使用matplotlib

Question

这是第一次发布海报，所以请保持温柔。 ：）

我正在尝试在Networkx中绘制不同类型的字符网络，并希望为每种类型设置不同的节点形状。例如，我希望角色是圆形，生物是三角形等等。我试图想出这几个小时并且已经广泛搜索过了，但我还没有找到一种方法来实现这个目的而不是为每种类型的角色设置不同的node_lists并分别渲染它们，这看似违反直觉。

问题是我无法从以下内容访问node_shape字典值：

nx.draw_networkx_nodes(G, pos) 

我尝试了多种解决方案，包括尝试访问节点属性，创建外部字典或列表以及从调用中访问它，设置列表理解或迭代器，似乎没有任何工作。

要么我传递一个列表，一个是批量提取的，一个字典，该函数无法哈希，或者是一个列表的实例，如shape_list.pop(0)，在这种情况下，该函数只接受第一个值并将其应用于所有节点。

我可以通过创建一个单独的node_colors列表来设置颜色，该列表由函数迭代，甚至尝试创建一个字典，以便node_shape由node_color触发，但这也不起作用。

我希望将代码用作在Python 3.4和Django 1.8中开发的Web应用程序的附加组件，因此Graphviz不是一个选项。

提前感谢您对替代图书馆的任何帮助或参考。

这是我的代码：

import json
import requests
import networkx as nx
import matplotlib.pyplot as plt

personas = 'http://story-chronicles.herokuapp.com/storyobjects/'
target = requests.get(personas)
x = target.json()

story_objects = {}
labels = {}
node_colors = []

for character in x:
    name = character["name"]
    story = character["story"]
    c_type = character["c_type"]
    story_objects[name] = {}
    story_objects[name]['name'] = name
    story_objects[name]['story'] = story
    story_objects[name]['c_type'] = c_type
    story_objects[name]['to_relationships'] = []
    if character['c_type'] == "Character":
        story_objects[name]['node_shape'] = 'o'
        story_objects[name]['node_color'] = 'r'
    elif character['c_type'] == "Organization":
        story_objects[name]['node_shape'] = 'h'
        story_objects[name]['node_color'] = 'b'
    elif character['c_type'] == "Creature":
        story_objects[name]['node_shape'] = '^'
        story_objects[name]['node_color'] = 'g'
    elif character['c_type'] == "Force":
        story_objects[name]['node_shape'] = 'v'
        story_objects[name]['node_color'] = 'c'
    elif character['c_type'] == "Thing":
        story_objects[name]['node_shape'] = 's'
        story_objects[name]['node_color'] = 'y'

    for relationship in character["to_relationships"]:
        break_1 = relationship.find(">>")
        break_2 = relationship.find("weight:")
        sub_1 = relationship[0:break_1].strip()
        context = relationship[break_1:break_2]
        weight = relationship[break_2+8:-1]
        story_objects[name]['to_relationships'].append([sub_1, context, weight])

G=nx.MultiDiGraph()

for sub in story_objects:
    s = story_objects[sub]
    if s['story'] == "http://story-chronicles.herokuapp.com/story/1/":
        G.add_node(s['name'], node_shape=s['node_shape'])
        labels[s['name']] = s['name']

        node_colors.append(s['node_color'])

        print("***", s['name'], "***", s['c_type'])
        print("details:", s['node_color'], s['node_shape'])
        for i in s['to_relationships']:
            print('target:', i[0])
            print('context:', i[1])
            print('weight:', i[2])
            G.add_edge(s['name'], i[0], weight=int(i[2]))
        print("")

node_shapes=nx.get_node_attributes(G, 'node_shape') # Latest attempt at getting this to work
node_shapes = [v for k,v in node_shapes.items()]

pos=nx.spring_layout(G)
G.degree(weight=weight)

nx.draw_networkx_nodes(G, pos, node_color=node_colors, node_shape=node_shapes.pop(0)) # <--- This is where I'm having problems
nx.draw_networkx_edges(G, pos)
nx.draw_networkx_labels(G, pos, labels)

plt.show()

Answer 1

我担心这必须使用多次通过。

主要思想是使用layout获取节点的位置，然后对n个不同类别的节点重复使用draw_networkx_nodes。

例如：

import networkx
import pylab

#Build a graph (Node attribute 's' determines the node shape here)
G = networkx.Graph()
G.add_node(0, s="^", b=1)
G.add_node(1, s="^", b=2)

G.add_node(2, s="o", b=3)
G.add_node(3, s="o", b=4)

G.add_node(4, s="v", b=5)
G.add_node(5, s="v", b=6)

G.add_path([0,2,5])
G.add_path([1,4,3,0])
G.add_path([2,4,0,5])

#Drawing the graph
#First obtain the node positions using one of the layouts
nodePos = networkx.layout.spring_layout(G)

#The rest of the code here attempts to automate the whole process by
#first determining how many different node classes (according to
#attribute 's') exist in the node set and then repeatedly calling 
#draw_networkx_node for each. Perhaps this part can be optimised further.

#Get all distinct node classes according to the node shape attribute
nodeShapes = set((aShape[1]["s"] for aShape in G.nodes(data = True)))

#For each node class...
for aShape in nodeShapes:
    #...filter and draw the subset of nodes with the same symbol in the positions that are now known through the use of the layout.
    networkx.draw_networkx_nodes(G,nodePos,node_shape = aShape, nodelist = [sNode[0] for sNode in filter(lambda x: x[1]["s"]==aShape,G.nodes(data = True))])

#Finally, draw the edges between the nodes
networkx.draw_networkx_edges(G,nodePos)

#And show the final result
pylab.show()

最终结果如下所示：

希望这有帮助。

Answer 2

以下帖子略有变化（由A_A提供）。我使用了大致相同的东西只放了一条画线并包含一些标签，并显示了一个二分图。它可能对互联网上的其他人有用，所以我把它放在这里。enter image description here

import networkx as nx
% matplotlib inline
from networkx.algorithms import bipartite
B = nx.Graph()
B.add_nodes_from(['$x_1$','$x_2$','$x_3$'], s='o', bipartite=0) # Add the node attribute 'bipartite'
B.add_nodes_from(['$f_a$','$f_b$','$f_c$','$f_d$'], s='s', bipartite=1)
B.add_edges_from([('$x_1$','$f_a$'),('$x_1$','$f_b$'),('$x_2$','$f_a$'),('$x_2$','$f_b$'),('$x_2$','$f_c$'),('$x_3$','$f_c$'),('$x_3$','$f_d$')])

pos = dict()
X, Y = bipartite.sets(B)
pos.update((n, (i,1)) for i, n in enumerate(X))
pos.update((n, (i+0.5,2)) for i, n in enumerate(Y))
nodeShapes = set((aShape[1]["s"] for aShape in B.nodes(data=True)))

for aShape in nodeShapes:
    nx.draw(
        B, 
        pos, 
        with_labels=True, 
        node_shape = aShape, 
        node_color = '0.95', 
        node_size=1000, 
        nodelist = [
            sNode[0] for sNode in filter(lambda x: x[1]["s"]==aShape, B.nodes(data=True))
        ]
    )

plt.savefig("img/15_Graphical_Models_12b.png") # save as png