[BUG] fix predict output conversion failure in BaseClassifier, BaseRegressor, if y_inner_mtype tag is a list

sktime/base/_base_panel.py

@@ -265,22 +265,29 @@ def _convert_X(self, X, X_mtype):
         )
         return X
 
-    def _check_y(self, y=None):
+    def _check_y(self, y=None, return_to_mtype=False):
         """Check and coerce X/y for fit/transform functions.
 
         Parameters
         ----------
         y : pd.DataFrame, pd.Series or np.ndarray
+        return_to_mtype : bool
+            whether to return the mtype of y output
 
         Returns
         -------
-        y : object of sktime compatible time series type
+        y_inner : object of sktime compatible time series type
             can be Series, Panel, Hierarchical
         y_metadata : dict
             metadata of y, retured by check_is_scitype
+        y_mtype : str, only returned if return_to_mtype=True
+            mtype of y_inner, after convert
         """
         if y is None:
-            return None
+            if return_to_mtype:
+                return None, None, None
+            else:
+                return None, None
 
         capa_multioutput = self.get_tag("capability:multioutput")
         y_inner_mtype = self.get_tag("y_inner_mtype")
@@ -314,24 +321,32 @@ def _check_y(self, y=None):
                 store=self._converter_store_y,
             )
             y_vec = VectorizedDF([y_df], iterate_cols=True)
-            return y_vec, y_metadata
+            if return_to_mtype:
+                return y_vec, y_metadata, "pd_DataFrame_Table"
+            else:
+                return y_vec, y_metadata
 
-        y_inner = convert(
+        y_inner, y_inner_mtype = convert(
             y,
             from_type=y_mtype,
             to_type=y_inner_mtype,
             as_scitype="Table",
             store=self._converter_store_y,
+            return_to_mtype=True,
         )
 
-        return y_inner, y_metadata
+        if return_to_mtype:
+            return y_inner, y_metadata, y_inner_mtype
+        else:
+            return y_inner, y_metadata
 
     def _convert_output_y(self, y):
         """Convert output y to original format.
 
         Parameters
         ----------
         y : np.ndarray or pd.DataFrame
+            output to convert
 
         Returns
         -------
@@ -346,9 +361,18 @@ def _convert_output_y(self, y):
             output_mtype = "numpy1D"
             converter_store = None
 
+        # inner return mtype is what we convert from
+        # special treatment for 1D numpy array
+        # this can be returned in composites due to
+        # current downwards compatible choice "1D return is always numpy"
+        if isinstance(y, np.ndarray) and y.ndim == 1:
+            inner_return_mtype = "numpy1D"
+        else:
+            inner_return_mtype = self._y_inner_mtype
+
         y = convert(
             y,
-            from_type=self.get_tag("y_inner_mtype"),
+            from_type=inner_return_mtype,
             to_type=output_mtype,
             as_scitype="Table",
             store=converter_store,

sktime/classification/base.py

@@ -180,8 +180,9 @@ class labels for fitting
         start = int(round(time.time() * 1000))
 
         # check and convert y for multioutput vectorization
-        y, y_metadata = self._check_y(y)
+        y, y_metadata, y_inner_mtype = self._check_y(y, return_to_mtype=True)
         self._y_metadata = y_metadata
+        self._y_inner_mtype = y_inner_mtype
         self._is_vectorized = isinstance(y, VectorizedDF)
 
         if self._is_vectorized:

sktime/classification/dictionary_based/_boss.py

@@ -667,6 +667,7 @@ def _shorten_bags(self, word_len, y):
         new_boss.classes_ = self.classes_
         new_boss._class_dictionary = self._class_dictionary
         new_boss._y_metadata = self._y_metadata
+        new_boss._y_inner_mtype = self._y_inner_mtype
         new_boss._is_fitted = True
 
         return new_boss

sktime/classification/early_classification/base.py

@@ -630,20 +630,26 @@ def _convert_X(self, X, X_mtype):
         _convert_X = BaseClassifier._convert_X
         return _convert_X(self, X, X_mtype)
 
-    def _check_y(self, y=None):
+    def _check_y(self, y=None, return_to_mtype=False):
         """Check and coerce X/y for fit/transform functions.
 
         Parameters
         ----------
         y : pd.DataFrame, pd.Series or np.ndarray
+        return_to_mtype : bool
+            whether to return the mtype of y output
 
         Returns
         -------
-        y : object of sktime compatible time series type
+        y_inner : object of sktime compatible time series type
             can be Series, Panel, Hierarchical
+        y_metadata : dict
+            metadata of y, retured by check_is_scitype
+        y_mtype : str, only returned if return_to_mtype=True
+            mtype of y_inner, after convert
         """
         _check_y = BaseClassifier._check_y
-        return _check_y(self, y)
+        return _check_y(self, y, return_to_mtype=return_to_mtype)
 
     def _check_input(self, X, y=None, enforce_min_instances=1, return_metadata=True):
         """Check whether input X and y are valid formats with minimum data.

sktime/datatypes/_convert.py

@@ -301,8 +301,8 @@ def _get_first_mtype_of_same_scitype(from_mtype, to_mtypes, varname="to_mtypes")
     ]
     if len(same_scitype_mtypes) == 0:
         raise TypeError(
-            f"{varname} contains no mtype compatible with the scitype of obj,"
-            f"which is {scitype}"
+            f"{varname} contains no mtype compatible with the scitype of obj, "
+            f"which is {scitype}. Value of {varname} is: {to_mtypes}"
         )
     to_type = same_scitype_mtypes[0]
     return to_type

sktime/regression/base.py

@@ -174,8 +174,9 @@ class labels for fitting
         start = int(round(time.time() * 1000))
 
         # check and convert y for multioutput vectorization
-        y, y_metadata = self._check_y(y)
+        y, y_metadata, y_inner_mtype = self._check_y(y, return_to_mtype=True)
         self._y_metadata = y_metadata
+        self._y_inner_mtype = y_inner_mtype
         self._is_vectorized = isinstance(y, VectorizedDF)
 
         if self._is_vectorized: