# Licensed to the Apache Software Foundation (ASF) under one
# or more contributor license agreements. See the NOTICE file
# distributed with this work for additional information
# regarding copyright ownership. The ASF licenses this file
# to you 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 pyarrow.pandas_compat import _pandas_api # noqa
from pyarrow.lib import (Codec, Table, # noqa
concat_tables, schema)
import pyarrow.lib as ext
from pyarrow import _feather
from pyarrow._feather import FeatherError # noqa: F401
class FeatherDataset:
"""
Encapsulates details of reading a list of Feather files.
Parameters
----------
path_or_paths : List[str]
A list of file names
validate_schema : bool, default True
Check that individual file schemas are all the same / compatible
"""
def __init__(self, path_or_paths, validate_schema=True):
self.paths = path_or_paths
self.validate_schema = validate_schema
def read_table(self, columns=None):
"""
Read multiple feather files as a single pyarrow.Table
Parameters
----------
columns : List[str]
Names of columns to read from the file
Returns
-------
pyarrow.Table
Content of the file as a table (of columns)
"""
_fil = read_table(self.paths[0], columns=columns)
self._tables = [_fil]
self.schema = _fil.schema
for path in self.paths[1:]:
table = read_table(path, columns=columns)
if self.validate_schema:
self.validate_schemas(path, table)
self._tables.append(table)
return concat_tables(self._tables)
def validate_schemas(self, piece, table):
if not self.schema.equals(table.schema):
raise ValueError('Schema in {!s} was different. \n'
'{!s}\n\nvs\n\n{!s}'
.format(piece, self.schema,
table.schema))
def read_pandas(self, columns=None, use_threads=True):
"""
Read multiple Parquet files as a single pandas DataFrame
Parameters
----------
columns : List[str]
Names of columns to read from the file
use_threads : bool, default True
Use multiple threads when converting to pandas
Returns
-------
pandas.DataFrame
Content of the file as a pandas DataFrame (of columns)
"""
return self.read_table(columns=columns).to_pandas(
use_threads=use_threads)
def check_chunked_overflow(name, col):
if col.num_chunks == 1:
return
if col.type in (ext.binary(), ext.string()):
raise ValueError("Column '{}' exceeds 2GB maximum capacity of "
"a Feather binary column. This restriction may be "
"lifted in the future".format(name))
else:
# TODO(wesm): Not sure when else this might be reached
raise ValueError("Column '{}' of type {} was chunked on conversion "
"to Arrow and cannot be currently written to "
"Feather format".format(name, str(col.type)))
_FEATHER_SUPPORTED_CODECS = {'lz4', 'zstd', 'uncompressed'}
[docs]def write_feather(df, dest, compression=None, compression_level=None,
chunksize=None, version=2):
"""
Write a pandas.DataFrame to Feather format.
Parameters
----------
df : pandas.DataFrame or pyarrow.Table
Data to write out as Feather format.
dest : str
Local destination path.
compression : string, default None
Can be one of {"zstd", "lz4", "uncompressed"}. The default of None uses
LZ4 for V2 files if it is available, otherwise uncompressed.
compression_level : int, default None
Use a compression level particular to the chosen compressor. If None
use the default compression level
chunksize : int, default None
For V2 files, the internal maximum size of Arrow RecordBatch chunks
when writing the Arrow IPC file format. None means use the default,
which is currently 64K
version : int, default 2
Feather file version. Version 2 is the current. Version 1 is the more
limited legacy format
"""
if _pandas_api.have_pandas:
if (_pandas_api.has_sparse and
isinstance(df, _pandas_api.pd.SparseDataFrame)):
df = df.to_dense()
if _pandas_api.is_data_frame(df):
# Feather v1 creates a new column in the resultant Table to
# store index information if index type is not RangeIndex
if version == 1:
preserve_index = False
elif version == 2:
preserve_index = None
else:
raise ValueError("Version value should either be 1 or 2")
table = Table.from_pandas(df, preserve_index=preserve_index)
if version == 1:
# Version 1 does not chunking
for i, name in enumerate(table.schema.names):
col = table[i]
check_chunked_overflow(name, col)
else:
table = df
if version == 1:
if len(table.column_names) > len(set(table.column_names)):
raise ValueError("cannot serialize duplicate column names")
if compression is not None:
raise ValueError("Feather V1 files do not support compression "
"option")
if chunksize is not None:
raise ValueError("Feather V1 files do not support chunksize "
"option")
else:
if compression is None and Codec.is_available('lz4_frame'):
compression = 'lz4'
elif (compression is not None and
compression not in _FEATHER_SUPPORTED_CODECS):
raise ValueError('compression="{}" not supported, must be '
'one of {}'.format(compression,
_FEATHER_SUPPORTED_CODECS))
try:
_feather.write_feather(table, dest, compression=compression,
compression_level=compression_level,
chunksize=chunksize, version=version)
except Exception:
if isinstance(dest, str):
try:
os.remove(dest)
except os.error:
pass
raise
[docs]def read_feather(source, columns=None, use_threads=True,
memory_map=False, **kwargs):
"""
Read a pandas.DataFrame from Feather format. To read as pyarrow.Table use
feather.read_table.
Parameters
----------
source : str file path, or file-like object
You can use MemoryMappedFile as source, for explicitly use memory map.
columns : sequence, optional
Only read a specific set of columns. If not provided, all columns are
read.
use_threads : bool, default True
Whether to parallelize reading using multiple threads. If false the
restriction is used in the conversion to Pandas as well as in the
reading from Feather format.
memory_map : boolean, default False
Use memory mapping when opening file on disk, when source is a str.
**kwargs
Additional keyword arguments passed on to `pyarrow.Table.to_pandas`.
Returns
-------
df : pandas.DataFrame
The contents of the Feather file as a pandas.DataFrame
"""
return (read_table(
source, columns=columns, memory_map=memory_map,
use_threads=use_threads).to_pandas(use_threads=use_threads, **kwargs))
[docs]def read_table(source, columns=None, memory_map=False, use_threads=True):
"""
Read a pyarrow.Table from Feather format
Parameters
----------
source : str file path, or file-like object
You can use MemoryMappedFile as source, for explicitly use memory map.
columns : sequence, optional
Only read a specific set of columns. If not provided, all columns are
read.
memory_map : boolean, default False
Use memory mapping when opening file on disk, when source is a str
use_threads : bool, default True
Whether to parallelize reading using multiple threads.
Returns
-------
table : pyarrow.Table
The contents of the Feather file as a pyarrow.Table
"""
reader = _feather.FeatherReader(
source, use_memory_map=memory_map, use_threads=use_threads)
if columns is None:
return reader.read()
column_types = [type(column) for column in columns]
if all(map(lambda t: t == int, column_types)):
table = reader.read_indices(columns)
elif all(map(lambda t: t == str, column_types)):
table = reader.read_names(columns)
else:
column_type_names = [t.__name__ for t in column_types]
raise TypeError("Columns must be indices or names. "
"Got columns {} of types {}"
.format(columns, column_type_names))
# Feather v1 already respects the column selection
if reader.version < 3:
return table
# Feather v2 reads with sorted / deduplicated selection
elif sorted(set(columns)) == columns:
return table
else:
# follow exact order / selection of names
return table.select(columns)
