Question

我使用2包生成了G 大型网络G1和networkx。我想计算所有节点之间的 jaccard相似性索引。

一种可能的方法如下：

def returnJaccardNetworks(G, G1):
    tmp =   list(G.nodes())
    tmp1 =  list(G1.nodes())
    tmp2 =  np.unique([tmp, tmp1]) ### Find nodes in the networks
    jc = []
    for i in tmp2:
    ## if the node i is in G and in G1 compute 
    ## the similarity between the lists of the ajacent nodes
    ## otherwise append 0
        if (i in G) and (i in G1):  
            k1 = list(G[i]) ## adjacent nodes of i in the network G     
            k2 = list(G1[i]) ## adjacent nodes of i in the network G1 
            ### Start Jaccard Similarity
            intersect = list(set(k1) & set(k2))
            n = len(intersect)
            jc.append(n / float(len(k1) + len(k2) - n))
            ### End Jaccard Similariy
        else:
            jc.append(0)
    return jc

我想知道是否有更有效的方法。我注意到包中有一个名为jaccard_coefficient的函数，但我不确定它是如何工作的。

https://networkx.github.io/documentation/networkx-1.9/reference/generated/networkx.algorithms.link_prediction.jaccard_coefficient.html

Answer 1

你的实施非常有效（尽管不是很漂亮，IMO）。我可以使用此版本在我的机器上节省15％的执行时间：

def get_jaccard_coefficients(G, H):
    for v in G:
        if v in H:
            n = set(G[v]) # neighbors of v in G
            m = set(H[v]) # neighbors of v in H
            length_intersection = len(n & m)
            length_union = len(n) + len(m) - length_intersection
            yield v, float(length_intersection) / length_union
        else:
            yield v, 0. # should really yield v, None as measure is not defined for these nodes

这个版本更紧凑，更易于维护，但执行时间增加了 30％

：

def get_jaccard_coefficients(G, H): for v in set(G.nodes) & set(H.nodes): # i.e. the intersection n = set(G[v]) # neighbors of v in G m = set(H[v]) # neighbors of v in H yield v, len(n & m) / float(len(n | m))

Python：两个网络之间的jaccard similartity？

1 个答案: