scikit-learn · OmarManzoor · May 15, 2024 · Nov 2, 2023 · Nov 6, 2023 · Nov 6, 2023
diff --git a/doc/modules/array_api.rst b/doc/modules/array_api.rst
@@ -106,6 +106,7 @@ Metrics
 -------
 
 - :func:`sklearn.metrics.accuracy_score`
+- :func:`sklearn.metrics.mean_absolute_error`
 - :func:`sklearn.metrics.mean_tweedie_deviance`
 - :func:`sklearn.metrics.r2_score`
 - :func:`sklearn.metrics.zero_one_loss`

diff --git a/doc/whats_new/v1.6.rst b/doc/whats_new/v1.6.rst
@@ -35,6 +35,7 @@ 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:**
 

diff --git a/sklearn/metrics/_regression.py b/sklearn/metrics/_regression.py
@@ -189,7 +189,7 @@ 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
@@ -213,19 +213,35 @@ def mean_absolute_error(
     >>> mean_absolute_error(y_true, y_pred, multioutput=[0.3, 0.7])
     0.85...
     """
-    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 = 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, xp=xp
+    )
     if isinstance(multioutput, str):
         if multioutput == "raw_values":
             return output_errors
         elif multioutput == "uniform_average":
             # pass None as weights to np.average: uniform mean
             multioutput = None
 
-    return np.average(output_errors, weights=multioutput)
+    # 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)
 
 
 @validate_params(

diff --git a/sklearn/metrics/tests/test_common.py b/sklearn/metrics/tests/test_common.py
@@ -1879,6 +1879,13 @@ def check_array_api_regression_metric_multioutput(
     )
 
 
+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,
@@ -1893,6 +1900,10 @@ def check_array_api_regression_metric_multioutput(
         check_array_api_regression_metric,
         check_array_api_regression_metric_multioutput,
     ],
+    mean_absolute_error: [
+        check_array_api_regression_metric,
+        check_array_api_multioutput_regression_metric,
+    ],
 }