由于 numpy.float64 值，将 NetworkX 图导出为 graphml 会引发异常

Question

我正在使用以下代码分析加权化学反应网络：

import networkx as nx
import pandas as pd

edge_data = pd.read_table('VULCAN 800.dat', sep=',')
edge_list=[]
edge_data = edge_data.fillna(value=0.1)
col1=edge_data['Column_1']
col2=edge_data['Column_2']
col3=edge_data['Column_3']
col4=edge_data['Column_4']
col5=edge_data['Column_5']
col6=edge_data['Column_6']
for i in range(1,560):
    edge_list.append((col1.iloc[i],col2.iloc[i],col5.iloc[i]))
if pd.isnull(col3.iloc[i]) != True:
    edge_list.append((col1.iloc[i],col3.iloc[i],col5.iloc[i]))
    edge_list.append((col2.iloc[i],col3.iloc[i],col5.iloc[i]))
if pd.isnull(col4.iloc[i]) != True:
    edge_list.append((col1.iloc[i],col4.iloc[i],col5.iloc[i]))
    edge_list.append((col2.iloc[i],col4.iloc[i],col5.iloc[i]))
if pd.isnull(col3.iloc[i]) != True and pd.isnull(col4.iloc[i]) != True:
    edge_list.append((col3.iloc[i],col4.iloc[i],col5.iloc[i]))

G=nx.Graph()
G.add_weighted_edges_from(edge_list)
G.remove_node(0.1)
nx.write_graphml(G, '/home/tessa/Git/Network_biosignatures/VULCAN 800k.graphml')

我想将结果图导出为 graphml 文件，以便在 Cytoscape 中对其进行可视化，但由于某种原因，它会抛出错误消息：

Traceback (most recent call last):
 File "topo_measure_pn_hot jupiter_weighted.py", line 196, in <module>
nx.write_graphml(G, '/home/tessa/Git/Network_biosignatures/VULCAN 800k.graphml')
 File "<decorator-gen-202>", line 2, in write_graphml
 File "/usr/lib/python2.7/dist-packages/networkx/utils/decorators.py",      line 220, in _open_file
 result = func(*new_args, **kwargs)
 File "/usr/lib/python2.7/dist-packages/networkx/readwrite/graphml.py", line 82, in write_graphml
 writer.add_graph_element(G)
 File "/usr/lib/python2.7/dist-packages/networkx/readwrite/graphml.py", line 351, in add_graph_element
 self.add_edges(G,graph_element)
 File "/usr/lib/python2.7/dist-packages/networkx/readwrite/graphml.py", line 325, in add_edges
 self.add_attributes("edge", edge_element, data, default)
 File "/usr/lib/python2.7/dist-packages/networkx/readwrite/graphml.py", line 300, in add_attributes
 scope=scope, default=default_value)
 File "/usr/lib/python2.7/dist-packages/networkx/readwrite/graphml.py", line 288, in add_data
 '%s as data values.'%element_type)
 networkx.exception.NetworkXError: GraphML writer does not support <type 'numpy.float64'> as data values.

显然，graphml 编写器不能很好地处理除了简单的图形属性之外的任何事情，我猜是它的权重引发了错误。但是，我不确定 在哪里将权重值分配为 numpy.float64，或者如何将其从该分配中剥离。

正在读入的数据格式如下：

1,H,H2O,OH,H2,1.75116588E-16,,,,,,,,,,,,,,,,,,,,,,,,,,,
2,H,H2O,OH,H2,4.00975292E-13,,,,,,,,,,,,,,,,,,,,,,,,,,,
3,O,H2,OH,H,9.25180896E-14,,,,,,,,,,,,,,,,,,,,,,,,,,,
4,O,H2,OH,H,1.04176774E-13,,,,,,,,,,,,,,,,,,,,,,,,,,,
5,O,H2O,OH,OH,1.04560994E-15,,,,,,,,,,,,,,,,,,,,,,,,,,,
6,O,H2O,OH,OH,2.6959031E-12,,,,,,,,,,,,,,,,,,,,,,,,,,,

感谢任何和所有建议。谢谢！

Answer 1

这似乎是 networkx 中的一个错误。转换为基本的 python 类型可以解决这个问题。

import numpy as np
import networkx as nx

total_nodes = 20
nodes = range(total_nodes)

p = 0.1
total_edges = int(p * total_nodes ** 2 )
sources = np.random.choice(nodes, total_edges)
targets = np.random.choice(nodes, total_edges)
weights = np.random.rand(total_edges)

# edge_list = zip(sources, targets, weights) # doesn't work
edge_list = [(int(s), int(t), float(w)) for s, t, w in zip(sources, targets, weights)] # works

G = nx.Graph()
G.add_weighted_edges_from(edge_list)
nx.write_graphml(G, 'test.graphml')

编辑：

显然，他们很清楚这个问题 (https://github.com/networkx/networkx/issues/1556)，但还没有解决。