从 pandas DataFrame 加载 pyarrow parquet 时保留索引

Question

我需要将带有字典值的字典转换为镶木地板，我的数据如下所示：

{"KEY":{"2018-12-06":250.0,"2018-12-07":234.0}}

我正在转换为 pandas 数据帧，然后写入 pyarrow table：

import pandas as pd
import pyarrow as pa
import pyarrow.parquet as pq

data = {"KEY":{"2018-12-06":250.0,"2018-12-07":234.0}}
df = pd.DataFrame.from_dict(data, orient='index')
table = pa.Table.from_pandas(df, preserve_index=False)
pq.write_table(table, 'file.parquet', flavor='spark')

我得到的数据只有日期和值，但没有字典的键。:

{"2018-12-06":250.0,"2018-12-07":234.0}

我需要的是也有数据的key：

{"KEY": {"2018-12-06":250.0,"2018-12-07":234.0}}

Answer 1

如果您想保留索引，那么您应该这样指定；设置 preserve_index=True:

table = pa.Table.from_pandas(df, preserve_index=True)

pq.write_table(table, 'file.parquet', flavor='spark')
pq.read_table('file.parquet').to_pandas()  # Index is preserved.

     2018-12-06  2018-12-07
KEY       250.0       234.0

Answer 2

我观察到一个相关但独立的问题，即 DateTimeIndex 的频率类型在从 pandas 到 table 的往返过程中未保留。

例如：

    >>> import pandas as pd
    >>> import pyarrow as pa
    >>> from collections import OrderedDict
    >>>
    >>>
    >>> pd.__version__
    '1.1.5'
    >>>
    >>> pa.__version__
    '4.0.1'
    >>>
    >>> dates = pd.date_range(start='2016-04-01', periods=4, name='DATE')
    >>> dict_data = OrderedDict()
    >>> dict_data['A'] = list('AABB')
    >>> dict_data['B'] = list('abab')
    >>> dict_data['C'] = list('wxyz')
    >>> dict_data['D'] = range(0, 4)
    >>> df = pd.DataFrame.from_dict(dict_data)
    >>> df = df.set_index(dates)
    >>>
    >>> df.index
    DatetimeIndex(['2016-04-01', '2016-04-02', '2016-04-03', '2016-04-04'], dtype='datetime64[ns]', name='DATE', freq='D')
    >>>
    >>> table = pa.Table.from_pandas(df, preserve_index=True)
    >>> df2 = table.to_pandas()
    >>> df2.index
    DatetimeIndex(['2016-04-01', '2016-04-02', '2016-04-03', '2016-04-04'], dtype='datetime64[ns]', name='DATE', freq=None)

从 pandas DataFrame 加载 pyarrow parquet 时保留索引

Preserve index when loading pyarrow parquet from pandas DataFrame

python

dictionary

dataframe

pandas

pyarrow