STRtree Nearest method always returns one geometry

Create two points that are equidistant to a Line.

from shapely.strtree import STRtree
from shapely import wkt
from shapely.geometry import box, Point, LineString

keys = []
keys.append(Point(1.5, 3))
keys.append(Point(1, 2.5))

query = LineString([Point(1,2),Point(3,4)])

for i, g in enumerate(keys):
    bbox_intersects = box(*g.bounds).intersects(query)
    print(i, bbox_intersects, g.distance(query))

# 0 False 0.3535533905932738
# 1 False 0.3535533905932738

Create the tree and query for the nearest neighbor. Only one geometry will be returned.

tree = STRtree(keys)
result = tree.nearest(query)

print(result)

# POINT (1 2.5)

Does anyone know of an efficient workaround that always guarantees the nearest neighbor between two geometries and doesn’t yield false negatives?

[tomplex]

Ah, I see what you’re saying. In that case I would probably try something like:

• start by querying the tree for the bounding box of the geometry of interest
• if no results are returned, increase the size bounding box by some set amount and re-query. Figuring out the right amount will probably be the difficult part here and will depend on the density and distribution of your dataset.
• if results are returned, find all results which are the minimum distance from the geometry of interest. This will require a distance calculation for all returned results.

this gist describes what I mean. It returns both points for a couple of different symmetric situations like the one you posted above:

[carlosg-m]

@tomplex, I’ve adapted your fantastic solution to my problem.

This is probably python’s only rtree solution for exact nearest neighbors search that guarantees 100% recall and generalizes to multiple geometries.

I wasn’t able to test performance, but I hope it scales to couple million geometries.

Once again thank you!

Example dataset:

from shapely.strtree import STRtree
from shapely import wkt
from shapely.geometry import box, Point, LineString
import numpy as np

keys = []
keys.append(Point(1.5, 3))
keys.append(Point(1, 2.5))
keys.append(Point(3.3, 5.5))
keys.append(LineString([Point(-1,-2),Point(6,7)]))

query = LineString([Point(1,2),Point(3,4)])

for i, g in enumerate(keys):
    bbox_intersects = box(*g.bounds).intersects(query)
    print(i, bbox_intersects, g.distance(query))

# 0 False 0.3535533905932738
# 1 False 0.3535533905932738
# 2 False 1.5297058540778354
# 3 True 0.5262348115842175

• Create STRtree
• Query STRtree for the first nearest neighbor
• Get distance to first nearest neighbor
• Expand query geometry bounding box to include minimum distance
• Query STRtree for all geometries that intersect previously created bounding box
• Search for false positives, filter candidate geometries that are greater than minimum distance

tree = STRtree(keys)
result = tree.nearest(query)
min_distance = result.distance(query)

def _expand(geom, amt):
    amt = amt + np.finfo(np.float64).eps
    bounds = geom.bounds
    bounds = (bounds[0] - amt, bounds[1] - amt, bounds[2] + amt, bounds[3] + amt)
    return box(*bounds)

bbox_search = _expand(query, min_distance)
result = tree.query(bbox_search)

print([g.wkt for g in result])

# ['POINT (1 2.5)', 'LINESTRING (-1 -2, 6 7)', 'POINT (1.5 3)']