Spatial indexing Approaches

geohash_based_spatial_radius_query_approach.py

RADIANT_TO_KM_CONSTANT = 6367
import proximitypyhash as ppyh
import pygeohash as pgh
from sklearn.neighbors import DistanceMetric
import numpy as np
from collections import defaultdict


class GeoHashIndexer:
    def __init__(self, precision, lat_longs):
        self.index = defaultdict(list)
        # build the index
        for lat_long in lat_longs:
            lat = lat_long[0]
            long = lat_long[1]
            geo_hash = pgh.encode(lat, long, precision=precision)
            lat_long_radian = np.radians(np.array(lat_long))
            self.index[geo_hash].append(lat_long_radian)
        self.precision = precision
        self.haversine = DistanceMetric.get_metric('haversine')
        self.lat_longs = lat_longs

    def query_radius(self, query, radius):
        candidate_geohashes = ppyh.get_geohash_radius_approximation(latitude=query[0],
                                                                    longitude=query[1],
                                                                    radius=radius,
                                                                    precision=self.precision,
                                                                    georaptor_flag=False)
        candidate_points = []
        for geohash in candidate_geohashes:
            if geohash in self.index.keys():
                candidate_points.extend(self.index[geohash])

        if not candidate_points:
            return []

        query = np.radians(np.array([query]))
        result = self.haversine.pairwise(candidate_points, query)
        # convert radiant radius to meters
        radius_km = radius / 1e3
        radius_radiant = radius_km / RADIANT_TO_KM_CONSTANT
        result = result[result < radius_radiant]
        return result * RADIANT_TO_KM_CONSTANT * 1000  # get meters again

On line 31 is there a reason you switched to using the keys method here? Doesn't this change the complexity from O(1) to O(n) in python 2 & something worse than O(1) in python 3?
https://stackoverflow.com/a/17539425