譜聚類演算法Matlab快速實現
阿新 • • 發佈:2019-01-02
%Ncut譜聚類完整函式定義(儲存為.m檔案):
function C = SpectralClustering(data,k,a) %data是資料點矩陣 K是聚類個數 a代表高斯核函式的引數
%UNTITLED Summary of this function goes here
% Detailed explanation goes here
d = pdist(data)
d2 = squareform(d)
d3 = d2.^2
W(:,:) = exp(-d3(:,:)/(2*a^2))
[n,m] = size(W);
s = sum(W)
D = full(sparse(1:n,1:n,s))
E = D^(-1/2)*W*D^(-1/2)
[X,B] = eig(E)
[Q,V] = eigs(E,k) %選的是前K個最大特徵值對應的特徵向量
C = kmeans(Q,k)
end
呼叫該函式時,首先載入資料,然後給定相關引數後直接呼叫:
load C:\Users\wangyuanw\Desktop\data\data1new.txt;
SpectralClustering(data1new,3,3);
備註:matlab相關函式功能可用help進行查詢
function C = SpectralClustering(data,k,a) %data是資料點矩陣 K是聚類個數 a代表高斯核函式的引數
%UNTITLED Summary of this function goes here
% Detailed explanation goes here
d = pdist(data)
d2 = squareform(d)
d3 = d2.^2
W(:,:) = exp(-d3(:,:)/(2*a^2))
[n,m] = size(W);
s = sum(W)
D = full(sparse(1:n,1:n,s))
E = D^(-1/2)*W*D^(-1/2)
[X,B] = eig(E)
[Q,V] = eigs(E,k) %選的是前K個最大特徵值對應的特徵向量
C = kmeans(Q,k)
end
呼叫該函式時,首先載入資料,然後給定相關引數後直接呼叫:
load C:\Users\wangyuanw\Desktop\data\data1new.txt;
SpectralClustering(data1new,3,3);
備註:matlab相關函式功能可用help進行查詢