fix methods to infer geom and bbox

[FlorisCalkoen]

Hi @TomAugspurger, I recently used stac table to create a STAC collection for a parquet dataset. Doing so I made some minor changes to the package. Please have a look if you would like to keep them.

In the source code I saw some TODO comments about 'maybe converting the geometries to EPSG:4326'. Since my data was in EPSG:3857, I added the reprojection for the geometries that fall directly under the pystac.Item.properties. So the geometries/bbox are now in 4326, whereas the ones under the projection prefix are in the CRS of the source data.

item.bbox # in 4326
item.geometry # in 4326
item.properties["proj:<geom>"] # in projection of source data

For some conditions, the data was being load using dask_geopandas.read_paquet(); but at least for my dataset the spatial partitions were not available without computing them first. What do you think about adding a call to calculate_spatial_partitions()?

        # some condition
        data = dask_geopandas.read_parquet(uri, storage_options=storage_options)
        data.calculate_spatial_partitions()

[TomAugspurger]

Thanks for the PR!

What do you think about adding a call to calculate_spatial_partitions()?

That sounds reasonable. We'll just want to make sure that .calculate_spatial_partitions is smart about not recomputing the values unless necessary.

I'm not immediately sure why CI failed.

>       assert result.properties["table:columns"] == expected_columns
E       AssertionError: assert [{'name': 'po...'byte_array'}] == [{'name': 'po...'byte_array'}]
E         At index 0 diff: {'name': 'pop_est', 'type': 'double'} != {'name': 'pop_est', 'type': 'int64'}
E         Full diff:
E           [
E         -  {'name': 'pop_est', 'type': 'int64'},
E         ?                               ^^^^^
E         +  {'name': 'pop_est', 'type': 'double'},
E         ?                               ^^^^^^
E            {'name': 'continent', 'type': 'byte_array'},
E            {'name': 'name', 'type': 'byte_array'},
E            {'name': 'iso_a3', 'type': 'byte_array'},
E         -  {'name': 'gdp_md_est', 'type': 'double'},
E         ?                                  ^^^^^^
E         +  {'name': 'gdp_md_est', 'type': 'int64'},
E         ?                                  ^^^^^
E            {'name': 'geometry', 'type': 'byte_array'},
E           ]

Probably unrelated to your change. Maybe pandas used to cast these to float and now it doesn't?

[FlorisCalkoen]

Hi @TomAugspurger, many thanks for the swift response! So I made the following changes to the PR:

Datatypes

I've changed the expected datetypes in the tests to double for "pop_est" and int64 for "gdp_md_est". Although this matches with what you find in Pandas, it's a bit weird because I would expect population to be an integer and gdp a float?

These are the dtypes in the test data:

print(type(ds))  # <class 'pyarrow.parquet.core._ParquetDatasetV2'>
fragment = ds.fragments[0]

print(ds.schema)

pop_est: double
continent: string
name: string
iso_a3: string
gdp_md_est: int64
geometry: binary
-- schema metadata --
pandas: '{"index_columns": [{"kind": "range", "name": null, "start": 0, "' + 945
geo: '{"primary_column": "geometry", "columns": {"geometry": {"encoding":' + 1355


print(fragment.metadata.schema)
<pyarrow._parquet.ParquetSchema object at 0x16cefd140>
required group field_id=-1 schema {
  optional double field_id=-1 pop_est;
  optional binary field_id=-1 continent (String);
  optional binary field_id=-1 name (String);
  optional binary field_id=-1 iso_a3 (String);
  optional int64 field_id=-1 gdp_md_est;
  optional binary field_id=-1 geometry;
  optional int64 field_id=-1 __null_dask_index__;
}



print(gpd.read_parquet("path/to/written/data.parquet").dtypes)  # same as gpd.read_file(gpd.datasets.get_path("naturalearth_lowres")).dtypes

pop_est        float64
continent       object
name            object
iso_a3          object
gdp_md_est       int64
geometry      geometry
dtype: object

Spatial partitions

I've added a condititon that checks if the spatial_partitions are None:

       ...
       if data.spatial_partitions is None:
            data.calculate_spatial_partitions()

Projection EPSG test

One of the tests was checking if the proj:epsg == 4326. Since this proj:epsg can now be other projections, I changed it to check if the EPSG is valid.

def is_valid_epsg(epsg_code):
    try:
        pyproj.CRS.from_user_input(epsg_code)
        return True
    except pyproj.exceptions.CRSError:
        return False

assert is_valid_epsg(result.properties["proj:epsg"])