python drawing directed graph in SPYDER

Question

I am using SPYDER. I have below code. The code produces the graph. But the graph is not clean. I would like to see layers in the graph - my first layer has 'start' node, second layer has 1,2,3 node, third layer has 'a', 'b', 'c' nodes and the last layer has 'end' node. Any idea how could i achieve my objective?

import networkx as nx
import matplotlib.pyplot as plt
import os.path as path

G=nx.DiGraph()
G.add_nodes_from(['start',1,2,3,'a','b','c','end'])

G.nodes(data=True)

G.add_edge('start',2)
G.add_edge('start',3)
G.add_edge(2,'b')
G.add_edge('b','end')
G.add_edge('a','end')
G.add_edge('f','g')
nx.draw(G)

------------------------update------------------

by layer i mean I should see the network like in the attached figure. I want the network drawn that way because a node in layer X will be connected directly ONLY to nodes in layer X+1 or layer X-1

Answer 1

It [seems that networkx does not provide much for arranging graphs but I have only touched it a few times so I can't say for sure. I couldn't find a way to do what you are asking.

However, I tried using graphviz (which has to be installed first) to accomplish it. Credit for the ranking method goes to this SO answer.

Copy-Paste Example

import networkx as nx

ranked_node_names = [['start'],
                     [1, 2, 3],
                     ['a', 'b', 'c'],
                     ['end']]
node_edges = [('start', 2),
              ('start', 3),
              (2, 'b'),
              ('b', 'end'),
              ('a', 'end')]

# graph and base nodes/edges in networkx
G = nx.DiGraph()
for rank_of_nodes in ranked_node_names:
    G.add_nodes_from(rank_of_nodes)
G.nodes(data=True)
G.add_edges_from(node_edges)
# I don't know a way to automatically arrange in networkx so using graphviz
A = nx.to_agraph(G)
A.graph_attr.update(rankdir='LR')  # change direction of the layout
for rank_of_nodes in ranked_node_names:
    A.add_subgraph(rank_of_nodes, rank='same')
# draw
A.draw('example.png', prog='dot')

Default Output

LR (left-right) Output

Answer 2

You have at least two options: You can create a dictionary with node positions yourself if you like. First assign nodes to layers and then:

position = dict()
for (i, layer) in enumerate(layers):
    for (j, node) in enumerate(layer):
        position[node] = (i, j / float(len(layer))

and then you draw the network with the given node positions nx.draw(G, pos=position).

The other option is to use graphviz and the python AGraph class which can be used to produce hierarchical layouts.