networkx怎么显示图_显示带有标签的networkx图

I'm trying to create a labeled graph using networkx but am having trouble getting the nodes and labels to turn out correctly. In short, the labels don't line up over the right nodes and there are some nodes which have no edges when displayed.

First I created a graph, added nodes and edges, then added labels.

The graph data comes from a pandas DataFrame object with two columns, employee and manager names:

emp_name mgr_name

0 Marianne Becker None

1 Evan Abbott Marianne Becker

2 Jay Page Marianne Becker

3 Seth Reese Marianne Becker

4 Maxine Collier Marianne Becker

...

Each node is a name and the edges are the mgr_name to emp_name relationship.

My graph code:

import networkx as nx

G=nx.DiGraph()

#set layout

pos=nx.spring_layout(G)

#add nodes

G.add_nodes_from(df.emp_name)

G.nodes()

G.add_node('None')

#create tuples for edges

subset = df[['mgr_name','emp_name']]

tuples = [tuple(x) for x in subset.values]

#add edges

G.add_edges_from(tuples)

G.number_of_edges()

#draw graph

import matplotlib.pyplot as plt

nx.draw(G, labels = True)

plt.show()

Ideally I would have a tree-like structure with employee names as the labels for each of the nodes.

Output image is

解决方案

Networkx has a number of functions to draw graphs but also allow the user fine control over the whole process.

draw is basic and its docstring specifically mentions:

Draw the graph as a simple representation with no nodeabels or edge

labels and using the full Matplotlib figure areas labels by default.

See draw_networkx() for more fatured drawing that allows title, axis

labels

The functions prefixed by draw_networkx followed by edges, nodes, edge_labels and edge_nodes allow finer control over the whole drawing process.

Your example worked fine when using draw_networkx.

In addition, if you are looking for an output that resembles an organogram, I would suggest the use of graphviz through networkx. Graphviz's dot is ideal for this kind of diagrams (please also see this for dot).

In what follows, I have tried to modify your code slightly to demonstrate the use of both functions:

import networkx as nx

import matplotlib.pyplot as plt

import pandas

#Build the dataset

df = pandas.DataFrame({'emp_name':pandas.Series(['Marianne Becker', 'Evan Abbott', 'Jay Page', 'Seth Reese', 'Maxine Collier'], index=[0,1,2,3,4]), 'mgr_name':pandas.Series(['None', 'Marianne Becker', 'Marianne Becker', 'Marianne Becker', 'Marianne Becker'], index = [0,1,2,3,4])})

#Build the graph

G=nx.DiGraph()

G.add_nodes_from(df.emp_name)

G.nodes()

G.add_node('None')

#

#Over here, you are manually adding 'None' but in reality

#your nodes are the unique entries of the concatenated

#columns, i.e. emp_name, mgr_name. You could achieve this by

#doing something like

#

#G.add_nodes_from(list(set(list(D.emp_name.values) + list(D.mgr_name.values))))

#

# Which does exactly that, retrieves the contents of the two columns

#concatenates them and then selects the unique names by turning the

#combined list into a set.

#Add edges

subset = df[['mgr_name','emp_name']]

tuples = [tuple(x) for x in subset.values]

G.add_edges_from(tuples)

G.number_of_edges()

#Perform Graph Drawing

#A star network (sort of)

nx.draw_networkx(G)

plt.show()

t = raw_input()

#A tree network (sort of)

nx.draw_graphviz(G, prog = 'dot')

plt.show()

You could also try using graphviz's dot from the command line directly, by saving your networkx network via nx.write_dot. To do this:

From within your python script:

nx.write_dot(G, 'test.dot')

After this, from your (linux) command line and assuming that you have graphviz installed:

dot test.dot -Tpng>test_output.png

feh test_output.png #Feh is just an image viewer.

firefox test_output.png & #In case you don't have feh installed.

For a more typical organogram format, you can force orthogonal edge routing by

dot test.dot -Tpng -Gsplines=ortho>test_output.png

Finally, here are the outputs

Output of draw_networkx

Output of draw_graphviz

Output of dot without orthogonal edges

Output of dot with orthogonal edges

Hope this helps.