Link_Predictors.py 文件源码

def structure_dependent_index(G, ebunch=None):
    if ebunch is None:
        ebunch = nx.non_edges(G)

    #C = nx.average_clustering(G)
    #d = nx.average_shortest_path_length(G)
    path_range = max(2, math.ceil(nx.average_shortest_path_length(G)))
    #print path_range

    def predict(u, v):
        #NeighborSet = nx.all_neighbors(G, u)
        #len( sorted(nx.common_neighbors(G, u, v) ))
        SD_Index = {}
        #Generate all simple paths in the graph G from source to target,  length <= cutoff .
        paths = list( nx.all_simple_paths(G, source=u, target=v, cutoff = path_range))
        print paths
        for path in paths:
            if SD_Index.has_key( len(path) ):
                SD_Index[len(path)] = SD_Index[len(path)] + 1.0
            else:
                SD_Index[len(path)] = 1.0
        #end for
        print SD_Index

        #Sum up the num of paths with different length
        Coefficient = 0.6
        SD_Value = 0.0
        key_Sequence = list(sorted(SD_Index.keys()))
        for key in key_Sequence:
            if key != 2:
                SD_Value = SD_Value + math.pow(Coefficient, key-2.0) * SD_Index[key]
        #end for
        return  SD_Value #Coefficient = 0.6

    Rank_List = []
    for u, v in ebunch:
        Rank_List.append((u, v, predict(u, v)))
    return Rank_List #((u, v, predict(u, v)) for u, v in ebunch)



##======================================================================##