kmeans_val.m

来自「動態聚類k-means演算 將輸入在程式中的數據資料 給予適當的分群」· M 代码 · 共 49 行

function [val, db] = kmeans_val(data, clusters, nc);

%

% Compute the distance between each datapoint to the clusters
dists = kmeans_dist(data, clusters);

% Assign datapoints to clusters and compute updates
[minval, dbelong] = min(dists');

% Intra mean distance
intra = 0.0;
for i=1:nc
  idxtmp = find(dbelong == i);
  
  if length(idxtmp > 0)
    cpoints = data(idxtmp,:);
    ctmp = clusters(i,:);
    cdists = kmeans_dist(cpoints, ctmp);
    intra = intra + sum(cdists);
    err(i) = mean(cdists);
  else
    err(i) = 0.0;
  end  
end

% Normalize with number of data
[ndata, nvar] = size(data);
intra = intra / ndata;

% Inter distance
interdist = dist(clusters');
tmp = find(interdist > 0.0);
inter = min(interdist(tmp));

val = intra / inter;

% Compute the Davies-Bouldin index
Rk = zeros(nc);
for i=1:nc
  for j=1:nc
    if( j ~= i )
      R(i,j) =(err(i) + err(j))/ interdist(i,j);
    end
  end
end
Rk = max(R);
db = sum(Rk) / nc;