feat: hyperparameter optimization for RNNs and CNNs

src/safeds/data/labeled/containers/_time_series_dataset.py

@@ -231,7 +231,7 @@ def _into_dataloader_with_window(
         batch_size:
             The size of data batches that should be loaded at one time.
         continuous:
-            Whether or not to continue the forecast in the steps before forecast horizon.
+            Whether to continue the forecast in the steps before forecast horizon.
 
         Raises
         ------

src/safeds/ml/metrics/_classification_metrics.py

@@ -1,15 +1,12 @@
 from __future__ import annotations
 
 from abc import ABC, abstractmethod
-from typing import TYPE_CHECKING, Any
+from typing import Any
 
-from safeds.data.labeled.containers import TabularDataset
-from safeds.data.tabular.containers import Table
+from safeds.data.labeled.containers import TabularDataset, TimeSeriesDataset
+from safeds.data.tabular.containers import Column, Table
 from safeds.exceptions import ColumnLengthMismatchError
 
-if TYPE_CHECKING:
-    from safeds.data.tabular.containers import Column
-
 
 class ClassificationMetrics(ABC):
     """A collection of classification metrics."""
@@ -18,7 +15,11 @@ class ClassificationMetrics(ABC):
     def __init__(self) -> None: ...
 
     @staticmethod
-    def summarize(predicted: Column | TabularDataset, expected: Column | TabularDataset, positive_class: Any) -> Table:
+    def summarize(
+        predicted: Column | TabularDataset | TimeSeriesDataset,
+        expected: Column | TabularDataset | TimeSeriesDataset,
+        positive_class: Any,
+    ) -> Table:
         """
         Summarize classification metrics on the given data.
 
@@ -53,7 +54,10 @@ def summarize(predicted: Column | TabularDataset, expected: Column | TabularData
         )
 
     @staticmethod
-    def accuracy(predicted: Column | TabularDataset, expected: Column | TabularDataset) -> float:
+    def accuracy(
+        predicted: Column | TabularDataset | TimeSeriesDataset,
+        expected: Column | TabularDataset | TimeSeriesDataset,
+    ) -> float:
         """
         Compute the accuracy on the given data.
 
@@ -87,7 +91,11 @@ def accuracy(predicted: Column | TabularDataset, expected: Column | TabularDatas
             return 0.0  # Types are not compatible, so no prediction can be correct
 
     @staticmethod
-    def f1_score(predicted: Column | TabularDataset, expected: Column | TabularDataset, positive_class: Any) -> float:
+    def f1_score(
+        predicted: Column | TabularDataset | TimeSeriesDataset,
+        expected: Column | TabularDataset | TimeSeriesDataset,
+        positive_class: Any,
+    ) -> float:
         """
         Compute the F₁ score on the given data.
 
@@ -122,7 +130,11 @@ def f1_score(predicted: Column | TabularDataset, expected: Column | TabularDatas
         return 2 * true_positives / (2 * true_positives + false_positives + false_negatives)
 
     @staticmethod
-    def precision(predicted: Column | TabularDataset, expected: Column | TabularDataset, positive_class: Any) -> float:
+    def precision(
+        predicted: Column | TabularDataset | TimeSeriesDataset,
+        expected: Column | TabularDataset | TimeSeriesDataset,
+        positive_class: Any,
+    ) -> float:
         """
         Compute the precision on the given data.
 
@@ -156,7 +168,11 @@ def precision(predicted: Column | TabularDataset, expected: Column | TabularData
         return true_positives / predicted_positives
 
     @staticmethod
-    def recall(predicted: Column | TabularDataset, expected: Column | TabularDataset, positive_class: Any) -> float:
+    def recall(
+        predicted: Column | TabularDataset | TimeSeriesDataset,
+        expected: Column | TabularDataset | TimeSeriesDataset,
+        positive_class: Any,
+    ) -> float:
         """
         Compute the recall on the given data.
 
@@ -190,9 +206,9 @@ def recall(predicted: Column | TabularDataset, expected: Column | TabularDataset
         return true_positives / actual_positives
 
 
-def _extract_target(column_or_dataset: Column | TabularDataset) -> Column:
+def _extract_target(column_or_dataset: Column | TabularDataset | TimeSeriesDataset) -> Column:
     """Extract the target column from the given column or dataset."""
-    if isinstance(column_or_dataset, TabularDataset):
+    if isinstance(column_or_dataset, TabularDataset | TimeSeriesDataset):
         return column_or_dataset.target
     else:
         return column_or_dataset

src/safeds/ml/metrics/_regression_metrics.py

@@ -2,7 +2,7 @@
 
 from abc import ABC, abstractmethod
 
-from safeds.data.labeled.containers import TabularDataset
+from safeds.data.labeled.containers import TabularDataset, TimeSeriesDataset
 from safeds.data.tabular.containers import Column, Table
 from safeds.exceptions import ColumnLengthMismatchError
 
@@ -14,7 +14,10 @@ class RegressionMetrics(ABC):
     def __init__(self) -> None: ...
 
     @staticmethod
-    def summarize(predicted: Column | TabularDataset, expected: Column | TabularDataset) -> Table:
+    def summarize(
+        predicted: Column | TabularDataset | TimeSeriesDataset,
+        expected: Column | TabularDataset | TimeSeriesDataset,
+    ) -> Table:
         """
         Summarize regression metrics on the given data.
 
@@ -57,7 +60,10 @@ def summarize(predicted: Column | TabularDataset, expected: Column | TabularData
         )
 
     @staticmethod
-    def coefficient_of_determination(predicted: Column | TabularDataset, expected: Column | TabularDataset) -> float:
+    def coefficient_of_determination(
+        predicted: Column | TabularDataset | TimeSeriesDataset,
+        expected: Column | TabularDataset | TimeSeriesDataset,
+    ) -> float:
         """
         Compute the coefficient of determination (R²) on the given data.
 
@@ -92,6 +98,20 @@ def coefficient_of_determination(predicted: Column | TabularDataset, expected: C
         predicted = _extract_target(predicted)
         _check_equal_length(predicted, expected)
 
+        # For TimeSeries Predictions, where the output is a list of values.
+        # Expected results are internally converted to a column containing multiple Columns for each prediction window
+        # Currently only used in fit_by_exhaustive_search, where prediction metrics have to be calculated internally.
+        if isinstance(expected.get_value(0), Column):
+            sum_of_coefficient_of_determination = 0.0
+            for i in range(expected.row_count):
+                predicted_row_as_col: Column = Column("predicted", predicted[i])
+                expected_row_as_col = expected.get_value(i)
+                sum_of_coefficient_of_determination += RegressionMetrics.coefficient_of_determination(
+                    predicted_row_as_col,
+                    expected_row_as_col,
+                )
+            return sum_of_coefficient_of_determination / expected.row_count
+
         residual_sum_of_squares = (expected._series - predicted._series).pow(2).sum()
         total_sum_of_squares = (expected._series - expected._series.mean()).pow(2).sum()
 
@@ -104,7 +124,10 @@ def coefficient_of_determination(predicted: Column | TabularDataset, expected: C
         return 1 - residual_sum_of_squares / total_sum_of_squares
 
     @staticmethod
-    def mean_absolute_error(predicted: Column | TabularDataset, expected: Column | TabularDataset) -> float:
+    def mean_absolute_error(
+        predicted: Column | TabularDataset | TimeSeriesDataset,
+        expected: Column | TabularDataset | TimeSeriesDataset,
+    ) -> float:
         """
         Compute the mean absolute error (MAE) on the given data.
 
@@ -131,10 +154,27 @@ def mean_absolute_error(predicted: Column | TabularDataset, expected: Column | T
         if expected.row_count == 0:
             return 0.0  # Everything was predicted correctly (since there is nothing to predict)
 
+        # For TimeSeries Predictions, where the output is a list of values.
+        # Expected results are internally converted to a column containing multiple Columns for each prediction window
+        # Currently only used in fit_by_exhaustive_search, where prediction metrics have to be calculated internally.
+        if isinstance(expected.get_value(0), Column):
+            sum_of_mean_absolute_errors = 0.0
+            for i in range(expected.row_count):
+                predicted_row_as_col: Column = Column("predicted", predicted[i])
+                expected_row_as_col = expected.get_value(i)
+                sum_of_mean_absolute_errors += RegressionMetrics.mean_absolute_error(
+                    predicted_row_as_col,
+                    expected_row_as_col,
+                )
+            return sum_of_mean_absolute_errors / expected.row_count
+
         return (expected._series - predicted._series).abs().mean()
 
     @staticmethod
-    def mean_directional_accuracy(predicted: Column | TabularDataset, expected: Column | TabularDataset) -> float:
+    def mean_directional_accuracy(
+        predicted: Column | TabularDataset | TimeSeriesDataset,
+        expected: Column | TabularDataset | TimeSeriesDataset,
+    ) -> float:
         """
         Compute the mean directional accuracy (MDA) on the given data.
 
@@ -172,7 +212,10 @@ def mean_directional_accuracy(predicted: Column | TabularDataset, expected: Colu
         return predicted_directions.eq(expected_directions).mean()
 
     @staticmethod
-    def mean_squared_error(predicted: Column | TabularDataset, expected: Column | TabularDataset) -> float:
+    def mean_squared_error(
+        predicted: Column | TabularDataset | TimeSeriesDataset,
+        expected: Column | TabularDataset | TimeSeriesDataset,
+    ) -> float:
         """
         Compute the mean squared error (MSE) on the given data.
 
@@ -201,10 +244,27 @@ def mean_squared_error(predicted: Column | TabularDataset, expected: Column | Ta
         if expected.row_count == 0:
             return 0.0  # Everything was predicted correctly (since there is nothing to predict)
 
+        # For TimeSeries Predictions, where the output is a list of values.
+        # Expected results are internally converted to a column containing multiple Columns for each prediction window
+        # Currently only used in fit_by_exhaustive_search, where prediction metrics have to be calculated internally.
+        if isinstance(expected.get_value(0), Column):
+            sum_of_mean_squared_errors = 0.0
+            for i in range(expected.row_count):
+                predicted_row_as_col: Column = Column("predicted", predicted[i])
+                expected_row_as_col = expected.get_value(i)
+                sum_of_mean_squared_errors += RegressionMetrics.mean_squared_error(
+                    predicted_row_as_col,
+                    expected_row_as_col,
+                )
+            return sum_of_mean_squared_errors / expected.row_count
+
         return (expected._series - predicted._series).pow(2).mean()
 
     @staticmethod
-    def median_absolute_deviation(predicted: Column | TabularDataset, expected: Column | TabularDataset) -> float:
+    def median_absolute_deviation(
+        predicted: Column | TabularDataset | TimeSeriesDataset,
+        expected: Column | TabularDataset | TimeSeriesDataset,
+    ) -> float:
         """
         Compute the median absolute deviation (MAD) on the given data.
 
@@ -231,12 +291,26 @@ def median_absolute_deviation(predicted: Column | TabularDataset, expected: Colu
         if expected.row_count == 0:
             return 0.0
 
+        # For TimeSeries Predictions, where the output is a list of values.
+        # Expected results are internally converted to a column containing multiple Columns for each prediction window
+        # Currently only used in fit_by_exhaustive_search, where prediction metrics have to be calculated internally.
+        if isinstance(expected.get_value(0), Column):
+            sum_of_median_absolute_deviation = 0.0
+            for i in range(expected.row_count):
+                predicted_row_as_col: Column = Column("predicted", predicted[i])
+                expected_row_as_col = expected.get_value(i)
+                sum_of_median_absolute_deviation += RegressionMetrics.median_absolute_deviation(
+                    predicted_row_as_col,
+                    expected_row_as_col,
+                )
+            return sum_of_median_absolute_deviation / expected.row_count
+
         return (expected._series - predicted._series).abs().median()
 
 
-def _extract_target(column_or_dataset: Column | TabularDataset) -> Column:
+def _extract_target(column_or_dataset: Column | TabularDataset | TimeSeriesDataset) -> Column:
     """Extract the target column from the given column or dataset."""
-    if isinstance(column_or_dataset, TabularDataset):
+    if isinstance(column_or_dataset, TabularDataset | TimeSeriesDataset):
         return column_or_dataset.target
     else:
         return column_or_dataset