ENH: Support Multi-Index for columns in parquet format

[yohplala]

Is your feature request related to a problem?

I would like to save DataFrame with Multi-Index used for columns into parquet format.
This is currently not possible.

import pandas as pd
df = pd.DataFrame([[1,2,3],[4,5,6],[7,8,9]], columns= pd.MultiIndex.from_product([['1'],['a', 'b', 'c']]))
df.to_parquet('test.parquet')
""" doesn't work, whatever the engine """
ValueError: parquet must have string column names

Describe alternatives you've considered

To do so with actual state of library, a piece of code has been shared on SO.
I will use it for now.

import pyarrow as pa
import pyarrow.parquet as pq
table = pa.Table.from_pandas(df)
pq.write_table(table, 'test.parquet')
df_test_read = pd.read_parquet('test.parquet')

Thanks for your help!
Bests,

[rhshadrach]

Simply removing the code

# must have value column names (strings only)
if df.columns.inferred_type not in {"string", "empty"}:
    raise ValueError("parquet must have string column names")

this example then works, although I'm not sure what the limitations are. I've tested a couple of examples, such as columns=[1, 2, 3] and things seem to generally work. Even weird things like columns=[(1, 2), 3, 4] "work", although this is converted to ['('1', '2')', '3', '4'] when reading back in (with a warning message).

[TomAugspurger]

The check needs to be updated to handle MultiIndexes. Something like

if df.index.nlevels > 1:
    if not all(x.inferred_type in {"string", "empty"} for x in df.index.levels):
        raise ValueError(...)

[hweecat]

Hello! I could make an attempt on updating the check to handle MultiIndexes for parquet format, if that's okay. :)

[TomAugspurger]

That'd be great, thanks.