This is a Matlab implementation of our paper "Fast Multi-view Graph-based Clustering via Hierarchical Initialization and Supercluster Similarity Minimization". Code will be updated after paper is published. The proposed HISSM model includes two steps named as Hierarchical Initialization(HI) and Supercluster Similarity Minimization(SSM).
The main body of code see figure
- Y_Initialize: different initial methods.
- Results_y_ini: record the results of initial label of proposed HI method.
- Results_y: record the results of refined label of proposed SSM method.
- Results_y_compare: record the results after HI and SSM on all testing datasets.
- funs: include some used functions.
- Dataset: include some testing datasets.
First, set runtimes and test datasets in Run_HISSM.m
runtimes = 1; % running times on each dataset, default: 1
dataname = {'MSRCV1','COIL20-3v','3Sources','HW2sources','ORL','BBC','BRCA','Hdigit','yaleA'};
Then, run Run_HISSM.m. ALL the results in the paper are obtained simultaneously and recorded as the following forms.
- For y_ini and y, we record the results of each dataset separately. The results include accuracy (ACC),Normalized Mutual Information(NMI),Purity(Pu)
,Fscore,Precision(Pre),Recall(Rec)，Adjusted Rand Index (ARI)，average value (AVE.) of different running and Standard deviation (Std.) of different running. Suppose the runtimes = 2 is set, the data record is shown in the figure.
- For better comparing y_ini and y, we also record the results of all dataset together. The data record is shown in the figure.
- For time cost, we record the results of each dataset separately. The data record is shown in the figure.
