Question

In networkx,

shortest_path(G, source=None, target=None, weight=None)  
# weight/distance/cost 1 by default

can support edge proprety “weight” as operator to calculate the shortest path between two node in graph.

However, if i have other meta class attached with a node/edge, for example:

class meta( object ):
    def __init__( self, weight_shift = 1 ):
        self.weight_shift = weight_shift

G.add_node('A', meta_data = meta( weight_shift = 100 ) )
G.add_node('B', meta_data = meta( weight_shift = 200 ) )
G.add_node('C')
...
G.add_edge("A", "C", weight=10, meta_data = meta( weight_shift = -5 ))  
G.add_edge("B", "C", weight=10, meta_data = meta( weight_shift = -10 ))
G.add_edge("A","B")

is it possible to define a function as weight parameter for shortest_path()?

def weight_sum():
    ...

Which can calculate the “weight” in “runtime”, for example, use logic:

weight_sum = edge.weight + edge.meta.weight_shift + node_left.meta.weight_shift + node_right.meta.weight_shift

then

shortest_path(G, source="A", target="B", weight=weight_sum())

to get the shortest path?

THANKS.

Answer 1

It’s not possible with the current implementation in NetworkX.

Two options are:

1) Before your shortest path search process all of the edges to add a new edge attribute according to your weight function. Use that attribute as the ‘weight’ argument to shortest_path()

2) Replace the lines like

vw_dist = dist[v] + edgedata.get(weight,1)

in the code for Dijkstra’s algorithm to use your custom function for the edge weight instead of getting the ‘weight’ attribute.