Make Frame._dtype an iterator instead of a dict (#15920)

A lot of the usages of `Frame._dtype` didn't require the previous `dict` return type since that was just re-iterated over anyways.

Also removed a redundant `tuple` call in `Frame._column_names` and `Frame._columns`

Authors:
  - Matthew Roeschke (https://github.com/mroeschke)

Approvers:
  - Lawrence Mitchell (https://github.com/wence-)

URL: #15920

python/cudf/cudf/core/dataframe.py

@@ -1231,7 +1231,7 @@ def dtypes(self):
         string              object
         dtype: object
         """
-        return pd.Series(self._dtypes, dtype="object")
+        return pd.Series(dict(self._dtypes), dtype="object")
 
     @property
     def ndim(self) -> int:
@@ -2834,7 +2834,7 @@ def reindex(
 
         return df._reindex(
             column_names=columns,
-            dtypes=self._dtypes,
+            dtypes=dict(self._dtypes),
             deep=copy,
             index=index,
             inplace=False,

python/cudf/cudf/core/frame.py

@@ -79,18 +79,16 @@ def _num_rows(self) -> int:
         return self._data.nrows
 
     @property
-    def _column_names(self) -> Tuple[Any, ...]:  # TODO: Tuple[str]?
-        return tuple(self._data.names)
+    def _column_names(self) -> Tuple[Any, ...]:
+        return self._data.names
 
     @property
-    def _columns(self) -> Tuple[Any, ...]:  # TODO: Tuple[Column]?
-        return tuple(self._data.columns)
+    def _columns(self) -> Tuple[ColumnBase, ...]:
+        return self._data.columns
 
     @property
-    def _dtypes(self):
-        return dict(
-            zip(self._data.names, (col.dtype for col in self._data.columns))
-        )
+    def _dtypes(self) -> abc.Iterator:
+        return zip(self._data.names, (col.dtype for col in self._data.columns))
 
     @property
     def ndim(self) -> int:
@@ -1969,7 +1967,7 @@ def __dask_tokenize__(self):
 
         return [
             type(self),
-            str(self._dtypes),
+            str(dict(self._dtypes)),
             normalize_token(self.to_pandas()),
         ]
 

python/cudf/cudf/core/groupby/groupby.py

@@ -22,12 +22,7 @@
 from cudf._lib.types import size_type_dtype
 from cudf._typing import AggType, DataFrameOrSeries, MultiColumnAggType
 from cudf.api.extensions import no_default
-from cudf.api.types import (
-    is_bool_dtype,
-    is_float_dtype,
-    is_list_like,
-    is_numeric_dtype,
-)
+from cudf.api.types import is_bool_dtype, is_list_like, is_numeric_dtype
 from cudf.core._compat import PANDAS_LT_300
 from cudf.core.abc import Serializable
 from cudf.core.column.column import ColumnBase, StructDtype, as_column
@@ -335,12 +330,8 @@ def dtypes(self):
             FutureWarning,
         )
         index = self.grouping.keys.unique().sort_values().to_pandas()
-        obj_dtypes = self.obj._dtypes
         return pd.DataFrame(
-            {
-                name: [obj_dtypes[name]] * len(index)
-                for name in self.obj._data.names
-            },
+            {name: [dtype] * len(index) for name, dtype in self.obj._dtypes},
             index=index,
         )
 
@@ -499,8 +490,7 @@ def rank(
         # treats NaNs the way we treat nulls.
         if cudf.get_option("mode.pandas_compatible"):
             if any(
-                is_float_dtype(typ)
-                for typ in self.grouping.values._dtypes.values()
+                col.dtype.kind == "f" for col in self.grouping.values._columns
             ):
                 raise NotImplementedError(
                     "NaNs are not supported in groupby.rank."

python/cudf/cudf/core/indexed_frame.py

@@ -891,7 +891,7 @@ def replace(
             ) = _get_replacement_values_for_columns(
                 to_replace=to_replace,
                 value=value,
-                columns_dtype_map=self._dtypes,
+                columns_dtype_map=dict(self._dtypes),
             )
 
             for name, col in self._data.items():
@@ -6313,11 +6313,11 @@ def __dask_tokenize__(self):
 
         return [
             type(self),
-            str(self._dtypes),
+            str(dict(self._dtypes)),
             *[
-                normalize_token(cat.categories)
-                for cat in self._dtypes.values()
-                if cat == "category"
+                normalize_token(col.dtype.categories)
+                for col in self._columns
+                if col.dtype == "category"
             ],
             normalize_token(self.index),
             normalize_token(self.hash_values().values_host),

python/cudf/cudf/io/csv.py

@@ -132,10 +132,9 @@ def read_csv(
         # There exists some dtypes in the result columns that is inferred.
         # Find them and map them to the default dtypes.
         specified_dtypes = {} if dtype is None else dtype
-        df_dtypes = df._dtypes
         unspecified_dtypes = {
-            name: df_dtypes[name]
-            for name in df._column_names
+            name: dtype
+            for name, dtype in df._dtypes
             if name not in specified_dtypes
         }
         default_dtypes = {}

python/cudf/cudf/io/json.py

@@ -147,10 +147,9 @@ def read_json(
         # There exists some dtypes in the result columns that is inferred.
         # Find them and map them to the default dtypes.
         specified_dtypes = {} if dtype is True else dtype
-        df_dtypes = df._dtypes
         unspecified_dtypes = {
-            name: df_dtypes[name]
-            for name in df._column_names
+            name: dtype
+            for name, dtype in df._dtypes
             if name not in specified_dtypes
         }
         default_dtypes = {}