R/generateSampleInputs.R

   1 #' Generate a sample of (X,Y) of size n
   2 #' @param meanX matrix of group means for covariates (of size p*K)
   3 #' @param covX covariance for covariates (of size p*p*K)
   4 #' @param covY covariance for the response vector (of size m*m*K)
   5 #' @param pi     proportion for each cluster
   6 #' @param beta regression matrix
   7 #' @param n     sample size
   8 #'
   9 #' @return list with X and Y
  10 #' @export
  11 generateXY = function(meanX, covX, covY, pi, beta, n)
  12 {
  13     p = dim(covX)[1]
  14     m = dim(covY)[1]
  15     k = dim(covX)[3]
  16
  17     X = matrix(nrow=n,ncol=p)
  18     Y = matrix(nrow=n,ncol=m)
  19
  20     require(MASS) #simulate from a multivariate normal distribution
  21     for (i in 1:n)
  22     {
  23         class = sample(1:k, 1, prob=pi)
  24         X[i,] = mvrnorm(1, meanX[,class], covX[,,class])
  25         Y[i,] = mvrnorm(1, X[i,] %*% beta[,,class], covY[,,class])
  26     }
  27
  28     return (list(X=X,Y=Y))
  29 }
  30
  31 #' Generate a sample of (X,Y) of size n with default values
  32 #' @param n sample size
  33 #' @param p number of covariates
  34 #' @param m size of the response
  35 #' @param k number of clusters
  36 #' @return list with X and Y
  37 #' @export
  38 generateXYdefault = function(n, p, m, k)
  39 {
  40     rangeX = 100
  41     meanX = rangeX * matrix(1 - 2*runif(p*k), ncol=k)
  42     covX = array(dim=c(p,p,k))
  43     covY = array(dim=c(m,m,k))
  44     for(r in 1:k)
  45     {
  46         covX[,,r] = diag(p)
  47         covY[,,r] = diag(m)
  48     }
  49     pi = rep(1./k,k)
  50     #initialize beta to a random number of non-zero random value
  51     beta = array(0, dim=c(p,m,k))
  52     for (j in 1:p)
  53     {
  54         nonZeroCount = sample(1:m, 1)
  55         beta[j,1:nonZeroCount,] = matrix(runif(nonZeroCount*k), ncol=k)
  56     }
  57
  58     sample_IO = generateXY(meanX, covX, covY, pi, beta, n)
  59     return (list(X=sample_IO$X,Y=sample_IO$Y))
  60 }
  61
  62 #' Initialize the parameters in a basic way (zero for the conditional mean, uniform for weights,
  63 #' identity for covariance matrices, and uniformly distributed for the clustering)
  64 #' @param n sample size
  65 #' @param p number of covariates
  66 #' @param m size of the response
  67 #' @param k number of clusters
  68 #' @return list with phiInit, rhoInit,piInit,gamInit
  69 #' @export
  70 basicInitParameters = function(n,p,m,k)
  71 {
  72     phiInit = array(0, dim=c(p,m,k))
  73
  74     piInit = (1./k)*rep(1,k)
  75
  76     rhoInit = array(dim=c(m,m,k))
  77     for (i in 1:k)
  78         rhoInit[,,i] = diag(m)
  79
  80     gamInit = 0.1 * matrix(1, nrow=n, ncol=k)
  81     R = sample(1:k, n, replace=TRUE)
  82     for (i in 1:n)
  83         gamInit[i,R[i]] = 0.9
  84     gamInit = gamInit/sum(gamInit[1,])
  85
  86     return (list("phiInit" = phiInit, "rhoInit" = rhoInit, "piInit" = piInit, "gamInit" = gamInit))
  87 }