Closed davetapley closed 1 year ago
Yes, this can be fixed here. fastparquet does support storing metadata using write(fn, df, ..., custom_metadata=)
, so if df
has there attrs, we can extract them. The whole of this metadata appears in .key_value_metadata
when opening the file, so we can restore them.
I am disappointed in pandas that they have once again ignored fastparquet.
Would you like to implement the functionality, @davetapley ?
Pandas 2.10 has ⬇️, but it only works with
pyarrow
. Ifpyarrow
isn't installed then if falls back tofastparquet
(per docs), andattrs
is always loaded as{}
.I assume this need a fix here?