Copy spatial_partitions when copying GeoDataFrame

[DahnJ]

I would expect the spatial_partitions to be copied when using df.copy().

import geopandas
import dask
import dask_geopandas
dask.config.set(scheduler='single-threaded')

df = geopandas.read_file(geopandas.datasets.get_path("naturalearth_lowres"))
ddf = dask_geopandas.from_geopandas(df, npartitions=4)
ddf.calculate_spatial_partitions()
ddfc = ddf.copy()
ddfc.spatial_partitions  # is None

Is there a reason to not copy spatial partitions that I didn't think of?

I thought about how this could work and didn't think of anything better than overriding dask's copy method. Something like

def copy(self):
    self_copy = super().copy()
    self_copy.spatial_partitions = self.spatial_partitions
    return self_copy

It would be great if spatial_partitions could get copied automatically, as crs for example does. However, crs is actually a property of _meta, which does get automatically copied by dask. Not sure how spatial partitions could be attached to the _meta, or if there's another mechanism that could act similarly.

If I'm headed in the right direction, then I'm happy to unit test & PR.

[jorisvandenbossche]

Yes, I agree that would certainly be expected to be propagated to the copy. I suspect there will be other methods as well where we need to fix this. On the short term, I think your proposed fix of overriding the method like that is fine (maybe just copy the actual spatial_partitions as well). PR certainly welcome! Long term it would be good to have a nicer way to have the spatial paritions copied over, but probably something to discuss with the dask team what this could look like. Op za 12 jun. 2021 12:10 schreef Dahn ***@***.***>:

…

I would expect the spatial_partitions to be copied when using df.copy(). import geopandasimport daskimport dask_geopandasdask.config.set(scheduler='single-threaded') df = geopandas.read_file(geopandas.datasets.get_path("naturalearth_lowres"))ddf = dask_geopandas.from_geopandas(df, npartitions=4)ddf.calculate_spatial_partitions()ddfc = ddf.copy()ddfc.spatial_partitions # is None Is there a reason to not copy spatial partitions that I didn't think of? I thought about how this could work and didn't think of anything better than overriding dask's copy method. Something like def copy(self): self_copy = super().copy() self_copy.spatial_partitions = self.spatial_partitions return self_copy It would be great if spatial_partitions could get copied automatically, as crs for example does. However, crs is actually a property of _meta, which does get automatically copied by dask. Not sure how spatial partitions could be attached to the _meta, or if there's another mechanism that could act similarly. If I'm headed in the right direction, then I'm happy to unit test & PR. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#55>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAHZEUFSTGBLCCWHJWKMHYTTSMXCZANCNFSM46STCMDA> .

[martinfleis]

Closed by #56