Dynamically change size of arrowheads in networkx based on list/dict

Question

I can dynamically change the node size or node color by passing a list of values to the draw_network function. But how can i do it with ArrowStyle? Say I wanted to change the ArrowStyle (width & length) based on a list of values. The arrowsize doesn't either accept anything else than a single int value.

Here is an example code:

import matplotlib.patches
import networkx as nx
G = nx.DiGraph()
G.add_edge("reddit", "youtube")
G.add_edge("reddit", "google")
G.add_edge("google", "reddit")
G.add_edge("youtube", "reddit")
print(G.adj)

testArrow = matplotlib.patches.ArrowStyle.Fancy(head_length=.4, head_width=.4, tail_width=.1)


nx.draw_networkx(G,
             arrowsize=30,
             node_size=[5000,50,300],
             arrowstyle=testArrow)

Answer 1

For changing the dimensions of arrows:

According to the documentation there is no option to do that in a single call. However, it can be done by drawing the edges one by one as follows:

import networkx as nx
import matplotlib.pyplot as plt

G = nx.DiGraph()
G.add_edge("reddit", "youtube", w=5)
G.add_edge("reddit", "google", w=10)
G.add_edge("google", "reddit", w=30)
G.add_edge("youtube", "reddit", w=20)
print(G.adj)

pos = nx.spring_layout(G)
nx.draw_networkx_nodes(G, pos, node_size=[5000, 50, 300])
nx.draw_networkx_labels(G, pos)
for edge in G.edges(data=True):
    w = edge[2]['w']
    nx.draw_networkx_edges(G, pos, edgelist=[(edge[0],edge[1])], arrowsize=w, node_size=[5000, 50, 300])
plt.show()

Which will bring to something like this:

For changing the dimensions of edges:

There's a conceptual difference between edges' width and length. While width is configurable and can be easily set per edge, length is defined by the position of nodes.

For changing edges' width in a similar manner to nodes' size and color, you can call draw_networkx_edges, and the argument 'width' accepts either float, or an array of floats.

For changing the edges' length you'll have to change the layout, set by 'pos' argument in nx.draw_networkx. The default is using a spring layout positioning.