Group nodes together in networkx

Question

I have a visualization problem involving a graph. I have N nodes, which belong to say some M networks. The nodes can have inter-network edges (within the same network) and intra-network edges (edges from a node in one network to another one).

When I visualize the graph in networkx I am looking for a way to place/cluster the networks together so that I can easily make out the inter/intra network connections. So ideally all the blue nodes would be clustered together as network (in no particular order). Similarly for the orange or green ones.

By the way I am not trying to find hubs/clusters, I know which nodes are in which networks, I just am trying to find a way to visualize it neater. Is there some easy way to do this? Something like an advanced spring layout where I can specify some nodes should show up together regardless of the edge weights / spring force?

---

Minimal working generator

import string, random
import networkx as nx
import matplotlib.pyplot as plt
from scipy.sparse import random as sparse_random


# Random string generator
def rand_string(size=6, chars=string.ascii_uppercase):
    return ''.join(random.choice(chars) for _ in range(size))


# Set up a nodes and networks randomly
nodes = [rand_string() for _ in range(30)]
networks = [rand_string() for _ in range(5)]
networks_list = networks*6
random.shuffle(networks_list)

# Define what nodes belong to what network and what their color should be
node_network_map = dict(zip(nodes, networks_list))
colors = ['green', 'royalblue', 'red', 'orange', 'cyan']
color_map = dict(zip(networks, colors))

graph = nx.Graph()
graph.add_nodes_from(nodes)
nodes_by_color = {val: [node for node in graph if color_map[node_network_map[node]] == val]
                  for val in colors}

# Take random sparse matrix as adjacency matrix
mat = sparse_random(30, 30, density=0.3).todense()
for row, row_val in enumerate(nodes):
    for col, col_val in enumerate(nodes):
        if col > row and mat[row, col] != 0.0: # Stick to upper half triangle, mat is not symmetric
            graph.add_edge(row_val, col_val, weight=mat[row, col])

# Choose a layout to visualize graph
pos = nx.spring_layout(graph)
edges = graph.edges()

# Get the edge weights and normalize them 
weights = [abs(graph[u][v]['weight']) for u, v in edges]
weights_n = [5*float(i)/max(weights) for i in weights] # Change 5 to control thickness

# First draw the nodes 
plt.figure()
for color, node_names in nodes_by_color.items():
    nx.draw_networkx_nodes(graph, pos=pos, nodelist=node_names, node_color=color)

# Then draw edges with thickness defined by weights_n
nx.draw_networkx_edges(graph, pos=pos, width=weights_n)
nx.draw_networkx_labels(graph, pos=pos)
plt.show()

Answer 1

To get better nodes layout, I start with using circular layout (replacing your spring-layout). Then I move each group of the nodes to their new locations along perimeter of a bigger circle.

# --- Begin_myhack ---
# All this code should replace original `pos=nx.spring_layout(graph)`
import numpy as np
pos = nx.circular_layout(graph)   # replaces your original pos=...
# prep center points (along circle perimeter) for the clusters
angs = np.linspace(0, 2*np.pi, 1+len(colors))
repos = []
rad = 3.5     # radius of circle
for ea in angs:
    if ea > 0:
        #print(rad*np.cos(ea), rad*np.sin(ea))  # location of each cluster
        repos.append(np.array([rad*np.cos(ea), rad*np.sin(ea)]))
for ea in pos.keys():
    #color = 'black'
    posx = 0
    if ea in nodes_by_color['green']:
        #color = 'green'
        posx = 0
    elif ea in nodes_by_color['royalblue']:
        #color = 'royalblue'
        posx = 1
    elif ea in nodes_by_color['red']:
        #color = 'red'
        posx = 2
    elif ea in nodes_by_color['orange']:
        #color = 'orange'
        posx = 3
    elif ea in nodes_by_color['cyan']:
        #color = 'cyan'
        posx = 4
    else:
        pass
    #print(ea, pos[ea], pos[ea]+repos[posx], color, posx)
    pos[ea] += repos[posx]
# --- End_myhack ---

The output plot will be similar to this:

EDIT

Usually, no particular layout is best in all situations. So, I offer the second solution which uses concentric circles to separate individual groups of the nodes. Here are the relevant code and its sample output.

# --- Begin_my_hack ---
# All this code should replace original `pos=nx.spring_layout(graph)`
import numpy as np
pos = nx.circular_layout(graph)
radii = [7,15,30,45,60]  # for concentric circles

for ea in pos.keys():
    new_r = 1
    if ea in nodes_by_color['green']:
        new_r = radii[0]
    elif ea in nodes_by_color['royalblue']:
        new_r = radii[1]
    elif ea in nodes_by_color['red']:
        new_r = radii[2]
    elif ea in nodes_by_color['orange']:
        new_r = radii[3]
    elif ea in nodes_by_color['cyan']:
        new_r = radii[4]
    else:
        pass
    pos[ea] *= new_r   # reposition nodes as concentric circles
# --- End_my_hack ---