尽管有无穷大规则，我怎样才能得到稀疏图的连通分量的偏心？

Question

我有一个scipy Compressed Sparse Row（CSR）矩阵，我试图从中提取偏心率以查看信息传播的平均距离。不幸的是，在使用networkx将其转换为networkx图后使用networkx.convert_matrix.from_scipy_sparse_matrix（https://networkx.github.io/documentation/latest/reference/generated/networkx.convert_matrix.from_scipy_sparse_matrix.html）时，我一直无限。

有没有办法可以将连接组件生成的标签集转换回原始值，然后对它们执行单独的偏心公式？

Answer 1

由于图形偏心率是最大最短路径距离，因此使用scipy稀疏矩阵运算可能更容易，更快捷：

import numpy as np
from scipy.sparse.csgraph import connected_components, shortest_path
from scipy.sparse import csr_matrix

def sparse_component_eccentricity(graph, directed=False):

    n_components, labels = connected_components(csgraph=graph, directed=directed, return_labels=True)

    component_eccentricity = np.zeros(graph.shape[0])

    for icomp in range(n_components):
        subgraph_indices = np.where(labels == icomp)[0]
        subgraph = graph[subgraph_indices][:,subgraph_indices]
        dist_matrix = shortest_path(subgraph, directed=directed)
        component_eccentricity[subgraph_indices] = np.nanmax(dist_matrix, axis=1)
    return component_eccentricity