Remove or split up Frame methods that use the index

python/cudf/cudf/core/column_accessor.py

@@ -342,6 +342,26 @@ def select_by_label(self, key: Any) -> ColumnAccessor:
                     return self._select_by_label_with_wildcard(key)
             return self._select_by_label_grouped(key)
 
+    def get_labels_by_index(self, index: Any) -> tuple:
+        """Get the labels corresponding to the provided column indices.
+
+        Parameters
+        ----------
+        index : integer, integer slice, or list-like of integers
+            The column indexes.
+
+        Returns
+        -------
+        tuple
+        """
+        if isinstance(index, slice):
+            start, stop, step = index.indices(len(self._data))
+            return self.names[start:stop:step]
+        elif pd.api.types.is_integer(index):
+            return (self.names[index],)
+        else:
+            return tuple(self.names[i] for i in index)
+
     def select_by_index(self, index: Any) -> ColumnAccessor:
         """
         Return a ColumnAccessor composed of the columns
@@ -355,13 +375,7 @@ def select_by_index(self, index: Any) -> ColumnAccessor:
         -------
         ColumnAccessor
         """
-        if isinstance(index, slice):
-            start, stop, step = index.indices(len(self._data))
-            keys = self.names[start:stop:step]
-        elif pd.api.types.is_integer(index):
-            keys = (self.names[index],)
-        else:
-            keys = tuple(self.names[i] for i in index)
+        keys = self.get_labels_by_index(index)
         data = {k: self._data[k] for k in keys}
         return self.__class__(
             data, multiindex=self.multiindex, level_names=self.level_names,

python/cudf/cudf/core/dataframe.py

@@ -372,9 +372,9 @@ class _DataFrameIlocIndexer(_DataFrameIndexer):
     def _getitem_tuple_arg(self, arg):
         # Iloc Step 1:
         # Gather the columns specified by the second tuple arg
-        columns_df = self._frame._get_columns_by_index(arg[1])
-
-        columns_df._index = self._frame._index
+        columns_df = self._frame._from_data(
+            self._frame._data.select_by_index(arg[1]), self._frame._index
+        )
 
         # Iloc Step 2:
         # Gather the rows specified by the first tuple arg
@@ -422,9 +422,9 @@ def _getitem_tuple_arg(self, arg):
 
     @_cudf_nvtx_annotate
     def _setitem_tuple_arg(self, key, value):
-        columns = self._frame._get_columns_by_index(key[1])
-
-        for col in columns:
+        # TODO: Determine if this usage is prevalent enough to expose this
+        # selection logic at a higher level than ColumnAccessor.
+        for col in self._frame._data.get_labels_by_index(key[1]):
             self._frame[col].iloc[key[0]] = value
 
     def _getitem_scalar(self, arg):
@@ -612,7 +612,8 @@ def __init__(
                 new_df = self._from_arrays(data, index=index, columns=columns)
 
             self._data = new_df._data
-            self.index = new_df._index
+            self._index = new_df._index
+            self._check_data_index_length_match()
         elif hasattr(data, "__array_interface__"):
             arr_interface = data.__array_interface__
             if len(arr_interface["descr"]) == 1:
@@ -621,7 +622,8 @@ def __init__(
             else:
                 new_df = self.from_records(data, index=index, columns=columns)
             self._data = new_df._data
-            self.index = new_df._index
+            self._index = new_df._index
+            self._check_data_index_length_match()
         else:
             if is_list_like(data):
                 if len(data) > 0 and is_scalar(data[0]):
@@ -632,7 +634,8 @@ def __init__(
                     new_df = DataFrame(data=data, index=index)
 
                     self._data = new_df._data
-                    self.index = new_df._index
+                    self._index = new_df._index
+                    self._check_data_index_length_match()
                 elif len(data) > 0 and isinstance(data[0], Series):
                     self._init_from_series_list(
                         data=data, columns=columns, index=index
@@ -653,6 +656,15 @@ def __init__(
         if dtype:
             self._data = self.astype(dtype)._data
 
+    def _check_data_index_length_match(df: DataFrame) -> None:
+        # Validate that the number of rows in the data matches the index if the
+        # data is not empty. This is a helper for the constructor.
+        if df._data.nrows > 0 and df._data.nrows != len(df._index):
+            raise ValueError(
+                f"Shape of passed values is {df.shape}, indices imply "
+                f"({len(df._index)}, {df._num_columns})"
+            )
+
     @_cudf_nvtx_annotate
     def _init_from_series_list(self, data, columns, index):
         if index is None:
@@ -856,9 +868,7 @@ def _from_data(
         index: Optional[BaseIndex] = None,
         columns: Any = None,
     ) -> DataFrame:
-        out = super()._from_data(data, index)
-        if index is None:
-            out.index = RangeIndex(out._data.nrows)
+        out = super()._from_data(data=data, index=index)
         if columns is not None:
             out.columns = columns
         return out
@@ -5601,7 +5611,9 @@ def stack(self, level=-1, dropna=True):
         """
         assert level in (None, -1)
         repeated_index = self.index.repeat(self.shape[1])
-        name_index = Frame({0: self._column_names}).tile(self.shape[0])
+        name_index = cudf.DataFrame._from_data({0: self._column_names}).tile(
+            self.shape[0]
+        )
         new_index = list(repeated_index._columns) + [name_index._columns[0]]
         if isinstance(self._index, MultiIndex):
             index_names = self._index.names + [None]