Get warning even though exporting non-geodata to parquet

I use both geodata and normal data in the same python file.

Context: As I work with point data, I transform the data set to normal pandas by dropping the geometry and saving LAT/LON.

e.g.

out = out.drop("geometry", axis=1)
out.to_parquet("bla.parqet")

Even though this is not a geodataframe any more I receive the warning:

C:\Users\rados\Documents\adv\point_linkage\src\data_management\merge_shapefiles.py:249: UserWarning: this is an initial implementation of Parquet/Feather file support and associated metadata.  This is tracking version 0.1.0 of the metadata specification at https://github.com/geopandas/geo-arrow-spec

This metadata specification does not yet make stability promises.  We do not yet recommend using this in a production setting unless you are able to rewrite your Parquet/Feather files.

Does this have any implication for my case?

[jorisvandenbossche]

@raholler Thanks a lot for your report!

I moved your issue to the geopandas repo, as it’s related to the implementation of geopandas.

So what is happening is that the drop method removes the geometry column, but the result is (incorrectly) still a GeoDataFrame, and thus using the to_parquet implementation of GeoPandas instead of pandas.

And it seems that our implementation doesn’t really correctly handle the case of no geometry column:

In [29]: df = geopandas.read_file(geopandas.datasets.get_path("naturalearth_cities"))

In [30]: df = df.drop("geometry", axis=1)

In [31]: type(df)
Out[31]: geopandas.geodataframe.GeoDataFrame

In [32]: df.to_parquet("test_no_geometry_column.parquet")
UserWarning: this is an initial implementation ....

In [33]: import pyarrow.parquet as pq

In [34]: meta = pq.read_metadata("test_geo.parquet")

In [36]: meta.metadata[b'geo']
Out[36]: b'{"primary_column": "geometry", "columns": {}, "schema_version": "0.1.0", "creator": {"library": "geopandas", "version": "0.8.0+48.g1e975ab"}}'

So we included incorrect metadata in the parquet file in this case.

@raholler short-term work-around is converting the result of drop() explicitly to a DataFrame (with pd.DataFrame(df.drop(...)))

[jorisvandenbossche]

Given the changes with that we still keep it as a GeoDataFrame if there is any geometry column, and thus more explicitly allow a GeoDataFrame without active geometry column, we should probably still test that case (to ensure we write this correctly, and can read the resulting file)