feat: metrics as methods of models (#77)

Closes #64.

### Summary of Changes

Metrics are now methods of classifiers and regressors. They also take a
validation or test set as input now instead of two columns representing
predicated and expected values.

---------

Co-authored-by: lars-reimann <[email protected]>

src/safeds/ml/classification/_classifier.py

@@ -1,6 +1,7 @@
 from abc import ABC, abstractmethod
 
 from safeds.data.tabular.containers import Table, TaggedTable
+from sklearn.metrics import accuracy_score as sk_accuracy_score
 
 
 class Classifier(ABC):
@@ -44,3 +45,25 @@ def predict(self, dataset: Table) -> TaggedTable:
         PredictionError
             If prediction with the given dataset failed.
         """
+
+    def accuracy(self, validation_or_test_set: TaggedTable) -> float:
+        """
+        Predicts the target values for the features in the validation or test set and compares it to the expected
+        results.
+
+        Parameters
+        ----------
+        validation_or_test_set : TaggedTable
+            The validation or test set.
+
+        Returns
+        -------
+        accuracy : float
+            The calculated accuracy score, i.e. the percentage of equal data.
+        """
+
+        expected = validation_or_test_set.target
+        predicted = self.predict(validation_or_test_set.features).target
+
+        # noinspection PyProtectedMember
+        return sk_accuracy_score(expected._data, predicted._data)

src/safeds/ml/regression/_regressor.py

@@ -1,6 +1,9 @@
 from abc import ABC, abstractmethod
 
-from safeds.data.tabular.containers import Table, TaggedTable
+from safeds.data.tabular.containers import Column, Table, TaggedTable
+from safeds.exceptions import ColumnLengthMismatchError
+from sklearn.metrics import mean_absolute_error as sk_mean_absolute_error
+from sklearn.metrics import mean_squared_error as sk_mean_squared_error
 
 
 class Regressor(ABC):
@@ -44,3 +47,59 @@ def predict(self, dataset: Table) -> TaggedTable:
         PredictionError
             If prediction with the given dataset failed.
         """
+
+    def mean_squared_error(self, validation_or_test_set: TaggedTable) -> float:
+        """
+        Return the mean squared error, calculated from a given known truth and a column to compare.
+
+        Parameters
+        ----------
+        validation_or_test_set : TaggedTable
+            The validation or test set.
+
+        Returns
+        -------
+        mean_squared_error : float
+            The calculated mean squared error (the average of the distance of each individual row squared).
+        """
+
+        expected = validation_or_test_set.target
+        predicted = self.predict(validation_or_test_set.features).target
+
+        _check_metrics_preconditions(predicted, expected)
+        return sk_mean_squared_error(expected._data, predicted._data)
+
+    def mean_absolute_error(self, validation_or_test_set: TaggedTable) -> float:
+        """
+        Return the mean absolute error, calculated from a given known truth and a column to compare.
+
+        Parameters
+        ----------
+        validation_or_test_set : TaggedTable
+            The validation or test set.
+
+        Returns
+        -------
+        mean_absolute_error : float
+            The calculated mean absolute error (the average of the distance of each individual row).
+        """
+
+        expected = validation_or_test_set.target
+        predicted = self.predict(validation_or_test_set.features).target
+
+        _check_metrics_preconditions(predicted, expected)
+        return sk_mean_absolute_error(expected._data, predicted._data)
+
+
+def _check_metrics_preconditions(actual: Column, expected: Column) -> None:
+    if not actual.type.is_numeric():
+        raise TypeError(f"Column 'actual' is not numerical but {actual.type}.")
+    if not expected.type.is_numeric():
+        raise TypeError(f"Column 'expected' is not numerical but {expected.type}.")
+
+    if actual._data.size != expected._data.size:
+        raise ColumnLengthMismatchError(
+            "\n".join(
+                [f"{column.name}: {column._data.size}" for column in [actual, expected]]
+            )
+        )

tests/safeds/ml/classification/_classifier/_dummy_classifier.py

@@ -0,0 +1,26 @@
+from safeds.data.tabular.containers import Table, TaggedTable
+from safeds.ml.classification import Classifier
+
+
+class DummyClassifier(Classifier):
+    """
+    Dummy classifier to test metrics.
+
+    Metrics methods expect a `TaggedTable` as input with two columns:
+
+    - `predicted`: The predicted targets.
+    - `expected`: The correct targets.
+
+    `target_name` must be set to `"expected"`.
+    """
+
+    def fit(self, training_set: TaggedTable) -> None:
+        pass
+
+    def predict(self, dataset: Table) -> TaggedTable:
+        # Needed until https://github.com/Safe-DS/Stdlib/issues/75 is fixed
+        predicted = dataset.get_column("predicted")
+        feature = predicted.rename("feature")
+        dataset = Table.from_columns([feature, predicted])
+
+        return TaggedTable(dataset, target_name="predicted")

tests/safeds/ml/classification/_classifier/test_accuracy.py

@@ -0,0 +1,20 @@
+import pandas as pd
+from safeds.data.tabular.containers import Column, Table, TaggedTable
+
+from ._dummy_classifier import DummyClassifier
+
+
+def test_accuracy() -> None:
+    c1 = Column(pd.Series(data=[1, 2, 3, 4]), "predicted")
+    c2 = Column(pd.Series(data=[1, 2, 3, 3]), "expected")
+    table = TaggedTable(Table.from_columns([c1, c2]), target_name="expected")
+
+    assert DummyClassifier().accuracy(table) == 0.75
+
+
+def test_accuracy_different_types() -> None:
+    c1 = Column(pd.Series(data=["1", "2", "3", "4"]), "predicted")
+    c2 = Column(pd.Series(data=[1, 2, 3, 3]), "expected")
+    table = TaggedTable(Table.from_columns([c1, c2]), target_name="expected")
+
+    assert DummyClassifier().accuracy(table) == 0.0

tests/safeds/ml/regression/_regressor/_dummy_regressor.py

@@ -0,0 +1,26 @@
+from safeds.data.tabular.containers import Table, TaggedTable
+from safeds.ml.regression import Regressor
+
+
+class DummyRegressor(Regressor):
+    """
+    Dummy regressor to test metrics.
+
+    Metrics methods expect a `TaggedTable` as input with two columns:
+
+    - `predicted`: The predicted targets.
+    - `expected`: The correct targets.
+
+    `target_name` must be set to `"expected"`.
+    """
+
+    def fit(self, training_set: TaggedTable) -> None:
+        pass
+
+    def predict(self, dataset: Table) -> TaggedTable:
+        # Needed until https://github.com/Safe-DS/Stdlib/issues/75 is fixed
+        predicted = dataset.get_column("predicted")
+        feature = predicted.rename("feature")
+        dataset = Table.from_columns([feature, predicted])
+
+        return TaggedTable(dataset, target_name="predicted")

tests/safeds/ml/regression/metrics/test_check_metrics_preconditions.py → tests/safeds/ml/regression/_regressor/test_check_metrics_preconditions.py

@@ -2,9 +2,9 @@
 import pytest
 from safeds.data.tabular.containers import Column
 from safeds.exceptions import ColumnLengthMismatchError
-from safeds.ml.regression.metrics._module_level_functions import (
-    _check_metrics_preconditions,
-)
+
+# noinspection PyProtectedMember
+from safeds.ml.regression._regressor import _check_metrics_preconditions
 
 
 @pytest.mark.parametrize(

tests/safeds/ml/regression/_regressor/test_mean_absolute_error.py

@@ -0,0 +1,26 @@
+import pytest
+from safeds.data.tabular.containers import Column, Table, TaggedTable
+
+from ._dummy_regressor import DummyRegressor
+
+
+@pytest.mark.parametrize(
+    "predicted, expected, result",
+    [
+        ([1, 2], [1, 2], 0),
+        ([0, 0], [1, 1], 1),
+        ([1, 1, 1], [2, 2, 11], 4),
+        ([0, 0, 0], [10, 2, 18], 10),
+        ([0.5, 0.5], [1.5, 1.5], 1),
+    ],
+)
+def test_mean_absolute_error_valid(
+    predicted: list[float], expected: list[float], result: float
+) -> None:
+    predicted_column = Column(predicted, "predicted")
+    expected_column = Column(expected, "expected")
+    table = TaggedTable(
+        Table.from_columns([predicted_column, expected_column]), target_name="expected"
+    )
+
+    assert DummyRegressor().mean_absolute_error(table) == result

tests/safeds/ml/regression/_regressor/test_mean_squared_error.py

@@ -0,0 +1,20 @@
+import pytest
+from safeds.data.tabular.containers import Column, Table, TaggedTable
+
+from ._dummy_regressor import DummyRegressor
+
+
+@pytest.mark.parametrize(
+    "predicted, expected, result",
+    [([1, 2], [1, 2], 0), ([0, 0], [1, 1], 1), ([1, 1, 1], [2, 2, 11], 34)],
+)
+def test_mean_squared_error_valid(
+    predicted: list[float], expected: list[float], result: float
+) -> None:
+    predicted_column = Column(predicted, "predicted")
+    expected_column = Column(expected, "expected")
+    table = TaggedTable(
+        Table.from_columns([predicted_column, expected_column]), target_name="expected"
+    )
+
+    assert DummyRegressor().mean_squared_error(table) == result