使用 NetworkX 进行复杂网络分析与可视化
NetworkX 是 Python 生态中处理图结构数据的核心工具库，广泛应用于社交网络挖掘、路由优化、生物信息学等场景。本文将系统介绍其核心功能与实战技巧。
环境准备
pip install networkx matplotlibimport networkx as nx
import matplotlib.pyplot as plt图结构的构建与分类
NetworkX 支持四种基础图模型，分别对应不同的业务场景：
| 类型 | 类名 | 适用场景 |
|---|---|---|
| 无向图 | Graph | 双向关系（如好友关系） |
| 有向图 | DiGraph | 单向关系（如网页链接） |
| 多重无向图 | MultiGraph | 多通道连接（如航班线路） |
| 多重有向图 | MultiDiGraph | 多向多通道（如物流网络） |
# 初始化各类图实例
undirected = nx.Graph()
directed = nx.DiGraph()
multi_undirected = nx.MultiGraph()
multi_directed = nx.MultiDiGraph()节点与边的操作
图的元素支持任意可哈希对象作为标识，并允许挂载自定义属性：
# 批量添加节点并设置属性
undirected.add_nodes_from([
('node_1', {'category': 'source', 'priority': 1}),
('node_2', {'category': 'intermediate'}),
('node_3', {'category': 'target'})
])
# 创建带权重的连接
undirected.add_weighted_edges_from([
('node_1', 'node_2', 3.5),
('node_2', 'node_3', 2.8),
('node_1', 'node_3', 8.0) # 直连通道
], weight='distance')查询图的基本信息：
print(f"节点集合: {list(undirected.nodes())}")
print(f"边集合: {undirected.edges(data=True)}")
print(f"node_1 的邻接点: {list(undirected.adj['node_1'])}")
print(f"node_1 的度数: {undirected.degree['node_1']}")属性管理与元数据
属性系统支持节点、边、图三个层级的数据存储：
# 节点级属性访问与修改
undirected.nodes['node_1']['status'] = 'active'
print(undirected.nodes['node_1'].get('status', 'unknown'))
# 边级属性动态更新
undirected['node_1']['node_2']['traffic'] = 1200
# 图级元数据
undirected.graph['project'] = 'urban_mobility'
undirected.graph['version'] = '2.1.0'核心算法应用
路径优化
# 基于权重的最优路径
optimal_route = nx.shortest_path(
undirected,
source='node_1',
target='node_3',
weight='distance'
)
route_cost = nx.shortest_path_length(
undirected,
source='node_1',
target='node_3',
weight='distance'
)网络连通性分析
# 识别孤立子网络
subgraphs = [comp for comp in nx.connected_components(undirected)]
print(f"发现 {len(subgraphs)} 个连通分量")
# 检查特定节点的可达范围
reachable = nx.node_connected_component(undirected, 'node_1')中心性度量
centrality_metrics = {
'degree': nx.degree_centrality(undirected),
'closeness': nx.closeness_centrality(undirected),
'betweenness': nx.betweenness_centrality(undirected),
'eigenvector': nx.eigenvector_centrality(undirected, max_iter=1000)
}社区结构探测
from networkx.algorithms import community
# 使用 Girvan-Newman 算法进行层次化社区划分
hierarchy = community.girvan_newman(undirected)
first_partition = next(hierarchy)
print(f"第一层划分结果: {first_partition}")可视化定制
结合 Matplotlib 实现专业级网络可视化：
# 布局算法选择
layout = nx.kamada_kawai_layout(undirected) # 力导向布局
# 基于节点属性映射颜色
node_colors = [
'coral' if undirected.nodes[n].get('priority') == 1 else 'lightblue'
for n in undirected.nodes()
]
# 基于边权重映射宽度
edge_widths = [
d.get('distance', 1.0) * 0.8
for u, v, d in undirected.edges(data=True)
]
fig, ax = plt.subplots(figsize=(10, 8))
nx.draw_networkx_nodes(undirected, layout, node_color=node_colors, node_size=800, ax=ax)
nx.draw_networkx_edges(undirected, layout, width=edge_widths, alpha=0.6, ax=ax)
nx.draw_networkx_labels(undirected, layout, font_size=10, ax=ax)
ax.set_title("城市交通网络拓扑")
plt.tight_layout()
plt.show()数据持久化
# 导出为标准交换格式
nx.write_graphml(undirected, "network.graphml") # 通用格式
nx.write_gml(undirected, "network.gml") # 兼容 Cytoscape
nx.write_gexf(undirected, "network.gexf") # Gephi 专用
# 从文件恢复
loaded_network = nx.read_graphml("network.graphml")综合案例：供应链网络优化
以下示例构建一个包含工厂、仓库、零售节点的供应链网络，并进行关键节点识别：
# 构建供应链有向图
supply_chain = nx.DiGraph()
# 三层节点结构
factories = ['F1', 'F2', 'F3']
warehouses = ['W1', 'W2', 'W3', 'W4']
retailers = ['R1', 'R2', 'R3', 'R4', 'R5']
supply_chain.add_nodes_from(factories, layer='production', capacity=5000)
supply_chain.add_nodes_from(warehouses, layer='distribution', capacity=2000)
supply_chain.add_nodes_from(retailers, layer='sales', demand=800)
# 建立层级连接
factory_to_warehouse = [
('F1', 'W1', {'cost': 12, 'lead_time': 2}),
('F1', 'W2', {'cost': 15, 'lead_time': 3}),
('F2', 'W2', {'cost': 10, 'lead_time': 2}),
('F2', 'W3', {'cost': 14, 'lead_time': 4}),
('F3', 'W3', {'cost': 11, 'lead_time': 2}),
('F3', 'W4', {'cost': 13, 'lead_time': 3}),
]
warehouse_to_retail = [
('W1', 'R1', {'cost': 5, 'lead_time': 1}),
('W1', 'R2', {'cost': 6, 'lead_time': 1}),
('W2', 'R2', {'cost': 4, 'lead_time': 1}),
('W2', 'R3', {'cost': 7, 'lead_time': 2}),
('W3', 'R3', {'cost': 5, 'lead_time': 1}),
('W3', 'R4', {'cost': 6, 'lead_time': 2}),
('W4', 'R4', {'cost': 5, 'lead_time': 1}),
('W4', 'R5', {'cost': 8, 'lead_time': 2}),
]
supply_chain.add_edges_from(factory_to_warehouse)
supply_chain.add_edges_from(warehouse_to_retail)
# 识别关键中转节点（介数中心性）
hub_importance = nx.betweenness_centrality(supply_chain, weight='cost')
critical_hubs = sorted(hub_importance.items(), key=lambda x: x[1], reverse=True)[:3]
print(f"核心中转节点: {[n for n, _ in critical_hubs]}")
# 计算最小成本流
min_cost_flow = nx.min_cost_flow(supply_chain, demand='demand', capacity='capacity')性能优化建议
- 大规模网络（>10⁴节点）建议使用
graph-tool或igraph作为后端 - 频繁查询场景可预计算邻接矩阵转换为 SciPy 稀疏矩阵
- 并行计算可利用
joblib对独立子图进行分布式处理
