FIX-#4438: Fix reindex function that doesn't preserve initial index metadata (#4442)

Signed-off-by: Anatoly Myachev <[email protected]>

Co-authored-by: Dmitry Chigarev <[email protected]>

Author: anmyachev

docs/release_notes/release_notes-0.15.0.rst

@@ -16,6 +16,7 @@ Key Features and Updates
   * FIX-#4407: Align `insert` function with pandas in case of numpy array with several columns (#4408)
   * FIX-#4373: Fix invalid file path when trying `read_csv_glob` with `usecols` parameter (#4405)
   * FIX-#4394: Fix issue with multiindex metadata desync (#4395)
+  * FIX-#4438: Fix `reindex` function that doesn't preserve initial index metadata (#4442)
   * FIX-#4425: Add parameters to groupby pct_change (#4429)
   * FIX-#4457: Fix `loc` in case when need reindex item (#4457)
   * FIX-#4414: Add missing f prefix on f-strings found at https://codereview.doctor (#4415)

modin/pandas/base.py

@@ -26,6 +26,7 @@
     is_dtype_equal,
     is_object_dtype,
 )
+from pandas.core.indexes.api import ensure_index
 import pandas.core.window.rolling
 import pandas.core.resample
 import pandas.core.generic
@@ -2220,6 +2221,34 @@ def rank(
             )
         )
 
+    def _copy_index_metadata(self, source, destination):  # noqa: PR01, RT01, D200
+        """
+        Copy Index metadata from `source` to `destination` inplace.
+        """
+        if hasattr(source, "name") and hasattr(destination, "name"):
+            destination.name = source.name
+        if hasattr(source, "names") and hasattr(destination, "names"):
+            destination.names = source.names
+        return destination
+
+    def _ensure_index(self, index_like, axis=0):  # noqa: PR01, RT01, D200
+        """
+        Ensure that we have an index from some index-like object.
+        """
+        if (
+            self._query_compiler.has_multiindex(axis=axis)
+            and not isinstance(index_like, pandas.Index)
+            and is_list_like(index_like)
+            and len(index_like) > 0
+            and isinstance(index_like[0], tuple)
+        ):
+            try:
+                return pandas.MultiIndex.from_tuples(index_like)
+            except TypeError:
+                # not all tuples
+                pass
+        return ensure_index(index_like)
+
     def reindex(
         self,
         index=None,
@@ -2243,7 +2272,9 @@ def reindex(
         new_query_compiler = None
         if index is not None:
             if not isinstance(index, pandas.Index):
-                index = pandas.Index(index)
+                index = self._copy_index_metadata(
+                    source=self.index, destination=self._ensure_index(index, axis=0)
+                )
             if not index.equals(self.index):
                 new_query_compiler = self._query_compiler.reindex(
                     axis=0, labels=index, **kwargs
@@ -2253,7 +2284,9 @@ def reindex(
         final_query_compiler = None
         if columns is not None:
             if not isinstance(columns, pandas.Index):
-                columns = pandas.Index(columns)
+                columns = self._copy_index_metadata(
+                    source=self.columns, destination=self._ensure_index(columns, axis=1)
+                )
             if not columns.equals(self.columns):
                 final_query_compiler = new_query_compiler.reindex(
                     axis=1, labels=columns, **kwargs

modin/pandas/test/dataframe/test_indexing.py

@@ -768,6 +768,51 @@ def test_reindex():
     )
 
 
+def test_reindex_4438():
+    index = pd.date_range(end="1/1/2018", periods=3, freq="h", name="some meta")
+    new_index = list(reversed(index))
+
+    # index case
+    modin_df = pd.DataFrame([1, 2, 3], index=index)
+    pandas_df = pandas.DataFrame([1, 2, 3], index=index)
+    new_modin_df = modin_df.reindex(new_index)
+    new_pandas_df = pandas_df.reindex(new_index)
+    df_equals(new_modin_df, new_pandas_df)
+
+    # column case
+    modin_df = pd.DataFrame(np.array([[1], [2], [3]]).T, columns=index)
+    pandas_df = pandas.DataFrame(np.array([[1], [2], [3]]).T, columns=index)
+    new_modin_df = modin_df.reindex(columns=new_index)
+    new_pandas_df = pandas_df.reindex(columns=new_index)
+    df_equals(new_modin_df, new_pandas_df)
+
+    # multiindex case
+    multi_index = pandas.MultiIndex.from_arrays(
+        [("a", "b", "c"), ("a", "b", "c")], names=["first", "second"]
+    )
+    new_multi_index = list(reversed(multi_index))
+
+    modin_df = pd.DataFrame([1, 2, 3], index=multi_index)
+    pandas_df = pandas.DataFrame([1, 2, 3], index=multi_index)
+    new_modin_df = modin_df.reindex(new_multi_index)
+    new_pandas_df = pandas_df.reindex(new_multi_index)
+    df_equals(new_modin_df, new_pandas_df)
+
+    # multicolumn case
+    modin_df = pd.DataFrame(np.array([[1], [2], [3]]).T, columns=multi_index)
+    pandas_df = pandas.DataFrame(np.array([[1], [2], [3]]).T, columns=multi_index)
+    new_modin_df = modin_df.reindex(columns=new_multi_index)
+    new_pandas_df = pandas_df.reindex(columns=new_multi_index)
+    df_equals(new_modin_df, new_pandas_df)
+
+    # index + multiindex case
+    modin_df = pd.DataFrame([1, 2, 3], index=index)
+    pandas_df = pandas.DataFrame([1, 2, 3], index=index)
+    new_modin_df = modin_df.reindex(new_multi_index)
+    new_pandas_df = pandas_df.reindex(new_multi_index)
+    df_equals(new_modin_df, new_pandas_df)
+
+
 def test_reindex_like():
     df1 = pd.DataFrame(
         [