Set edge lengths according to weights in graphviz_layout

Question

I'm trying to use networkx to plot a weighted graph. I'm generating the positions of vertices using graphviz_layout. I want the edge lengths in the plot to be proportional to the edge weights I am using. Following is my code snippet to achieve this:

import networkx as nx
from networkx.drawing.nx_agraph import graphviz_layout

G = nx.Graph()
G.add_weighted_edges_from(edgelist)
pos = graphviz_layout(G)
nx.draw_networkx(G, pos=pos)

This does not seem to impact edge lengths in the plot. I have seen this post however it only describes how to pass a single default weight for all edges to graphviz_layout (or perhaps I haven't understood the solution).

How may I change my snippet to plot a graph with lengths proportional to edge weights?

In the code snippet above, edgelist is a list of lists of this format:

edgelist = [[4, 219, 1.414], [4, 7, 3.86]]

Answer 1

The key is to set the len attribute of the edges, as described in the Graphviz docs (and available for Neato). Note that this is a preferred edge length, so you will not have complete control. For example, you can add a single weighted edge (u, v, w) to a graph G like so:

G.add_edge(u, v, len=w)

Here is a full example using the Florentine families graph:

# Load modules
import networkx as nx
from networkx.drawing.nx_agraph import graphviz_layout
import matplotlib.pyplot as plt

# Construct the graph, choosing some edges to be longer
G = nx.generators.social.florentine_families_graph()
lengths = dict()
long_edges = { frozenset(['Salviati', 'Pazzi']),
               frozenset(['Ginori', 'Albizzi']) }
for edge in G.edges():
    if frozenset(edge) in long_edges:
        lengths[edge] = dict(len=2)
    else:
        lengths[edge] = dict(len=1)

nx.set_edge_attributes(G, lengths)

# Plot and show
pos = graphviz_layout(G, prog='neato')
nx.draw_networkx(G, pos=pos)

plt.tight_layout()
plt.show()

To illustrate the difference in layouts when using edge lengths, here is the result where I set two edges of degree one nodes ("Ginori" and "Pazzi") to be longer:

And here is the result not using any edge lengths: