vqCenterInit
Find initial centers for VQ of k-means
Contents
Syntax
- center = vqCenterInit(data, clusterNum)
- center = vqCenterInit(data, clusterNum, method)
Description
center=vqCenterInit(data, clusterNum) returns the initial centers for k-means clustering.
center=vqCenterInit(data, clusterNum, method) uses the given method for computing the initial centers.
- method=1: Randomly pick data points as cluster centers
- method=2: Choose data points closest to the mean vector
- method=3: Choose data points furthest to the mean vector
- method=4: Choose the first few data as the centers
References
- M. Al-Daoud and S. Roberts, "New methods for the initialisation of clusters", Technical Report 94.34, School of Computer Studies, University of Leeds, 1994
- J. He, M. Lan, C.-L. Tan, S.-Y. Sung, and H.-B. Low, "Initialization of Cluster Refinement Algorithms: A Review and Comparative Study", Proc. IEEE Int. Joint Conf. Neural Networks, pp. 297-302, 2004.
Example
data=dcData(6); data=data.input; clusterNum=10; for i=1:7 method=i; center=vqCenterInit(data, clusterNum, method); subplot(3,3,i); plot(data(1,:), data(2,:), '.'); axis image; for i=1:clusterNum line(center(1,i), center(2,i), 'linestyle', 'none', 'marker', 'o', 'color', 'r'); end if method==1, title('Random centers'); end if method==2, title('Centers nearest to the mean'); end if method==3, title('Centers farthest to the mean'); end if method==4, title('Centers from the beginning few data points of the dataset'); end if method==5, title('A greedy selection for centers'); end if method==6, title('KKZ'); end if method==7, title('Kevin'); end end
