feat: accept metric names without neg_ prefix in summarize()

skore/src/skore/_sklearn/_estimator/metrics_accessor.py

@@ -54,6 +54,18 @@ class _MetricsAccessor(_BaseAccessor[EstimatorReport], DirNamesMixin):
         "custom_metric": {"name": "Custom metric", "icon": ""},
     }
 
+    _METRIC_ALIASES: dict[str, str] = {
+        "mean_squared_error": "neg_mean_squared_error",
+        "mean_absolute_error": "neg_mean_absolute_error",
+        "root_mean_squared_error": "neg_root_mean_squared_error",
+        "mean_absolute_percentage_error": "neg_mean_absolute_percentage_error",
+        "median_absolute_error": "neg_median_absolute_error",
+        "log_loss": "neg_log_loss",
+        "brier_score": "neg_brier_score",
+        "mean_poisson_deviance": "neg_mean_poisson_deviance",
+        "mean_gamma_deviance": "neg_mean_gamma_deviance",
+    }
+
     def __init__(self, parent: EstimatorReport) -> None:
         super().__init__(parent)
 
@@ -208,7 +220,8 @@ def summarize(
         for metric_name, metric_ in zip(metric_names, metrics, strict=True):
             if isinstance(metric_, str) and metric_ not in self._score_or_loss_info:
                 try:
-                    metric_ = sklearn.metrics.get_scorer(metric_)
+                    resolved_metric = self._METRIC_ALIASES.get(metric_, metric_)
+                    metric_ = sklearn_metrics.get_scorer(resolved_metric)
                 except ValueError as err:
                     raise ValueError(
                         f"Invalid metric: {metric_!r}. "

skore/tests/unit/reports/estimator/metrics/test_metric_aliases.py

@@ -0,0 +1,50 @@
+import pandas as pd
+import pytest
+from sklearn.datasets import make_regression
+from sklearn.linear_model import LinearRegression
+from sklearn.model_selection import train_test_split
+
+from skore import EstimatorReport
+
+
+@pytest.fixture
+def linear_regression_with_test():
+    X, y = make_regression(n_samples=100, n_features=5, random_state=0)
+    X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=0)
+    regressor = LinearRegression().fit(X_train, y_train)
+    return regressor, X_test, y_test
+
+
+def test_metric_aliases_without_neg_prefix(linear_regression_with_test):
+    """Check that metrics can be passed without the 'neg_' prefix and are
+    automatically resolved to their 'neg_' prefixed sklearn scorer equivalent.
+    This is the fix for https://github.com/probabl-ai/skore/issues/2607
+    """
+    regressor, X_test, y_test = linear_regression_with_test
+    report = EstimatorReport(regressor, X_test=X_test, y_test=y_test)
+
+    display = report.metrics.summarize(
+        metric=["mean_squared_error", "mean_absolute_error", "root_mean_squared_error"]
+    )
+
+    result = display.frame()
+    assert isinstance(result, pd.DataFrame)
+    assert "Mean Squared Error" in result.index
+    assert "Mean Absolute Error" in result.index
+    assert "Root Mean Squared Error" in result.index
+
+
+def test_metric_aliases_same_result_as_neg_prefix(linear_regression_with_test):
+    """Check that passing 'mean_squared_error' gives the same score as
+    passing 'neg_mean_squared_error'.
+    """
+    regressor, X_test, y_test = linear_regression_with_test
+    report = EstimatorReport(regressor, X_test=X_test, y_test=y_test)
+
+    display_with_neg = report.metrics.summarize(metric=["neg_mean_squared_error"])
+    display_without_neg = report.metrics.summarize(metric=["mean_squared_error"])
+
+    score_with_neg = display_with_neg.frame().values[0][0]
+    score_without_neg = display_without_neg.frame().values[0][0]
+
+    assert score_with_neg == pytest.approx(score_without_neg)

sphinx/user_guide/reporters.rst

@@ -84,7 +84,9 @@ regression). Nevertheless, you can specify the metrics you want to compute thank
 `scoring` parameter. We accept different types:
 
 1. A string that corresponds to a scikit-learn scorer name or a built-in `skore`
-   metric name,
+   metric name. For error metrics, you can pass the name with or without the
+   ``neg_`` prefix (e.g., ``"mean_squared_error"`` or ``"neg_mean_squared_error"``
+   are both valid),
 
 2. A callable,