ExampleData: Illustrative Example

In brodsa/IClust: Imbalanced clustering

Description

A simple illustrative data example constisting of 4 highly imbalanced groups. The groups are described be a large number of variables - both informative and noise. The informative variables follows a finite mixture of Gaussian distributions.

Usage

data("ExampleData")

Format

Two objects are uploaded.

data

A data frame consisting of standardized values with 343 observations and 300 variables. The observations form 4 groups of highly different size. The group structure is given by the first 100 variables and the remaining variables represent noise.

label

A numeric vector reflecting a group memberhip of each observation.

Details

The code generating data is visible in example, see below.

Examples

data(ExampleData)
dim(data)
table(label)

## Not run: 
 require(MASS)     # multivariate normal distribution
 require(mixAK)    # rotation matrix
 require(ggplot2)  # a grafical tool
 
 # dimensionality of data set
 p_inf <- 100 
 p_noise <- 200
 n_k <- c(300,30,10,3)
 
 # covariances of groups
 sigma1 <- matrix(0.3,nrow=p_inf,ncol=p_inf)
 diag(sigma1) <- 1
 set.seed(12)
 R <- rRotationMatrix(n=1, dim=p_inf)
 sigma1 <- R 
 
 sigma2 <- matrix(0.0,nrow=p_inf,ncol=p_inf)
 diag(sigma1) <- 5
 set.seed(35)
 R <- rRotationMatrix(n=1, dim=p_inf)
 sigma2 <- R 
 
 sigma3 <- diag(p_inf)
 diag(sigma3) <- 2
 set.seed(68)
 R <- rRotationMatrix(n=1, dim=p_inf)
 sigma3 <- R 
 
 sigma4 <- diag(p_inf)
 diag(sigma4) <- 3
 set.seed(98)
 R <- rRotationMatrix(n=1, dim=p)
 sigma4 <- R 
 
 # groups follow Gaussion distribution
 g1 <- mvrnorm(n_k[1],mu=rep(0,p), Sigma=sigma1)
 g2 <- mvrnorm(n_k[2],mu=rep(c(0,-3),p/2),Sigma=sigma2)
 g3 <- mvrnorm(n_k[3],mu=rep(-3,p),Sigma=sigma3)
 g4 <- mvrnorm(n_k[4],mu=rep(c(-5,5),p/2),Sigma=sigma4)
 X <- rbind(cl1,cl2,cl3,cl4)
 
 # group membership
 label <- c(rep(1,n_k[1]),rep(2,n_k[2]),rep(3,n_k[3]),rep(4,n_k[4])) 
 # additing noise part
 X <- cbind(X, matrix(rnorm(p_noise * sum(n_k), mean = 0), ncol = p_noise))
 data <- scale(X)

## End(Not run)

brodsa/IClust documentation built on April 7, 2020, 10:41 a.m.