R/initSmallEM.R

   1 #' initialization of the EM algorithm
   2 #'
   3 #' @param k number of components
   4 #' @param X matrix of covariates (of size n*p)
   5 #' @param Y matrix of responses (of size n*m)
   6 #' @param tau threshold to stop EM algorithm
   7 #'
   8 #' @return a list with phiInit, rhoInit, piInit, gamInit
   9 #' @export
  10 initSmallEM = function(k,X,Y,tau)
  11 {
  12     n = nrow(Y)
  13     m = ncol(Y)
  14     p = ncol(X)
  15
  16     Zinit1 = array(0, dim=c(n,20)) #doute sur la taille
  17     betaInit1 = array(0, dim=c(p,m,k,20))
  18     sigmaInit1 = array(0, dim = c(m,m,k,20))
  19     phiInit1 = array(0, dim = c(p,m,k,20))
  20     rhoInit1 = array(0, dim = c(m,m,k,20))
  21     Gam = matrix(0, n, k)
  22     piInit1 = matrix(0,20,k)
  23     gamInit1 = array(0, dim=c(n,k,20))
  24     LLFinit1 = list()
  25
  26     require(MASS) #Moore-Penrose generalized inverse of matrix
  27     require(mclust) # K-means with selection of K
  28     for(repet in 1:20)
  29     {
  30         clusters = Mclust(X,k) #default distance : euclidean  #Mclust(matrix(c(X,Y)),k)
  31         Zinit1[,repet] = clusters$classification
  32
  33         for(r in 1:k)
  34         {
  35             Z = Zinit1[,repet]
  36             Z_bin = vec_bin(Z,r)
  37             Z_vec = Z_bin$vec #vecteur 0 et 1 aux endroits o? Z==r
  38             Z_indice = Z_bin$indice #renvoit les indices o? Z==r
  39
  40             betaInit1[,,r,repet] = ginv(t(X[Z_indice,])%*%X[Z_indice,])%*%t(X[Z_indice,])%*%Y[Z_indice,]
  41             sigmaInit1[,,r,repet] = diag(m)
  42             phiInit1[,,r,repet] = betaInit1[,,r,repet]#/sigmaInit1[,,r,repet]
  43             rhoInit1[,,r,repet] = solve(sigmaInit1[,,r,repet])
  44             piInit1[repet,r] = sum(Z_vec)/n
  45         }
  46
  47         for(i in 1:n)
  48         {
  49             for(r in 1:k)
  50             {
  51                 dotProduct = 3 #(Y[i,]%*%rhoInit1[,,r,repet]-X[i,]%*%phiInit1[,,r,repet]) %*% (Y[i,]%*%rhoInit1[,,r,repet]-X[i,]%*%phiInit1[,,r,repet])
  52                 Gam[i,r] = piInit1[repet,r]*det(rhoInit1[,,r,repet])*exp(-0.5*dotProduct)
  53             }
  54             sumGamI = sum(Gam[i,])
  55             gamInit1[i,,repet]= Gam[i,] / sumGamI
  56         }
  57
  58         miniInit = 10
  59         maxiInit = 11
  60
  61         new_EMG = .Call("EMGLLF",phiInit1[,,,repet],rhoInit1[,,,repet],piInit1[repet,],
  62                                         gamInit1[,,repet],miniInit,maxiInit,1,0,X,Y,tau)
  63         LLFEessai = new_EMG$LLF
  64         LLFinit1[repet] = LLFEessai[length(LLFEessai)]
  65     }
  66
  67     b = which.max(LLFinit1)
  68     phiInit = phiInit1[,,,b]
  69     rhoInit = rhoInit1[,,,b]
  70     piInit = piInit1[b,]
  71     gamInit = gamInit1[,,b]
  72
  73     return (list(phiInit=phiInit, rhoInit=rhoInit, piInit=piInit, gamInit=gamInit))
  74 }