msm: Bayesian Mixture of Subspaces of Different Dimensions

In kyoustat/mosub: Learning Mixtures of Subspaces

Description

msm is a Bayesian model inferring mixtures of subspaces that are of possibly different dimensions. For simplicity, this function returns only a handful of information that are most important in representing the mixture model, including projection, location, and hard assignment parameters.

Usage

msm(
  X,
  K = 2,
  iter = 496,
  prop.var = 1,
  temperature = 1e-06,
  burn.in = round(iter/2),
  thin = 10,
  print.progress = TRUE
)

Arguments

X	an (n\times p) data matrix.
K	the number of mixtures.
iter	the number of MCMC runs.
prop.var	proposal variance parameter.
temperature	temperature value for Gibbs posterior.
burn.in	burn-in for MCMC runs.
thin	interval for recording MCMC runs
print.progress	a logical; TRUE to show completion of iterations by 10, FALSE otherwise.

Value

a list whose elements are also lists of following elements:

P

length-K list of projection matrices.

U

length-K list of orthonormal basis.

theta

length-K list of center locations of each mixture.

cluster

length-n vector of cluster label.

References

\insertRef

thomas_learning_2015mosub

Examples

## generate a toy example
set.seed(10)
tester = gen.LP(n=100, K=2, iso.var=0.1)
data   = tester$data
label  = tester$class

## do PCA for data reduction
proj = base::eigen(stats::cov(data))$vectors[,1:2]
dat2 = data%*%proj

## run MSM algorithm with K=2, 3, and 4
maxiter = 1000
output2 = msm(data, K=2, iter=maxiter)
output3 = msm(data, K=3, iter=maxiter)
output4 = msm(data, K=4, iter=maxiter)

## extract final clustering information
nrec  = length(output2)
finc2 = output2[[nrec]]$cluster
finc3 = output3[[nrec]]$cluster
finc4 = output4[[nrec]]$cluster

## visualize
opar <- par(mfrow=c(3,4))
plot(dat2[,1],dat2[,2],pch=19,cex=0.3,col=finc2+1,main="K=2:PCA")
plot(data[,1],data[,2],pch=19,cex=0.3,col=finc2+1,main="K=2:Axis(1,2)")
plot(data[,1],data[,3],pch=19,cex=0.3,col=finc2+1,main="K=2:Axis(1,3)")
plot(data[,2],data[,3],pch=19,cex=0.3,col=finc2+1,main="K=2:Axis(2,3)")

plot(dat2[,1],dat2[,2],pch=19,cex=0.3,col=finc3+1,main="K=3:PCA")
plot(data[,1],data[,2],pch=19,cex=0.3,col=finc3+1,main="K=3:Axis(1,2)")
plot(data[,1],data[,3],pch=19,cex=0.3,col=finc3+1,main="K=3:Axis(1,3)")
plot(data[,2],data[,3],pch=19,cex=0.3,col=finc3+1,main="K=3:Axis(2,3)")

plot(dat2[,1],dat2[,2],pch=19,cex=0.3,col=finc4+1,main="K=4:PCA")
plot(data[,1],data[,2],pch=19,cex=0.3,col=finc4+1,main="K=4:Axis(1,2)")
plot(data[,1],data[,3],pch=19,cex=0.3,col=finc4+1,main="K=4:Axis(1,3)")
plot(data[,2],data[,3],pch=19,cex=0.3,col=finc4+1,main="K=4:Axis(2,3)")
par(opar)

kyoustat/mosub documentation built on Feb. 25, 2020, 3:52 a.m.