[ENH] remove unneccesary conversion in TSFreshFeatureExtractor (#4571)

This PR:

* removes an unneccesary conversion in `TSFreshFeatureExtractor`, it
would convert first to `nested_univ` then to `pd-long`. This is
unnecessary, it should convert directly to `pd-long`.
* adds a direct converter from `pd-multiindex` to `pd-long`, removing
the need to rely on the dodgy `nested_univ` data format.

Depends on #4573

sktime/datatypes/_panel/_convert.py

@@ -569,6 +569,30 @@ def from_long_to_nested_adp(obj, store=None):
 convert_dict[("pd-long", "nested_univ", "Panel")] = from_nested_to_long_adp
 
 
+def from_multiindex_to_long(obj, store=None):
+
+    X_long = pd.melt(obj, value_vars=obj.columns, ignore_index=False)
+    X_long = X_long.reset_index()
+
+    return X_long
+
+
+convert_dict[("pd-multiindex", "pd-long", "Panel")] = from_multiindex_to_long
+
+
+def from_long_to_multiindex(obj, store=None):
+
+    ixcols = obj.columns[[0, 1]]
+    Xmi = pd.pivot(obj, columns=obj.columns[2], values=obj.columns[3], index=ixcols)
+
+    Xmi.columns.name = None
+
+    return Xmi
+
+
+convert_dict[("pd-long", "pd-multiindex", "Panel")] = from_long_to_multiindex
+
+
 def from_multi_index_to_3d_numpy(X):
     """Convert pandas multi-index Panel to numpy 3D Panel.
 
@@ -1097,6 +1121,10 @@ def from_numpyflat_to_numpy3d(obj, store=None):
     "numpyflat", "numpy3D", convert_dict, mtype_universe=MTYPE_LIST_PANEL
 )
 
+_extend_conversions(
+    "pd-long", "pd-multiindex", convert_dict, mtype_universe=MTYPE_LIST_PANEL
+)
+
 
 if _check_soft_dependencies("dask", severity="none"):
     from sktime.datatypes._adapter.dask_to_pd import (

sktime/transformations/panel/tsfresh.py

@@ -5,9 +5,6 @@
 __author__ = ["AyushmaanSeth", "mloning", "alwinw", "MatthewMiddlehurst"]
 __all__ = ["TSFreshFeatureExtractor", "TSFreshRelevantFeatureExtractor"]
 
-from warnings import warn
-
-from sktime.datatypes._panel._convert import from_nested_to_long
 from sktime.transformations.base import BaseTransformer
 from sktime.utils.validation import check_n_jobs
 
@@ -25,7 +22,6 @@ class _TSFreshFeatureExtractor(BaseTransformer):
         "y_inner_mtype": "None",  # which mtypes do _fit/_predict support for X?
         "fit_is_empty": True,  # is fit empty and can be skipped? Yes = True
         "python_dependencies": "tsfresh",
-        "python_version": "<3.10",
     }
 
     def __init__(
@@ -229,6 +225,8 @@ class TSFreshFeatureExtractor(_TSFreshFeatureExtractor):
     >>> X_transform2 = ts_custom.fit_transform(X_train) # doctest: +SKIP
     """
 
+    _tags = {"X_inner_mtype": "pd-long"}
+
     def __init__(
         self,
         default_fc_parameters="efficient",
@@ -275,34 +273,24 @@ def _transform(self, X, y=None):
             each cell of Xt contains pandas.Series
             transformed version of X
         """
-        # tsfresh requires unique index, returns only values for
-        # unique index values
-        if X.index.nunique() < X.shape[0]:
-            warn(
-                "tsfresh requires a unique index, but found "
-                "non-unique. To avoid this warning, please make sure the index of X "
-                "contains only unique values."
-            )
-            X = X.reset_index(drop=True)
-
-        Xt = from_nested_to_long(X)
-
         # lazy imports to avoid hard dependency
         from tsfresh import extract_features
 
         Xt = extract_features(
-            Xt,
-            column_id="index",
-            column_value="value",
-            column_kind="column",
-            column_sort="time_index",
+            X,
+            column_id=X.columns[0],
+            column_value=X.columns[3],
+            column_kind=X.columns[2],
+            column_sort=X.columns[1],
             **self.default_fc_parameters_,
         )
 
         # When using the long input format, tsfresh seems to sort the index,
         # here we make sure we return the dataframe in the sort order as the
         # input data
-        return Xt.reindex(X.index)
+        instances = X.iloc[:, 0].unique()
+        Xt = Xt.reindex(instances)
+        return Xt
 
     @classmethod
     def get_test_params(cls, parameter_set="default"):