Plotting repeated x-axis ranges in matplotlib, plotly and networkx

Question

I need to draw a networkx graph with x axis ranges are repeating, such as 0-100, 0-100,0-100 etc. Nodes are distributed along these x axis domains and there can be edges connecting the nodes between different domains. To make it easier to understand, I share an image below that is similar to how I want x-axis to be, but with matplotlib and/or plotly. Plotly is more important. I label each node with their corresponding domain number, 12 or 13, in a dataframe column. An example is

Node 1 | Node 2 | y axis | x1 | domain
1534   945             20    22803603   13
945    946             10    32494954   12
946    -               9     32530403   12

where the edges are represented with Node1-Node2 and the rest of the columns belong to Node1. The last row doesn't connect to another node. Example code is

import networkx as nx
import matplotlib.pyplot as plt

fig = plt.figure(figsize=(10,10))
ax = fig.add_subplot(111)

G = nx.DiGraph()
G.add_node(1534,pos=(22803603,20))
G.add_node(945,pos=(32494954,10))
G.add_node(946,pos=(32530403,9))
G.add_edge(1534,945)
G.add_edge(945,946)

pos=nx.get_node_attributes(G,'pos')

nx.draw_networkx(G,pos)
ax.tick_params(left=True, right=True, top=True, bottom=True, labelleft=True, labelbottom=True)

In plotly, I follow the example here by just replacing random graph with the G above: https://plotly.com/python/network-graphs/#color-node-points

This is the example figure. I don't care about the shape of the nodes or the curvature of edges. All I care is the x-axis formatting.

Answer 1

I think you can achieve what you want with the following method. It requires knowing when the second (or more) domain starts.

• Define your graph and add domain index (starting at 0)

G = nx.DiGraph()
G.add_node(1534, pos=(22803603, 20), domain=0)
G.add_node(945, pos=(32494954, 10), domain=1)
G.add_node(946, pos=(32530403, 9), domain=1)
G.add_edge(1534, 945)
G.add_edge(945, 946)

• Define the domains starting X positions

d_starts = [0, 140_000_000]

• Get node positions relative to the domain

positions = nx.get_node_attributes(G, "pos")
domains = nx.get_node_attributes(G, "domain")

final_positions = {}

for pos, dom in zip(positions.items(), domains.items()):
    label, (x, y) = pos
    _, d = dom
    final_positions[label] = [x + d_starts[d], y]

• Finally, draw the graph on one axis and "fake" 2 axis with ticklabels and axvline

fig, ax = plt.subplots(figsize=(18,9))
nx.draw_networkx(G, final_positions, ax=ax)

# Works properly if d_starts values are modulo of the tick range step (here 10M)
max_xlim = int(ax.get_xlim()[1])
tick_range = range(0, max_xlim, 10_000_000)
labels = [f"{(i % d_starts[1]) / 1e6  :.0f}M" for i in tick_range]

ax.set_xlim(0, max_xlim)
ax.set_xticks(tick_range, labels , rotation=-45)
ax.tick_params(bottom=True, labelbottom=True)

for ypos in d_starts:
    ax.axvline(ypos, c="k")

Putting that all together here is the figure you obtain :

NB: You can easily add more domains if needed.
For your title, you'll need a map between the domain indexes and the domain names.