Disaggregating population from block groups to blocks changes total population

[gabeschoenbach]

Using the latest version of maup (version 1.0.1), disaggregating CVAP data from block groups down to blocks gives strange behavior.

import geopandas as gpd
import maup
import math

bgs = gpd.read_file("bgs.shp")
blocks = gpd.read_file("blocks.shp")

Even though these shapefiles came from the Census, there are a couple geometry errors that need to be fixed:

blocks[blocks["geometry"].apply(lambda x: isinstance(x, (Polygon, MultiPolygon)))]
bgs["geometry"] = maup.close_gaps(bgs)

Cropping and dropping empty geometries (per @InnovativeInventor and @amybecker's suggestions) doesn't change anything — in this case there aren't any empty geometries:

blocks["geometry"] = maup.crop_to(blocks, bgs)
bgs["geometry"] = maup.crop_to(bgs, blocks)
blocks = blocks[~blocks["geometry"].is_empty]
bgs = bgs[~bgs["geometry"].is_empty]

After doing this, the area of the symmetric difference between these two regions is 0:

print(blocks.unary_union.symmetric_difference(bgs.unary_union).area)

Disaggregating CVAP from bgs -> blocks:

cvap_cols = ["CVAP", "WCVAP", "BCVAP", "HCVAP", "ASIANCVAP"]
assignment = maup.assign(blocks, bgs)
weights = blocks["VAP"] / assignment.map(bgs["VAP"])
prorated = maup.prorate(assignment, bgs[cvap_cols], weights)
blocks[cvap_cols] = prorated

The issue is that the following should be True, but it isn't:

math.isclose(blocks["CVAP"].sum(), bgs["CVAP"].sum())

[InnovativeInventor]

Fixed by #34