\(k\)-means++ algorithm is known to be a smart, careful initialization technique. It is originally intended to return a set of \(k\) points as initial centers though it can still be used as a rough clustering algorithm by assigning points to the nearest points.
kmeanspp(data, k = 2)Arguments
- data
an \((n\times p)\) matrix whose rows are observations.
- k
the number of clusters.
Value
a length-\(n\) vector of class labels.
References
Arthur D, Vassilvitskii S (2007). “K-Means++: The Advantages of Careful Seeding.” In In Proceedings of the 18th Annual ACM-SIAM Symposium on Discrete Algorithms.
Examples
## use simple example of iris dataset
data(iris)
mydata = as.matrix(iris[,1:4])
mycol = as.factor(iris[,5])
## find the low-dimensional embedding for visualization
my2d = cmds(mydata, ndim=2)$embed
## apply 'kmeanspp' with different numbers of k's.
k2 = kmeanspp(mydata, k=2)
k3 = kmeanspp(mydata, k=3)
k4 = kmeanspp(mydata, k=4)
k5 = kmeanspp(mydata, k=5)
k6 = kmeanspp(mydata, k=6)
## visualize
opar <- par(no.readonly=TRUE)
par(mfrow=c(2,3))
plot(my2d, col=k2, main="k=2", pch=19, cex=0.5)
plot(my2d, col=k3, main="k=3", pch=19, cex=0.5)
plot(my2d, col=k4, main="k=4", pch=19, cex=0.5)
plot(my2d, col=k5, main="k=5", pch=19, cex=0.5)
plot(my2d, col=k6, main="k=6", pch=19, cex=0.5)
plot(my2d, col=mycol, main="true cluster", pch=19, cex=0.5)
par(opar)