链路预测代码

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%% 链路预测 ---基于相似性的链路预测算法 %%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%% (1)基于局部结构-共同邻居 (2)基于路径 (3)基于随机游走的算法 (4)其他方法 %%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%% 对每个数据集： step-1 划分训练集和测试集 %%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%% step-2 基于此划分评估所有算法的精度（AUC） %%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%% 重复前面两个过程100次 并取平均值和方差 %%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% clear; %% 参数设定—— 设定训练集的比例 和 独立实验的次数 ratioTrain = 0.9; %训练集比例 numOfExperiment = 2; %独立实验的次数 %% 用到的数据集名称 dataname = strvcat('USAir','NS','PB','Yeast','Celegans','FWFB','Power','Router'); datapath = 'D:\research\completed\Link prediction\writting book\data\'; %数据集所在的路径 % 结果的格式 % 第一行：算法名 % 倒数第一行：每种算法在多次实验中的方差； % 倒数第二行：每种算法在多次实验中的平均值； % 中间数据：每种算法在多次实验中的具体值。 %% 链路预测过程 for ith_data = 1:2 % 遍历每一个数据 tempcont = strcat('正在处理第 ', int2str(ith_data), '个数据...', dataname(ith_data,:)); disp(tempcont); tic; thisdatapath = strcat(datapath,dataname(ith_data,:),'.txt'); % 第ith个数据的路径 linklist = load(thisdatapath); % 导入数据（边的list） net = FormNet(linklist); clear linklist; % 根据边的list构成邻接矩阵 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%% 每个数据做 numOfExperiment（100）次独立实验， 并将所有结果存到（实验次数×预测器个数）阶的矩阵中用以计算均值和方差 aucOfallPredictor = []; % 用于存储100次实验的结果，第j行表示第j次实验 PredictorsName = []; % 记录预测器的顺序 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %----- 开始100次实验的循环 for ith_experiment = 1:numOfExperiment if mod(ith_experiment,10) == 0 tempcont = strcat(int2str(ith_experiment),'%... '); disp(tempcont); end %%% step-1 划分训练集和测试集，保证训练集的连通性 [train, test] = DivideNet(net,ratioTrain); % 划分训练集和测试集，返回上三角矩阵 train = sparse(train); test = sparse(test); train = spones(train + train'); test = spones(test+test'); ithAUCvector = []; Predictors = []; % 用于存储当前实验中所有预测器的精度 %%% step-2 根据train set计算test set和nonexistent set中所有节点对产生（或存在）连边的可能性，并得出AUC %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%% 首先是基于Common Neighbor（共同邻居）的相似性算法 disp('CN...'); tempauc = CN(train, test); % Common Neighbor Predictors = [Predictors '%CN ']; ithAUCvector = [ithAUCvector tempauc]; disp('Salton...'); tempauc = Salton(train, test); % Salton Index Predictors = [Predictors 'Salton ']; ithAUCvector = [ithAUCvector tempauc]; disp('Jaccard...'); tempauc = Jaccard(train, test); % Jaccard Index Predictors = [Predictors 'Jaccard ']; ithAUCvector = [ithAUCvector tempauc]; disp('Sorenson...'); tempauc = Sorenson(train, test); % Sorenson Index Predictors = [Predictors 'Sorens ']; ithAUCvector = [ithAUCvector tempauc]; disp('HPI...'); tempauc = HPI(train, test); % Hub Promoted Index Predictors = [Predictors 'HPI ']; ithAUCvector = [ithAUCvector tempauc]; disp('HDI...'); tempauc = HDI(train, test); % Hub Depressed Index Predictors = [Predictors 'HDI ']; ithAUCvector = [ithAUCvector tempauc]; disp('LHN...'); tempauc = LHN(train, test); % Leicht-Holme-Newman Predictors = [Predictors 'LHN ']; ithAUCvector = [ithAUCvector tempauc]; disp('AA...'); tempauc = AA(train, test); % Adar-Adamic Index Predictors = [Predictors 'AA ']; ithAUCvector = [ithAUCvector tempauc]; disp('RA...'); tempauc = RA(train, test); % Resourse Allocation Predictors = [Predictors 'RA ']; ithAUCvector = [ithAUCvector tempauc]; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%% 偏好连接相似性算法 disp('PA...'); tempauc = PA(train, test); % Preferential Attachment Predictors = [Predictors 'PA ']; ithAUCvector = [ithAUCvector tempauc]; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%% 局部朴素贝叶斯模型 disp('LNBCN...'); tempauc = LNBCN(train, test); % Local naive bayes method - Common Neighbor Predictors = [Predictors 'LNBCN ']; ithAUCvector = [ithAUCvector tempauc]; disp('LNBAA...'); tempauc = LNBAA(train, test); % Local naive bayes method - Adar-Adamic Index Predictors = [Predictors 'LNBAA ']; ithAUCvector = [ithAUCvector tempauc]; disp('LNBRA...'); tempauc = LNBRA(train, test); % Local naive bayes method - Resource Allocation Predictors = [Predictors 'LNBRA ']; ithAUCvector = [ithAUCvector tempauc]; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%% 基于路径的相似性指标 disp('LocalPath...'); tempauc = LocalPath(train, test, 0.0001); % Local Path Index Predictors = [Predictors 'LocalP ']; ithAUCvector = [ithAUCvector tempauc]; disp('Katz 0.01...'); tempauc = Katz(train, test, 0.01); % Katz Index 参数取0.01 Predictors = [Predictors 'Katz.01 ']; ithAUCvector = [ithAUCvector tempauc]; disp('Katz 0.001...'); tempauc = Katz(train, test, 0.001); % Katz Index 参数取0.001 Predictors = [Predictors '~.001 ']; ithAUCvector = [ithAUCvector tempauc]; disp('LHNII 0.9...'); tempauc = LHNII(train, test, 0.9); % Leicht-Holme-Newman II 参数取0.9 Predictors = [Predictors 'LHNII.9 ']; ithAUCvector = [ithAUCvector tempauc]; disp('LHNII 0.95...'); tempauc = LHNII(train, test, 0.95); % Leicht-Holme-Newman II 参数取0.95 Predictors = [Predictors '~.95 ']; ithAUCvector = [ithAUCvector tempauc]; disp('LHNII 0.99...'); tempauc = LHNII(train, test, 0.99); % Leicht-Holme-Newman II 参数取0.99 Predictors = [Predictors '~.99 ']; ithAUCvector = [ithAUCvector tempauc]; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%