cluster.reg: Clustering analysis of regression coefficients.

In RegClust: Cluster analysis via regression coefficients.

Description

This package performs clustering on regression coefficients using the methods of clustering through linear regression models (CLM) (Qin and Self 2006). Maximum likelihood approach is used to infer the parameters for each cluster. Bayesian information criterion (BIC) combined with Bootstrapped maximum volume (BMV) criterion are used to determine the number of clusters. The Bootstrap method is used to estimate the uncertainty on the number of clusters.

Usage

cluster.reg(Y, X, loop = 1000)

Arguments

Y	An n x k data matrix for n number of observations and k dependent variables.
X	An n x m data matrix for n number of observations and m covariates of interest. Restriction of m < n.
loop	A numeric scalar identfying the number of iterations for the bootstrappng process. Default is 1000.

Value

cluster	A numeric vector of length k identifying which cluster the respective dependent variables belong to.
parm	Dataframe containing estimates of regression coefficients, &sigma^2;, and π for each cluster, where &sigma^2; is the variance in the random error, and π is the probability that a variable is in a cluster.
likelihood	Likelihood at final iteration or at convergence.
BIC	Bayesian information criterion at final interation or at convergence.

Note

Although the number of covariates is unlimited, it is recommended to only allow up to 3 covariates to prevent potential difficulty in clustering the variables.

Author(s)

Weichao Bao, Xin Tong, Meredith Ray, Hongmei Zhang

References

Qin, Li-Xuan, and Steven G. Self. The clustering of regression models method with applications in gene expression data. Biometrics 62.2 (2006): 526-533.

Examples

## Not run: 
beta0=1
beta1=0.02
beta2=-0.04
beta3=0.1
set.seed(1234)
sim=200
age=runif(sim, min=18, max=70)
Rerror=rnorm(sim,-3,3)

y1=matrix(NA,sim,4,dimnames = list(NULL,c("y1", "y2", "y3", "y4")))
for (g in 1:4){
set.seed(1234)
y1[,g]=beta0+beta1*runif(sim, min=18, max=70)+beta1*rnorm(sim,-3,3)
set.seed(1134+g)
y1[,g]=y1[,g]+rnorm(sim,0,1)
}
y2=matrix(NA,sim,5,dimnames = list(NULL,c("y5", "y6", "y7","y8","y9")))
for (g in 1:5){
set.seed(1234)
y2[,g]=beta0+beta2*runif(sim, min=18, max=70)+beta2*rnorm(sim,-3,3)
set.seed(2234+g)
y2[,g]=y2[,g]+rnorm(sim,0,0.5^0.5)
}
y3=matrix(NA,sim,6,dimnames = list(NULL,c("y10", "y11", "y12","y13","y14","y15")))
for (g in 1:6){
set.seed(1234)
y3[,g]=beta0+beta3*runif(sim, min=18, max=70)+beta3*rnorm(sim,-3,3)
set.seed(3334+g)
y3[,g]=y3[,g]+rnorm(sim,0,1)
}
X=data.frame(round(cbind(Rerror=Rerror,age=age),2))
Y=data.frame(round(cbind(y1,y2,y3),2))

run<-cluster.reg(Y,X)
run

## End(Not run)

Example output

Converged at iteration  4 , BIC= -8050.398 
$cluster
     y1 y2 y3 y4 y5 y6 y7 y8 y9 y10 y11 y12 y13 y14 y15
[1,]  1  1  1  1  2  2  2  2  2   2   2   2   2   2   2

$param
          Intercept     Rerror        age     sigma        pi
cluster 1 1.0657278 0.01482447 0.01896318 0.9957545 0.2666667
cluster 2 0.9485024 0.02880514 0.03706263 9.8203637 0.7333333

$likelihood
[1] -4019.901

$BIC
[1] -8050.398

RegClust documentation built on May 2, 2019, 5:56 a.m.