Mismatched numbers of nodes and edges using OSMnx

I'm downloading a road network using the OSMnx library for the San Francisco region. I get a different number of nodes and edges of the network using the graph_to_gdfs() function by directly converting to the nodes and edges from the graph compared to first exporting the graph using osmnx.io.save_graph_xml(), then reloading the same graph using osmnx.graph_from_xml() function, and finally converting the graph using the graph_to_gdfs() function. By directly converting the graph, I get 9671 nodes and 16030 edges. But, by first saving and then re-creating the same graph, I get 9649 nodes and 15975 edges. Thanks for your help!

Please see the code below:

import osmnx as ox

utn = ox.settings.useful_tags_node
oxna = ox.settings.osm_xml_node_attrs
oxnt = ox.settings.osm_xml_node_tags
utw = ox.settings.useful_tags_way
oxwa = ox.settings.osm_xml_way_attrs
oxwt = ox.settings.osm_xml_way_tags
utn = list(set(utn + oxna + oxnt))
utw = list(set(utw + oxwa + oxwt))
ox.settings.all_oneway = True
ox.settings.useful_tags_node = utn
ox.settings.useful_tags_way = utw
G = ox.graph_from_bbox(37.708300, 37.813300, -122.517100, -122.355100, network_type='drive')

nodes, edges = ox.utils_graph.graph_to_gdfs(G)
print(nodes.shape)
print(edges.shape)

ox.io.save_graph_xml(G, filepath="./data/SF_Road_Network.osm")

Recreated_G = ox.graph_from_xml('./data/SF_Road_Network.osm')
recreated_nodes, recreated_edges = ox.utils_graph.graph_to_gdfs(Recreated_G)
print(recreated_nodes.shape)
print(recreated_edges.shape)

Your code has three problems.

First, your north and south arguments are transposed in your graph_from_bbox call. Nevertheless, the function is probably smart enough to figure out which is supposed to be which in this case to make a valid polygon.

Second, you should not use the save_graph_xml to save a graph to disk for subsequent use in OSMnx. From the docs:

> This function exists only to allow serialization to the .osm file format for applications that require it, and has constraints to conform to that. As such, this function has a limited use case which does not include saving/loading graphs for subsequent OSMnx analysis. To save/load graphs to/from disk for later use in OSMnx, use the io.save_graphml and io.load_graphml functions instead. To load a graph from a .osm file that you have downloaded or generated elsewhere, use the graph.graph_from_xml function.

Third, you are simplifying your graph twice. Once (by default) when you call graph_from_bbox. Then you save it and simplify it again (by default) when you call graph_from_xml. Hence they diverge. If you pass simplify=False into your graph_from_xml call, you will have the same number of nodes (but not edges) as in the original graph.

Long story short, follow the documentation: "To save/load graphs to/from disk for later use in OSMnx, use the io.save_graphml and io.load_graphml functions instead."