osmnx: get_nearest_node returning distant nodes

@kaseyklimes see the documentation for this function: if the graph is projected, you should use the ‘euclidean’ method instead of the default ‘greatcircle’ method. Also, note that the point argument is passed as a tuple of (lat, lng) if in degrees or (y, x) if in projected planar units such as meters. This is kind of weird, but I think it’s necessary: we consistently treat degree coordinates as (lat, lng) and there may be times when we want to calculate euclidean distances between them. In which case, both algorithms (euclidean and great circle) must accept a consistent point format… (lat, lng) is equivalent to (y, x).

So, if we specify a (y, x) point and use the euclidean method:

import osmnx as ox, networkx as nx
ox.config(log_console=True, use_cache=True)
G = ox.graph_from_place('Buffalo, New York, USA', network_type='walk')
G = ox.project_graph(G, to_crs={'init': 'epsg:2163'})
center_point = (-13324.947275372, 1697219.79349577) #this tuple should be (y, x), like (lat, lng)
center_node, distance = ox.get_nearest_node(G, center_point, method='euclidean', return_dist=True)
print(distance)

Prints 0.454… so the nearest node is only half a meter from the point you specified.

Clarified usage in 29005df and dd32fa2

@kuanb that’s right. Great circle distance measures distances along the surface of a sphere while euclidean distance measures it along a two dimensional plane. Lat-long data is in spherical units of degrees: it represents points on the surface of a sphere (or an approximate spheroid). If you project your lat-long data to two-dimensional planar units such as meters, great circle distance no longer makes sense: you need to use a planar metric such as Euclidean distance.

RTFM Kasey, RTFM. Thanks Geoff 👍