xp
     )
     check_consistent_length(y_true, y_pred, sample_weight)
-    output_errors = _average(xp.abs(y_pred - y_true), weights=sample_weight, axis=0)
+
+    output_errors = _average(
+        xp.abs(y_pred - y_true), weights=sample_weight, axis=0, xp=xp
+    )
     if isinstance(multioutput, str):
         if multioutput == "raw_values":
             return output_errors

From 44038743c21348ac9f64145346eea92974bae537 Mon Sep 17 00:00:00 2001
From: Olivier Grisel <olivier.grisel@ensta.org>
Date: Mon, 25 Mar 2024 18:19:26 +0100
Subject: [PATCH 24/29] Trigger CI


From 3c311b02f5e6828200d03f0a2c3bc8151292e257 Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Mon, 25 Mar 2024 18:31:45 +0100
Subject: [PATCH 25/29] add missing :user:

---
 doc/whats_new/v1.5.rst | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/doc/whats_new/v1.5.rst b/doc/whats_new/v1.5.rst
index 53f8812fa2de7..d203707b54ab3 100644
--- a/doc/whats_new/v1.5.rst
+++ b/doc/whats_new/v1.5.rst
@@ -33,8 +33,8 @@ See :ref:`array_api` for more details.
 **Functions:**
 
 - :func:`sklearn.metrics.r2_score` now supports Array API compliant inputs.
-  :pr:`27904` by :user:`Eric Lindgren <elindgren>`, `Franck Charras <fcharras>`,
-  `Olivier Grisel <ogrisel>` and `Tim Head <betatim>`;
+  :pr:`27904` by :user:`Eric Lindgren <elindgren>`, :user:`Franck Charras <fcharras>`,
+  :user:`Olivier Grisel <ogrisel>` and :user:`Tim Head <betatim>`;
 - :func:`sklearn.metrics.mean_absolute_error` :pr:`27736` by :user:`Edoardo Abati <EdAbati>`.
 
 **Classes:**

From 5cc9c25826ce3e30f636bafa0d008ad17e86d917 Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Wed, 8 May 2024 12:16:49 +0200
Subject: [PATCH 26/29] moved to 1.6 whatsnew

---
 doc/whats_new/v1.5.rst | 1 -
 doc/whats_new/v1.6.rst | 3 ++-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/doc/whats_new/v1.5.rst b/doc/whats_new/v1.5.rst
index 05d2f1e036eb1..9ba894e0e4157 100644
--- a/doc/whats_new/v1.5.rst
+++ b/doc/whats_new/v1.5.rst
@@ -80,7 +80,6 @@ See :ref:`array_api` for more details.
 - :func:`sklearn.metrics.r2_score` now supports Array API compliant inputs.
   :pr:`27904` by :user:`Eric Lindgren <elindgren>`, :user:`Franck Charras <fcharras>`,
   :user:`Olivier Grisel <ogrisel>` and :user:`Tim Head <betatim>`;
-- :func:`sklearn.metrics.mean_absolute_error` :pr:`27736` by :user:`Edoardo Abati <EdAbati>`.
 
 **Classes:**
 
diff --git a/doc/whats_new/v1.6.rst b/doc/whats_new/v1.6.rst
index 6eda6717b3d1b..24fc62beb15e1 100644
--- a/doc/whats_new/v1.6.rst
+++ b/doc/whats_new/v1.6.rst
@@ -35,10 +35,11 @@ See :ref:`array_api` for more details.
 - :func:`sklearn.metrics.mean_tweedie_deviance` now supports Array API compatible
   inputs.
   :pr:`28106` by :user:`Thomas Li <lithomas1>`
+- :func:`sklearn.metrics.mean_absolute_error` :pr:`27736` by :user:`Edoardo Abati <EdAbati>`.
 
 **Classes:**
 
-- 
+-
 
 Changelog
 ---------

From 18d3ec320a0214296754505840e333754a4bf6f2 Mon Sep 17 00:00:00 2001
From: Edoardo Abati <29585319+EdAbati@users.noreply.github.com>
Date: Tue, 14 May 2024 18:20:46 +0200
Subject: [PATCH 27/29] removed reduntant conversion

---
 sklearn/metrics/tests/test_common.py | 3 ---
 1 file changed, 3 deletions(-)

diff --git a/sklearn/metrics/tests/test_common.py b/sklearn/metrics/tests/test_common.py
index f4c8f6a0eeb92..ae47ffe3d6a56 100644
--- a/sklearn/metrics/tests/test_common.py
+++ b/sklearn/metrics/tests/test_common.py
@@ -1758,9 +1758,6 @@ def check_array_api_metric(
     with config_context(array_api_dispatch=True):
         metric_xp = metric(y_true_xp, y_pred_xp, sample_weight=sample_weight)
 
-        if not isinstance(metric_xp, float):
-            metric_xp = _convert_to_numpy(metric_xp, xp)
-
         assert_allclose(
             _convert_to_numpy(xp.asarray(metric_xp), xp),
             metric_np,

From e0dea5b0c0a33e58ed6b8cb243fbe6ce0eb3c2d3 Mon Sep 17 00:00:00 2001
From: Olivier Grisel <olivier.grisel@ensta.org>
Date: Wed, 15 May 2024 11:06:34 +0200
Subject: [PATCH 28/29] Revert unrelated change to 1.5.

---
 doc/whats_new/v1.5.rst | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/doc/whats_new/v1.5.rst b/doc/whats_new/v1.5.rst
index 1b9a8c6d61f3e..4eb6ecd9264f1 100644
--- a/doc/whats_new/v1.5.rst
+++ b/doc/whats_new/v1.5.rst
@@ -82,7 +82,7 @@ See :ref:`array_api` for more details.
 **Functions:**
 
 - :func:`sklearn.metrics.r2_score` now supports Array API compliant inputs.
-  :pr:`27904` by :user:`Eric Lindgren <elindgren>`, :user:`Franck Charras <fcharras>`,
+  :pr:`27904` by :user:`Eric Lindgren <elindgren>`, `Franck Charras <fcharras>`,
   :user:`Olivier Grisel <ogrisel>` and :user:`Tim Head <betatim>`;
 
 **Classes:**

From 0fa23d311cf2ca2a4c355917bbc2dc363ecd5a6c Mon Sep 17 00:00:00 2001
From: Olivier Grisel <olivier.grisel@ensta.org>
Date: Wed, 15 May 2024 11:08:05 +0200
Subject: [PATCH 29/29] Revert more unrelated changelog.

---
 doc/whats_new/v1.5.rst | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/doc/whats_new/v1.5.rst b/doc/whats_new/v1.5.rst
index 4eb6ecd9264f1..5fdc0707ffbee 100644
--- a/doc/whats_new/v1.5.rst
+++ b/doc/whats_new/v1.5.rst
@@ -83,7 +83,7 @@ See :ref:`array_api` for more details.
 
 - :func:`sklearn.metrics.r2_score` now supports Array API compliant inputs.
   :pr:`27904` by :user:`Eric Lindgren <elindgren>`, `Franck Charras <fcharras>`,
-  :user:`Olivier Grisel <ogrisel>` and :user:`Tim Head <betatim>`;
+  `Olivier Grisel <ogrisel>` and `Tim Head <betatim>`.
 
 **Classes:**