Networkx: how to specify multiple roots for plotting multiple trees at once?

Question

In networkx there is a function to plot a tree using a radial layout (graphviz's "twopi"):

import pydot
from networkx.drawing.nx_pydot import graphviz_layout

pos = graphviz_layout(G, prog='twopi', root=root, args='')

One can specify the root node with the argument root (which is added to args under the hood as args += f" -Groot={root}").

But how do you specify multiple roots when the graph consists of multiple disconnected components? I.e. a forest of trees.

I get the following plot without providing the root parameter:

As you can see visually, while it has correctly chosen the true root node for 10 of the trees, for 12 it has picked a child of the true root node as the centre (and hence some branches seem shallower than they actually are, relative to other branches).

How do you manually specify the roots for multiple trees?

You might take a look at this for a purely networkx solution: https://stackoverflow.com/a/29597209/2966723 It would need some adaptation, but I think it's straightforward. — Joel, May 01 '20 at 23:10

yatu · Accepted Answer · 2020-05-01T18:36:27.187

I don't think there is a way to do this in a single graph with graphviz's twopi layout. Twopi should do a good job in general in setting the root nodes of each subgraph, since as mentioned in the docs it will randomly pick one of the nodes that are furthest from a leaf node as root, so in the case of having a single root node, this should lead to the expected topological arrangement. Though if that is not the case, and you want to set manually the roots for each subgraph, the way I'd approach this is by iterating over the graphs connected component subgraphs, and plotting each component into an separate axis in a subplot graph, creating a custom graphviz_layout for each.

Here's how this could be done using the following example graph:

from matplotlib import pyplot as plt
import pygraphviz
from networkx.drawing.nx_agraph import graphviz_layout

result_set = {('plant','tree'), ('tree','oak'), ('flower', 'rose'), ('flower','daisy'), ('plant','flower'), ('tree','pine'), ('plant','roots'), ('animal','fish'),('animal','bird'), ('bird','robin'), ('bird','falcon'), ('animal', 'homo'),('homo','homo-sapiens'), ('animal','reptile'), ('reptile','snake'),('fungi','mushroom'), ('fungi','mold'), ('fungi','toadstool'),('reptile','crocodile'), ('mushroom','Portabello'), ('mushroom','Shiitake'),('pine','roig'),('pine','pinyer'), ('tree','eucaliptus'),('rose','Floribunda'),('rose','grandiflora')}
G=nx.from_edgelist(result_set, create_using=nx.DiGraph)

In order to iterate over the existing subgraphs we must create a copy of the current graph as an undirected graph if it isn't one already, and create a list of subgraphs using nx.connected_component_subgraphs:

UG = G.to_undirected()
subgraphs = list(nx.connected_component_subgraphs(UG))

Let's say we know that we want the root nodes of the different components to be the nodes 'plant', 'animal' and 'mushroom', we can now create a set of subplots, and iterate over the respective axes, along with the subgraph objects and the list of roots (making sure they are in the same order), creating a new layout for each subgraph setting the corresponding root nodes:

n_cols = 2
roots = ['plant','animal','mushroom']
fig, axes = plt.subplots(nrows=int(np.ceil(len(subgraphs)/n_cols)), 
                         ncols=n_cols, 
                         figsize=(15,10))
plt.box(False)

for subgraph, root, ax in zip(subgraphs, roots, axes.flatten()):
    pos = graphviz_layout(G,  prog='twopi', args=f"-Groot={root}")
    nx.draw(subgraph, pos=pos, with_labels=True, 
            node_color='lightblue', node_size=500, ax=ax)

Thanks, yatu - this was indeed my workaround solution but I wanted to maintain the relative lengths of edges between graphs (as is kept when they are plotted simultaneously, but if the trees have different 'depths' they will be scaled differently when plotted separately, as shown in your figure). But I realised I can just force them to maintain the same scale with `subplots(sharex=True, sharey=True)`. — iacob, May 01 '20 at 18:44

Networkx: how to specify multiple roots for plotting multiple trees at once?

1 Answers1