maxframe 2.2.0__cp312-cp312-win32.whl

maxframe/dataframe/datasource/read_odps_table.py

@@ -0,0 +1,273 @@
+# Copyright 1999-2025 Alibaba Group Holding Ltd.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import logging
+from typing import List, MutableMapping, Optional, Union
+
+import numpy as np
+import pandas as pd
+from odps import ODPS
+from odps.models import Table
+from odps.utils import to_timestamp
+
+from ... import opcodes
+from ...config import options
+from ...core import OutputType
+from ...io.odpsio import odps_schema_to_pandas_dtypes
+from ...serialization.serializables import (
+    AnyField,
+    BoolField,
+    FieldTypes,
+    Int64Field,
+    ListField,
+    SeriesField,
+    StringField,
+)
+from ...utils import estimate_table_size, is_empty
+from ..core import DataFrame  # noqa: F401
+from ..utils import parse_index
+from .core import ColumnPruneSupportedDataSourceMixin, IncrementalIndexDatasource
+
+logger = logging.getLogger(__name__)
+
+
+class DataFrameReadODPSTable(
+    IncrementalIndexDatasource,
+    ColumnPruneSupportedDataSourceMixin,
+):
+    __slots__ = ("_odps_entry",)
+    _op_type_ = opcodes.READ_ODPS_TABLE
+
+    table_name = StringField("table_name")
+    partitions = ListField("partitions", field_type=FieldTypes.string, default=None)
+    dtypes = SeriesField("dtypes", default=None)
+    columns = AnyField("columns", default=None)
+    nrows = Int64Field("nrows", default=None)
+    use_arrow_dtype = BoolField("use_arrow_dtype", default=None)
+    string_as_binary = BoolField("string_as_binary", default=None)
+    append_partitions = BoolField("append_partitions", default=None)
+    last_modified_time = Int64Field("last_modified_time", default=None)
+    index_columns = ListField("index_columns", FieldTypes.string, default=None)
+    index_dtypes = SeriesField("index_dtypes", default=None)
+
+    def __init__(self, memory_scale=None, **kw):
+        output_type = kw.pop("output_type", OutputType.dataframe)
+        self._odps_entry = kw.pop("odps_entry", None)
+        super(DataFrameReadODPSTable, self).__init__(
+            memory_scale=memory_scale, _output_types=[output_type], **kw
+        )
+
+    @property
+    def partition(self):
+        return getattr(self, "partition_spec", None)
+
+    def get_columns(self):
+        return self.columns or list(self.dtypes.index)
+
+    def set_pruned_columns(self, columns, *, keep_order=None):  # pragma: no cover
+        self.columns = columns
+
+    def __call__(self, shape, chunk_bytes=None, chunk_size=None):
+        if is_empty(self.index_columns):
+            if np.isnan(shape[0]):
+                index_value = parse_index(pd.RangeIndex(0))
+            else:
+                index_value = parse_index(pd.RangeIndex(shape[0]))
+        elif len(self.index_columns) == 1:
+            index_value = parse_index(
+                pd.Index([], name=self.index_columns[0]).astype(
+                    self.index_dtypes.iloc[0]
+                )
+            )
+        else:
+            idx = pd.MultiIndex.from_frame(
+                pd.DataFrame([], columns=self.index_columns).astype(self.index_dtypes)
+            )
+            index_value = parse_index(idx)
+
+        if self.output_types[0] == OutputType.dataframe:
+            columns_value = parse_index(self.dtypes.index, store_data=True)
+            return self.new_tileable(
+                [],
+                None,
+                shape=shape,
+                dtypes=self.dtypes,
+                index_value=index_value,
+                columns_value=columns_value,
+                chunk_bytes=chunk_bytes,
+                chunk_size=chunk_size,
+            )
+        elif self.output_types[0] == OutputType.series:
+            return self.new_tileable(
+                [],
+                None,
+                shape=shape[:1],
+                name=self.dtypes.index[0],
+                dtype=self.dtypes.iloc[0],
+                index_value=index_value,
+                chunk_bytes=chunk_bytes,
+                chunk_size=chunk_size,
+            )
+        elif self.output_types[0] == OutputType.index:
+            return self.new_tileable(
+                [],
+                None,
+                shape=shape[:1],
+                name=getattr(index_value, "name", None),
+                names=getattr(index_value, "names", None),
+                dtype=self.index_dtypes.iloc[0],
+                index_value=index_value,
+                chunk_bytes=chunk_bytes,
+                chunk_size=chunk_size,
+            )
+
+    @classmethod
+    def estimate_size(
+        cls, ctx: MutableMapping[str, Union[int, float]], op: "DataFrameReadODPSTable"
+    ) -> None:
+        odps_entry = op._odps_entry or ODPS.from_global() or ODPS.from_environments()
+        if not odps_entry:  # pragma: no cover
+            ctx[op.outputs[0].key] = float("inf")
+            return
+        ctx[op.outputs[0].key] = estimate_table_size(
+            odps_entry, op.table_name, op.partitions
+        )
+
+
+def read_odps_table(
+    table_name: Union[str, Table],
+    partitions: Union[None, str, List[str]] = None,
+    columns: Optional[List[str]] = None,
+    index_col: Union[None, str, List[str]] = None,
+    odps_entry: ODPS = None,
+    string_as_binary: bool = None,
+    append_partitions: bool = False,
+    **kw,
+):
+    """
+    Read data from a MaxCompute (ODPS) table into DataFrame.
+
+    Supports specifying some columns as indexes. If not specified, RangeIndex
+    will be generated.
+
+    Parameters
+    ----------
+    table_name: Union[str, Table]
+        Name of the table to read from.
+    partitions: Union[None, str, List[str]]
+        Table partition or list of partitions to read from.
+    columns: Optional[List[str]]
+        Table columns to read from. You may also specify partition columns here.
+        If not specified, all table columns (or include partition columns if
+        `append_partitions` is True) will be included.
+    index_col: Union[None, str, List[str]]
+        Columns to be specified as indexes.
+    append_partitions: bool
+        If True, will add all partition columns as selected columns when
+        `columns` is not specified,
+
+    Returns
+    -------
+    result: DataFrame
+        DataFrame read from MaxCompute (ODPS) table
+    """
+    odps_entry = odps_entry or ODPS.from_global() or ODPS.from_environments()
+    schema = options.session.default_schema or odps_entry.schema
+    if odps_entry is None:
+        raise ValueError("Missing odps_entry parameter")
+    if isinstance(table_name, Table):
+        table = table_name
+    else:
+        table = odps_entry.get_table(table_name, schema=schema)
+
+    if not table.table_schema.partitions and (
+        partitions is not None or append_partitions
+    ):
+        raise ValueError("Cannot handle partitions on non-partition tables")
+
+    cols = (
+        table.table_schema.columns
+        if append_partitions or columns is not None
+        else table.table_schema.simple_columns
+    )
+    table_columns = [c.name.lower() for c in cols]
+    table_dtypes = odps_schema_to_pandas_dtypes(
+        table.table_schema, with_partitions=True
+    )
+    df_types = [table_dtypes[c] for c in table_columns]
+
+    if isinstance(index_col, str):
+        index_col = [index_col]
+    if index_col and columns is None:
+        index_col_set = set([c.lower() for c in index_col])
+        col_diff = sorted(index_col_set - set(table_columns))
+        if col_diff:
+            raise ValueError(
+                f"Cannot found column {', '.join(col_diff)} specified "
+                f"in index_columns argument in source table"
+            )
+        columns = [c for c in table_columns if c not in index_col_set]
+
+    if not index_col:
+        index_dtypes = None
+    else:
+        table_index_types = [df_types[table_columns.index(col)] for col in index_col]
+        index_dtypes = pd.Series(table_index_types, index=index_col)
+
+    if columns is not None:
+        new_columns = [c.lower() for c in columns]
+        table_col_set = set(new_columns)
+        col_diff = sorted(table_col_set - set(table_columns))
+        if col_diff:
+            raise ValueError(
+                f"Cannot found column {', '.join(col_diff)} specified "
+                f"in columns argument in source table"
+            )
+        if any(col in table_col_set for col in index_col or ()):
+            raise ValueError("Index columns and columns shall not overlap.")
+
+        # reorder columns
+        df_types = [df_types[table_columns.index(col)] for col in new_columns]
+        table_columns = new_columns
+        columns = new_columns
+
+    shape = kw.pop("shape", None) or (np.nan, len(df_types))
+    dtypes = pd.Series(df_types, index=table_columns)
+    chunk_bytes = kw.pop("chunk_bytes", None)
+    chunk_size = kw.pop("chunk_size", None)
+    use_arrow_dtype = kw.pop("use_arrow_dtype", True)
+
+    partitions = partitions or kw.get("partition")
+    if isinstance(partitions, str):
+        partitions = [partitions]
+
+    append_partitions = append_partitions or any(
+        pt.name in (columns if not is_empty(columns) else ())
+        for pt in (table.table_schema.partitions or ())
+    )
+    op = DataFrameReadODPSTable(
+        table_name=table.full_table_name,
+        partitions=partitions,
+        dtypes=dtypes,
+        columns=columns,
+        use_arrow_dtype=use_arrow_dtype,
+        string_as_binary=string_as_binary,
+        append_partitions=append_partitions,
+        last_modified_time=to_timestamp(table.last_data_modified_time),
+        index_columns=index_col,
+        index_dtypes=index_dtypes,
+        odps_entry=odps_entry,
+        **kw,
+    )
+    return op(shape, chunk_bytes=chunk_bytes, chunk_size=chunk_size)

maxframe/dataframe/datasource/read_parquet.py

@@ -0,0 +1,426 @@
+# Copyright 1999-2025 Alibaba Group Holding Ltd.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+from typing import Dict, MutableMapping, Union
+from urllib.parse import urlparse
+
+import numpy as np
+import pandas as pd
+
+try:
+    import pyarrow as pa
+    import pyarrow.parquet as pq
+except ImportError:
+    pa = None
+
+try:
+    import fastparquet
+except ImportError:
+    fastparquet = None
+
+from ... import opcodes
+from ...config import options
+from ...lib.filesystem import FileSystem, get_fs, glob, open_file
+from ...serialization.serializables import (
+    AnyField,
+    BoolField,
+    DictField,
+    Int32Field,
+    Int64Field,
+    ListField,
+    StringField,
+)
+from ...utils import lazy_import
+from ..arrays import ArrowStringDtype
+from ..operators import OutputType
+from ..utils import parse_index, to_arrow_dtypes
+from .core import ColumnPruneSupportedDataSourceMixin, IncrementalIndexDatasource
+
+PARQUET_MEMORY_SCALE = 15
+STRING_FIELD_OVERHEAD = 50
+cudf = lazy_import("cudf")
+
+
+def check_engine(engine):
+    if engine == "auto":
+        if pa is not None:
+            return "pyarrow"
+        elif fastparquet is not None:  # pragma: no cover
+            return "fastparquet"
+        else:  # pragma: no cover
+            raise RuntimeError("Please install either pyarrow or fastparquet.")
+    elif engine == "pyarrow":
+        if pa is None:  # pragma: no cover
+            raise RuntimeError("Please install pyarrow first.")
+        return engine
+    elif engine == "fastparquet":
+        if fastparquet is None:  # pragma: no cover
+            raise RuntimeError("Please install fastparquet first.")
+        return engine
+    else:  # pragma: no cover
+        raise RuntimeError("Unsupported engine {} to read parquet.".format(engine))
+
+
+def get_engine(engine):
+    if engine == "pyarrow":
+        return ArrowEngine()
+    elif engine == "fastparquet":
+        return FastpaquetEngine()
+    else:  # pragma: no cover
+        raise RuntimeError("Unsupported engine {}".format(engine))
+
+
+class ParquetEngine:
+    def get_row_num(self, f):
+        raise NotImplementedError
+
+    def read_dtypes(self, f, **kwargs):
+        raise NotImplementedError
+
+    def read_to_pandas(
+        self, f, columns=None, nrows=None, use_arrow_dtype=None, **kwargs
+    ):
+        raise NotImplementedError
+
+    def read_group_to_pandas(
+        self, f, group_index, columns=None, nrows=None, use_arrow_dtype=None, **kwargs
+    ):
+        raise NotImplementedError
+
+    def read_partitioned_to_pandas(
+        self,
+        f,
+        partitions: Dict,
+        partition_keys: Dict,
+        columns=None,
+        nrows=None,
+        use_arrow_dtype=None,
+        **kwargs,
+    ):
+        raw_df = self.read_to_pandas(
+            f, columns=columns, nrows=nrows, use_arrow_dtype=use_arrow_dtype, **kwargs
+        )
+        for col, value in partition_keys.items():
+            dictionary = partitions[col]
+            raw_df[col] = pd.Series(
+                value,
+                dtype=pd.CategoricalDtype(categories=dictionary.tolist()),
+                index=raw_df.index,
+            )
+        return raw_df
+
+    def read_partitioned_dtypes(self, fs: FileSystem, directory, storage_options):
+        # As ParquetDataset will iterate all files,
+        # here we just find one file to infer dtypes
+        current_path = directory
+        partition_cols = []
+        while fs.isdir(current_path):
+            _, dirs, files = next(fs.walk(current_path))
+            dirs = [d for d in dirs if not d.startswith(".")]
+            files = [f for f in files if not f.startswith(".")]
+            if len(files) == 0:
+                # directory as partition
+                partition_cols.append(dirs[0].split("=", 1)[0])
+                current_path = os.path.join(current_path, dirs[0])
+            elif len(dirs) == 0:
+                # parquet files in deepest directory
+                current_path = os.path.join(current_path, files[0])
+            else:  # pragma: no cover
+                raise ValueError(
+                    "Files and directories are mixed in an intermediate directory"
+                )
+
+        # current path is now a parquet file
+        with open_file(current_path, storage_options=storage_options) as f:
+            dtypes = self.read_dtypes(f)
+        for partition in partition_cols:
+            dtypes[partition] = pd.CategoricalDtype()
+        return dtypes
+
+
+def _parse_prefix(path):
+    path_prefix = ""
+    if isinstance(path, str):
+        parsed_path = urlparse(path)
+        if parsed_path.scheme:
+            path_prefix = f"{parsed_path.scheme}://{parsed_path.netloc}"
+    return path_prefix
+
+
+class ArrowEngine(ParquetEngine):
+    def get_row_num(self, f):
+        file = pq.ParquetFile(f)
+        return file.metadata.num_rows
+
+    def read_dtypes(self, f, **kwargs):
+        file = pq.ParquetFile(f)
+        return file.schema_arrow.empty_table().to_pandas().dtypes
+
+    @classmethod
+    def _table_to_pandas(cls, t, nrows=None, use_arrow_dtype=None):
+        if nrows is not None:
+            t = t.slice(0, nrows)
+        if use_arrow_dtype:
+            df = t.to_pandas(types_mapper={pa.string(): ArrowStringDtype()}.get)
+        else:
+            df = t.to_pandas()
+        return df
+
+    def read_to_pandas(
+        self, f, columns=None, nrows=None, use_arrow_dtype=None, **kwargs
+    ):
+        file = pq.ParquetFile(f)
+        t = file.read(columns=columns, **kwargs)
+        return self._table_to_pandas(t, nrows=nrows, use_arrow_dtype=use_arrow_dtype)
+
+    def read_group_to_pandas(
+        self, f, group_index, columns=None, nrows=None, use_arrow_dtype=None, **kwargs
+    ):
+        file = pq.ParquetFile(f)
+        t = file.read_row_group(group_index, columns=columns, **kwargs)
+        return self._table_to_pandas(t, nrows=nrows, use_arrow_dtype=use_arrow_dtype)
+
+
+class FastpaquetEngine(ParquetEngine):
+    def get_row_num(self, f):
+        file = fastparquet.ParquetFile(f)
+        return file.count()
+
+    def read_dtypes(self, f, **kwargs):
+        file = fastparquet.ParquetFile(f)
+        dtypes_dict = file._dtypes()
+        return pd.Series(dict((c, dtypes_dict[c]) for c in file.columns))
+
+    def read_to_pandas(
+        self, f, columns=None, nrows=None, use_arrow_dtype=None, **kwargs
+    ):
+        file = fastparquet.ParquetFile(f)
+        df = file.to_pandas(columns, **kwargs)
+        if nrows is not None:
+            df = df.head(nrows)
+        if use_arrow_dtype:
+            df = df.astype(to_arrow_dtypes(df.dtypes).to_dict())
+        return df
+
+
+class CudfEngine:
+    @classmethod
+    def read_to_cudf(cls, file, columns: list = None, nrows: int = None, **kwargs):
+        df = cudf.read_parquet(file, columns=columns, **kwargs)
+        if nrows is not None:
+            df = df.head(nrows)
+        return df
+
+    def read_group_to_cudf(
+        self, file, group_index: int, columns: list = None, nrows: int = None, **kwargs
+    ):
+        return self.read_to_cudf(
+            file, columns=columns, nrows=nrows, row_groups=group_index, **kwargs
+        )
+
+    @classmethod
+    def read_partitioned_to_cudf(
+        cls,
+        file,
+        partitions: Dict,
+        partition_keys: Dict,
+        columns=None,
+        nrows=None,
+        **kwargs,
+    ):
+        # cudf will read entire partitions even if only one partition provided,
+        # so we just read with pyarrow and convert to cudf DataFrame
+        file = pq.ParquetFile(file)
+        t = file.read(columns=columns, **kwargs)
+        t = t.slice(0, nrows) if nrows is not None else t
+        t = pa.table(t.columns, names=t.column_names)
+        raw_df = cudf.DataFrame.from_arrow(t)
+        for col, value in partition_keys.items():
+            dictionary = partitions[col].tolist()
+            codes = cudf.core.column.as_column(
+                dictionary.index(value), length=len(raw_df)
+            )
+            raw_df[col] = cudf.core.column.build_categorical_column(
+                categories=dictionary,
+                codes=codes,
+                size=codes.size,
+                offset=codes.offset,
+                ordered=False,
+            )
+        return raw_df
+
+
+class DataFrameReadParquet(
+    IncrementalIndexDatasource,
+    ColumnPruneSupportedDataSourceMixin,
+):
+    _op_type_ = opcodes.READ_PARQUET
+
+    path = AnyField("path")
+    engine = StringField("engine")
+    columns = ListField("columns")
+    use_arrow_dtype = BoolField("use_arrow_dtype")
+    groups_as_chunks = BoolField("groups_as_chunks")
+    group_index = Int32Field("group_index")
+    read_kwargs = DictField("read_kwargs")
+    incremental_index = BoolField("incremental_index")
+    storage_options = DictField("storage_options")
+    is_partitioned = BoolField("is_partitioned")
+    merge_small_files = BoolField("merge_small_files")
+    merge_small_file_options = DictField("merge_small_file_options")
+    # for chunk
+    partitions = DictField("partitions", default=None)
+    partition_keys = DictField("partition_keys", default=None)
+    num_group_rows = Int64Field("num_group_rows", default=None)
+    # as read meta may be too time-consuming when number of files is large,
+    # thus we only read first file to get row number and raw file size
+    first_chunk_row_num = Int64Field("first_chunk_row_num")
+    first_chunk_raw_bytes = Int64Field("first_chunk_raw_bytes")
+
+    def get_columns(self):
+        return self.columns
+
+    def set_pruned_columns(self, columns, *, keep_order=None):
+        self.columns = columns
+
+    def __call__(self, index_value=None, columns_value=None, dtypes=None):
+        self._output_types = [OutputType.dataframe]
+        shape = (np.nan, len(dtypes))
+        return self.new_dataframe(
+            None,
+            shape,
+            dtypes=dtypes,
+            index_value=index_value,
+            columns_value=columns_value,
+        )
+
+    @classmethod
+    def estimate_size(
+        cls, ctx: MutableMapping[str, Union[int, float]], op: "DataFrameReadParquet"
+    ):  # pragma: no cover
+        # todo implement this to facilitate local computation
+        ctx[op.outputs[0].key] = float("inf")
+
+
+def read_parquet(
+    path,
+    engine: str = "auto",
+    columns: list = None,
+    groups_as_chunks: bool = False,
+    use_arrow_dtype: bool = None,
+    incremental_index: bool = False,
+    storage_options: dict = None,
+    memory_scale: int = None,
+    merge_small_files: bool = True,
+    merge_small_file_options: dict = None,
+    gpu: bool = None,
+    **kwargs,
+):
+    """
+    Load a parquet object from the file path, returning a DataFrame.
+
+    Parameters
+    ----------
+    path : str, path object or file-like object
+        Any valid string path is acceptable. The string could be a URL.
+        For file URLs, a host is expected. A local file could be:
+        ``file://localhost/path/to/table.parquet``.
+        A file URL can also be a path to a directory that contains multiple
+        partitioned parquet files. Both pyarrow and fastparquet support
+        paths to directories as well as file URLs. A directory path could be:
+        ``file://localhost/path/to/tables``.
+        By file-like object, we refer to objects with a ``read()`` method,
+        such as a file handler (e.g. via builtin ``open`` function)
+        or ``StringIO``.
+    engine : {'auto', 'pyarrow', 'fastparquet'}, default 'auto'
+        Parquet library to use. The default behavior is to try 'pyarrow',
+        falling back to 'fastparquet' if 'pyarrow' is unavailable.
+    columns : list, default=None
+        If not None, only these columns will be read from the file.
+    groups_as_chunks : bool, default False
+        if True, each row group correspond to a chunk.
+        if False, each file correspond to a chunk.
+        Only available for 'pyarrow' engine.
+    incremental_index: bool, default False
+        If index_col not specified, ensure range index incremental,
+        gain a slightly better performance if setting False.
+    use_arrow_dtype: bool, default None
+        If True, use arrow dtype to store columns.
+    storage_options: dict, optional
+        Options for storage connection.
+    memory_scale: int, optional
+        Scale that real memory occupation divided with raw file size.
+    merge_small_files: bool, default True
+        Merge small files whose size is small.
+    merge_small_file_options: dict
+        Options for merging small files
+    **kwargs
+        Any additional kwargs are passed to the engine.
+
+    Returns
+    -------
+    MaxFrame DataFrame
+    """
+
+    engine_type = check_engine(engine)
+    engine = get_engine(engine_type)
+
+    single_path = path[0] if isinstance(path, list) else path
+    fs = get_fs(single_path, storage_options)
+    is_partitioned = False
+    if fs.isdir(single_path):
+        paths = fs.ls(path)
+        if all(fs.isdir(p) for p in paths):
+            # If all are directories, it is read as a partitioned dataset.
+            dtypes = engine.read_partitioned_dtypes(fs, path, storage_options)
+            is_partitioned = True
+        else:
+            with fs.open(paths[0], mode="rb") as f:
+                dtypes = engine.read_dtypes(f)
+    else:
+        if not isinstance(path, list):
+            file_path = glob(path, storage_options=storage_options)[0]
+        else:
+            file_path = path[0]
+
+        with open_file(file_path, storage_options=storage_options) as f:
+            dtypes = engine.read_dtypes(f)
+
+    if columns:
+        dtypes = dtypes[columns]
+
+    if use_arrow_dtype is None:
+        use_arrow_dtype = options.dataframe.use_arrow_dtype
+    if use_arrow_dtype:
+        dtypes = to_arrow_dtypes(dtypes)
+
+    index_value = parse_index(pd.RangeIndex(-1))
+    columns_value = parse_index(dtypes.index, store_data=True)
+    op = DataFrameReadParquet(
+        path=path,
+        engine=engine_type,
+        columns=columns,
+        groups_as_chunks=groups_as_chunks,
+        use_arrow_dtype=use_arrow_dtype,
+        read_kwargs=kwargs,
+        incremental_index=incremental_index,
+        storage_options=storage_options,
+        is_partitioned=is_partitioned,
+        memory_scale=memory_scale,
+        merge_small_files=merge_small_files,
+        merge_small_file_options=merge_small_file_options,
+        gpu=gpu,
+    )
+    return op(index_value=index_value, columns_value=columns_value, dtypes=dtypes)