geoarrow-based multithreaded coordinate reprojection

[kylebarron]

Change list

• Include PROJJSON CRS on geometry array metadata when converting from GeoPandas
• Move CRS checking and reprojection into GeoArrow level instead of GeoPandas level

Example

Low level example and benchmark. 4.6x reprojection speedup across 8 threads!

import geodatasets
import pyarrow as pa
import geopandas as gpd
import numpy as np

from lonboard._geoarrow.geopandas_interop import geopandas_to_geoarrow
from lonboard._geoarrow.ops import reproject


path = geodatasets.get_path('nybb')
gdf = gpd.read_file(path)
table = geopandas_to_geoarrow(gdf)

field = table.schema.field(4)
column = table['geometry']
chunk = column.chunk(0)

new_chunked_array = pa.chunked_array([chunk] * 100)
%timeit out = reproject(field, new_chunked_array, max_workers=8)
# 693 ms ± 85.6 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
%timeit out = reproject(field, new_chunked_array, max_workers=1)
# 3.22 s ± 22 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)