[ENH] Standardise check_y usage via separate _convert_y method

aeon/anomaly_detection/collection/base.py

@@ -109,7 +109,8 @@ def fit(self, X, y=None):
 
         X = self._preprocess_collection(X)
         if y is not None:
-            y = self._check_y(y, self.metadata_["n_cases"])
+            self._check_y(y, self.metadata_["n_cases"])
+            y = self._convert_y(y)
 
         self._fit(X, y)
 
@@ -208,7 +209,8 @@ def fit_predict(self, X, y=None, axis=1) -> np.ndarray:
             return self._predict(X)
 
         if y is not None:
-            y = self._check_y(y)
+            self._check_y(y, self.metadata_["n_cases"])
+            y = self._convert_y(y)
 
         pred = self._fit_predict(X, y)
 
@@ -251,7 +253,10 @@ def _check_y(self, y, n_cases):
                 f"Mismatch in number of cases. Found X = {n_cases} and y = {n_labels}"
             )
 
+        return y
+
+    def _convert_y(self, y):
+        """Convert y to correct anomaly detection format after validation."""
         if isinstance(y, pd.Series):
             y = pd.Series.to_numpy(y)
-
-        return y
+        return y.astype(bool)

aeon/anomaly_detection/series/base.py

@@ -123,7 +123,8 @@ def fit(self, X, y=None, axis=1):
 
         X = self._preprocess_series(X, axis, True)
         if y is not None:
-            y = self._check_y(y)
+            self._check_y(y)
+            y = self._convert_y(y)
 
         self._fit(X=X, y=y)
 
@@ -203,7 +204,8 @@ def fit_predict(self, X, y=None, axis=1) -> np.ndarray:
             return self._predict(X)
 
         if y is not None:
-            y = self._check_y(y)
+            self._check_y(y)
+            y = self._convert_y(y)
 
         pred = self._fit_predict(X, y)
 
@@ -284,5 +286,12 @@ def _check_y(self, y: VALID_SERIES_INPUT_TYPES) -> np.ndarray:
                 f"{VALID_SERIES_INPUT_TYPES}, saw {type(y)}"
             )
 
-        new_y = new_y.astype(bool)
         return new_y
+
+    def _convert_y(self, y):
+        """Convert y to correct anomaly detection format after validation."""
+        if isinstance(y, pd.Series):
+            y = y.values
+        elif isinstance(y, pd.DataFrame):
+            y = y.squeeze().values
+        return y.astype(bool)

aeon/regression/base.py

@@ -231,7 +231,8 @@ def score(self, X, y, metric="r2", metric_params=None) -> float:
             MSE score of predict(X) vs y
         """
         self._check_is_fitted()
-        y = self._check_y(y, len(X))
+        self._check_y(y, len(X))
+        y = self._convert_y(y)
         _metric_params = metric_params
         if metric_params is None:
             _metric_params = {}
@@ -351,7 +352,8 @@ def _fit_setup(self, X, y):
         self.reset()
 
         X = self._preprocess_collection(X)
-        y = self._check_y(y, self.metadata_["n_cases"])
+        self._check_y(y, self.metadata_["n_cases"])
+        y = self._convert_y(y)
 
         # return processed X and y
         return X, y
@@ -380,13 +382,16 @@ def _check_y(self, y, n_cases):
                 f"sklearn.utils.multiclass.type_of_target"
             )
 
-        if isinstance(y, pd.Series):
-            y = pd.Series.to_numpy(y)
-
         if any([isinstance(label, str) for label in y]):
             raise ValueError(
                 "y contains strings, cannot fit a regressor. If suitable, convert "
                 "to floats or consider classification."
             )
 
+        return y
+
+    def _convert_y(self, y):
+        """Convert y to the correct regression format after validation."""
+        if isinstance(y, pd.Series):
+            y = pd.Series.to_numpy(y)
         return y.astype(float)

aeon/regression/tests/test_base.py

@@ -98,7 +98,8 @@ def test__check_y():
     reg._check_y(y, 100)
     assert isinstance(y, np.ndarray)
     y = pd.Series(y)
-    y = reg._check_y(y, 100)
+    reg._check_y(y, 100)
+    y = reg._convert_y(y)
     assert isinstance(y, np.ndarray)
     # Test error raising
     # y wrong length