R/Anthropometry-internalTDDclust.R

Defines functions NNDDVQEstart NNDDVQE NNDDVQA1 pamsil DDcalc2 DDapplyloo DDfcnadj DDapply Weiszfeldadj Weisziteradj initW2 unitvec l1norm Distvec rowweight is.empty

Documented in DDapply DDapplyloo DDcalc2 DDfcnadj Distvec initW2 is.empty l1norm NNDDVQA1 NNDDVQE NNDDVQEstart pamsil rowweight unitvec Weiszfeldadj Weisziteradj

#assign("last.warning", NULL, envir = baseenv())

is.empty <- function(x, mode=NULL){
  if (is.null(mode)) mode <- class(x)
  identical(vector(mode,1),c(x,vector(class(x),1)))
}

rowweight <- function(X,a){ #Product function. 
  xm <- X * a 
  return(xm) 
} 

Distvec <- function(Xmatr,y){ #L1 distance from y to all data points. 
  Dv <- apply(Xmatr,1,l1norm,b=y) 
  return(Dv) 
} 

l1norm <- function(a,b){ #L1 distance from a to b. 
  l1 <- sqrt(sum((a-b)^2)) 
  return(l1) 
} 

unitvec <- function(a,b){ #Unit vector from a to b. 
  
  l1 <- sqrt(sum((a-b)^2)) 
  
  if (l1 > 10^(-10)){ 
    uv <- (b-a)/l1 
  }
  
  if (l1 < 10^(-10)){ 
    uv <- matrix(0,1,length(a)) 
  } 
  
  return(uv) 
} 


initW2 <- function(Xmatr,Nvec){ #Initial y for Weiszfeld algorithm. 
  #Xmatr = cell data matrix, Nmat = multiplicities. 
  dimX <- dim(Xmatr) 
  mm <- apply(Xmatr,2,mean) 
  ystart <- mm 
  return(ystart) 
} 


Weisziteradj <- function(Xmatr,Nvec,ystart,B){ #Iterates the Weiszfeld step B times # starting value ystart. 
  
  y <- ystart 
  
  for(k in (1:B)){ 
    W <- Weiszfeldadj(Xmatr,Nvec,y) 
    y <- W$y 
  } 
  
  DDy <- W$DDy 
  
  return(list(y=y,DDy=DDy))
}


Weiszfeldadj <- function(Xmatr,Nvec,ystart){ 
  #Xmatr distinct data matrix, Nvec vector of multiplicities, ystart is the starting point. 
  #Also gives data depth of new y. 
  l <- rbind(Xmatr,ystart)
  y <- ystart 
  
  lcorr <- cor(t(l)) 
  diag(lcorr) <- 0 
  uc <- upper.tri(lcorr) 
  lcorr[uc==0] <- 0 
  lcorr[abs(lcorr-1) < 10^(-10)] <- 1 
  lcorr[abs(lcorr-1) > 10^(-10)] <- 0 
  ccl <- apply(lcorr,1,sum) #Default for y weight. 
  
  ny <- 0 
  
  if (max(ccl)==1){ #y is in the set of observations. 
    cpp <-ccl[1:(length(ccl)-1)] 
    ny <- Nvec[cpp==1] 
  } 
  
  dvec <- Distvec(Xmatr,y) #Distance from Xmatr elements to y. 
  
  dind <- seq(1,length(dvec)) 
  
  dinds <- dind[abs(dvec) > 10^(-10)]  
  
  wvec <- matrix(0,1,dim(Xmatr)[1]) 
  
  wvec[dinds] <- (Nvec[dinds] / dvec[dinds]) 
  
  wvec <- wvec / sum(wvec) 
  
  Ttilde <- apply(apply(Xmatr,2,rowweight,a=wvec),2,sum) 
  
  unitvecs <- t(apply(Xmatr,1,unitvec,b=y)) 
  
  wunitvecs <- apply(unitvecs,2,rowweight,a=Nvec) 
  
  ry <- sqrt(sum(apply(wunitvecs,2,sum)^2)) 
  
  ynew <- Ttilde * max(0,(1-ny/ry)) + y * min(1,ny/ry) 
  
  unitvecs <- t(apply(Xmatr,1,unitvec,b=ynew)) 
  
  ebar <- sqrt(sum(apply(apply(unitvecs,2,rowweight,a=Nvec/sum(Nvec)),2,sum)^2)) 
  
  l <- rbind(ynew,Xmatr) 
  lcorr <- cor(t(l)) 
  diag(lcorr) <- 0 
  uc<-upper.tri(lcorr) 
  lcorr[uc==0] <- 0 
  lcorr[abs(lcorr-1) < 10^(-10)] <- 1 
  lcorr[abs(lcorr-1) > 10^(-10)] <- 0 
  ccl <- apply(lcorr,1,sum) 
  
  ny<-0 
  
  if(max(ccl)==1){ 
    ny<-Nvec[ccl==1] 
  } 
  if(ny==0){ 
    DDy <- 1 - ebar
  } 
  if(ny>0) { 
    DDy <- 1 - max(0,(ebar-ny/sum(Nvec))) 
    #DDy <- 1 - ebar
  } 
  y<-ynew 
  
  return(list(y=y,DDy=DDy))
} 


##TO RUN THE PROGRAM## 
#yr <- initW2(Xmatr,Nvecr) 
#W <- Weisziteradj(Xmatr,Nvecr,yr,B) #Usually enough to run B=5, can check by DDy to see that it is equal to 1. 
#Y <- W$Y #This is the multivariate median. 
#Computing depth of an observations with respect to a cluster. 

DDapply <- function(Xv,Xmat) { #Computes the depths of a vector of observations Xv with respect to
#a cloud of observations Xmat. 
  Nvec <- rep(1,dim(Xmat)[1]) 
  dd <- DDfcnadj(Xmat,Nvec,Xv) 
  return(dd) 
} 

DDfcnadj <- function(Xmatr,Nvec,y){ #Computes the data depth of point y in cloud Xmatr. 
 
  aux <- apply(Xmatr,1,unitvec,b=y)
  #if(class(aux) == "list"){
  #  unitvecs <- t(matrix(unlist(aux),dim(Xmatr)[1],dim(Xmatr)[2])) 
  #}else{
  #  unitvecs <- t(aux) 
  #}
  if(inherits(aux, "list")){
    unitvecs <- t(matrix(unlist(aux),dim(Xmatr)[1],dim(Xmatr)[2])) 
  }else{
    unitvecs <- t(aux) 
  }
  
  #unitvec<-function(a,b){ #Unit vector from a to b. 
  #l1<-sqrt(sum((a-b)^2)) 
  #if (l1 > 10^(-10)){ #10^(-10) 
  # uv <- (b-a)/l1 
  #}
  # if (l1 < 10^(-10)){ 
  #  uv <- matrix(0,1,length(a)) 
  # } 
  # return(uv) 
  #} 
  
  gvv = matrix(as.numeric(unitvecs), dim(unitvecs)[1], dim(unitvecs)[2])
 
  a <- Nvec/sum(Nvec)  
  ebar <- sqrt( sum ( apply( rowweight(gvv,a) ,2,sum )^2))   
  
  #rowweight<-function(X,a){ #Product function 
  # xm<-X*a 
  # return(xm) 
  #}
  
  l <- rbind(y,Xmatr) 
  
  lcorr <- cor(t(l)) 
  
  diag(lcorr) <- 0 
  
  uc <- upper.tri(lcorr) 
  
  lcorr[uc==0] <- 0 
  lcorr[abs(lcorr-1) < 10^(-10)] <- 1 
  lcorr[abs(lcorr-1) > 10^(-10)] <- 0 
  ccl <- apply(lcorr,1,sum) 
  
  ny <- 0 
  
  if(max(ccl)==1){ 
    ny <- sum(Nvec[ccl==1])
  } 
  
  if(ny == 0){ 
    DDy <- 1-ebar
  }
  
  if(ny > 0){ 
    DDy <- 1-max(0,(ebar-ny/sum(Nvec))) 
    #DDy <- 1-ebar 
  } 
  return(DDy) 
} 

DDapplyloo <- function(Xv,Xmat){ 
  
  dl <- apply(Xmat,1,l1norm,b=Xv) 
  
  Xmatu <- Xmat[dl>0,] 
  
  Nvec <- rep(1,dim(Xmatu)[1]) 
  
  dd <- DDfcnadj(Xmatu,Nvec,Xv) 
  
  return(dd) 
} 

##TO RUN## 
#x is a vector, X is a group of data,N is the vector of multiplicities.
#Dx <- apply(x,1,DDapply,Xmat=X,Nvec=N) #Depth of a single observation x. 
#Dx <- DDfcnadj(Xmat,Nvec,x) 


DDcalc2 <- function(DDi,NN,K,Km,norm=1){ 
  
  n <- NN[1,] 
  nvec <- matrix(0,K,1) 
  ovec <- matrix(0,length(n),1) 
  svec <- matrix(0,length(n),1) 
  indvec <- seq(1,length(n)) 
  normvec <- matrix(0,K,1) 
  
  for(kz in (1:K)){ 
    nvec[kz] <- length(n[n==kz]) 
    Vin <- DDi[n==kz,kz] 
    normvec[kz] <- mean(Vin) 
  } 
  if(norm==0){ 
    normvec <- normvec/normvec
  } 
  DDb <- matrix(0,length(n),1) 
  DDw <- DDb 
  DD <- DDb 
  Vt <- 0 
  VZD <- DDi 
  
  for(ml in (1:length(n))){ 
    if(ml==1){ 
      Vt <- DDi[ml,c(NN[2:Km,ml])] / normvec[c(NN[2:Km,ml])]
      
      VZD[ml,] <- DDi[ml,] / normvec[c(NN[,ml])] 
    }
    if(ml>1){ 
      Vt <- c(Vt,DDi[ml,c(NN[2:Km,ml])] / normvec[c(NN[2:Km,ml])]) 
      
      VZD[ml,] <- DDi[ml,] / normvec[c(NN[,ml])] 
    }
  } 
  TT <- Vt[rev(sort.list(Vt))[length(NN[1,])+1]] #VZD2<-VZD #VZD[VZD<-0 
  
  for(kt in (1:length(n))){ 
    Nv <- NN[2:Km,kt]
    
    Vv <- sum(VZD[kt,Nv]) 
    DDw[kt] <- DDi[kt,NN[1,kt]] / normvec[NN[1,kt]] 
    DDb[kt] <- VZD[kt,NN[2,kt]] / normvec[c(NN[2,kt])] 
    ti <- DDi[kt,NN[1,kt]] / normvec[NN[1,kt]]
    DD[kt] <- (ti-Vv) 
  } 
  
  return(list(DDw=DDw,DD=DD))
} 


pamsil <- function(X,pc,K) { #Silwidth calculation. 
  
  sil <- matrix(0,length(pc),1) 
  
  Kvec <- seq(min(pc):max(pc)) 
  
  Nuvec <- pc 
  
  ccl <- rep(0,length(pc)) 
  
  clvec <- matrix(0,length(pc),K) 
  
  clvec[,1] <- pc 
  
  clvec2 <- clvec 
  
  indvec <- seq(1:length(Nuvec)) 
  
  for(kt in (1:length(pc))){ 
    kk <- Nuvec[kt] 
    
    kc <- setdiff(indvec,kt) 
    
    Nt <- Nuvec[kc] 
    
    ai <- mean(Distvec(X[kc[Nt==kk],],X[kt,])) 
    
    Ks <- setdiff(Kvec,kk) 
    
    for(kz in (1:length(Ks))){ 
      if(kz==1){ 
        bi <- mean(Distvec(X[Nuvec==Ks[kz],],X[kt,])) 
        
        ccl[kt] <- Ks[kz] 
        
        clvec2[kt,2] <- bi 
        
        clvec[kt,2] <- Ks[kz] 
      } 
      if(kz>1){ 
        nnd <- mean(Distvec(X[Nuvec==Ks[kz],],X[kt,])) 
        if(bi>nnd){ 
          ccl[kt] <- Ks[kz] 
        } 
        bi <- min(bi,nnd) 
        clvec2[kt,kz+1] <- nnd 
        clvec[kt,kz+1] <- Ks[kz] 
      } 
    } 
    
    sili <- (bi-ai) / max(bi,ai) 
    
    sil[kt] <- sili 
  } 
  clvec <- clvec[,2:K] 
  clvec2 <- clvec2[,2:K] 
  
  if(K>2){ 
    Sm <- apply(clvec2,1,sort.list) 
    clvec3 <- clvec 
    
    for(kk in (1:length(Nuvec))){ 
      clvec3[kk,] <- clvec[kk,Sm[,kk]] 
    } 
    clvec <- clvec3 
  } 

  return(list(sil=sil,ccl=ccl,clvec=clvec)) 
} 


NNDDVQA1 <- function(X,pcc,lambda,norm){

  #X is the data matrix. 
  #pcc particion actual.
  #MM is the matrix of multivariate medians. 
  #T is current temperature. 
  
  K <- dim(pcc$med)[1] 
  Km <- 2 
  Kmat <- matrix(0,dim(X)[1],K) 
  
  for(kk in (1:K)){ 
    Kmat[,kk]<-apply(X,1,l1norm,b=pcc$med[kk,]) 
  } 
  
  Smat <- apply(Kmat,1,sort.list) 
  
  Kmat <- apply(Kmat,1,sort) 
  NNuse <- Smat           
  
  Nvec <- rep(1,dim(X)[1]) 
  DDi <- matrix(0,dim(X)[1],K) 
  
  for(kz in (1:K)){ 
    for(ky in (1:dim(X)[1])){ 
      Xmatr <- X[NNuse[1,] == kz,]
      
      Nvecr <- Nvec[NNuse[1,] == kz] 
      
      DDi[ky,kz] <- DDfcnadj(Xmatr,Nvecr,X[ky,]) 
    } 
  } 
  
  DD <- DDcalc2(DDi,NNuse,K,Km,norm) 
    
  pcc <- pamsil(X,NNuse[1,],K)   
  
  Kmata <- pcc$sil * (1 - lambda) + lambda * DD$DD 
  
  NN <- NNuse 
  NN[2:K,] <- t(pcc$clvec) 
  
  Nuvec <- NNuse[1,] 
  
  Kmatb <- 0 
  prs <- 0 
  stopnow <- 0 
  
  return(list(NN=NN,DDi=DDi,DD=DD,pcc=pcc,Kmata=Kmata,Nuvec=Nuvec,Kmatb=Kmatb,prs=prs,stopnow=stopnow)) 
} 


NNDDVQE <- function(X,MM,prT,lambda,NNold,Km,Th,Trimm,kl,NNold_old,verbose){ 
  
  #X is the data matrix. 
  #MM is the matrix of multivariate medians. 
  #Th is current temperature. 
  
  change <- c()
  iter_while <- 0
  improv <- c()
  norm <- 0 
  K <- dim(MM)[1] 
  
  if(K <= 3){ 
    Km <- 2 
  } 
    
  aux <- sort(NNold$Kmata)
  
  tri <- Trimm*(dim(X)[1])
  if(tri < 1){
   tri1 <- ceiling(tri)  
  }else{
    tri1 <- round(tri)
  }
  
  phi <- aux[abs(tri1)]
  trimm <- which(NNold$Kmata <= phi)
  if(verbose){
   cat("Trimmed women: ")
   print(trimm)
   cat("\n")
  } 
  X <- X[-trimm,]
  
  sil <- as.matrix(NNold$pcc$sil[-trimm,])
  ccl <- NNold$pcc$ccl[-trimm]
  clvec <- as.matrix(NNold$pcc$clvec[-trimm,])
  pcc_aux <- list(sil=sil,ccl=ccl,clvec=clvec)  
  
  pcc <- pcc_aux
  NNuse <- NNold$NN[,-trimm]
  NNuse[2:K,] <- t(pcc$clvec) 
  
  Nvec <- rep(1,dim(X)[1])
  DDi <- NNold$DDi[-trimm,] 
  DDw <- as.matrix(NNold$DD$DDw[-trimm,])
  DD <- as.matrix(NNold$DD$DD[-trimm,])
  DD <- list(DDw=DDw,DD=DD)
  
  Kmata <- pcc$sil * (1 - lambda) + lambda * DD$DD 
  
  if((kl-1) == 0){
    if(verbose){
     cat("Initial value of the partition:")
     print(mean(Kmata))
     cat("\n")
    } 
  }else{
    if(verbose){
     cat("Initial value of the partition:")
     print(NNold$Cost)
     cat("\n")
    } 
  } 
  
  Th <- as.numeric(quantile(Kmata,0.25))
  
  Kmat0 <- Kmata 
  Kmatb <- Kmata
  Kmatb[Kmata > Th] <- 1 
  Kmatb[Kmata <= Th] <- 0 
  
  NNnew <- NNuse 
  NNusea <- NNuse 
  indvec <- seq(1,dim(X)[1]) 
  
  Iuse <- indvec[Kmatb==0] #Elements that can be moved. 
  Iuse <- setdiff(Iuse,trimm)
  if(verbose){
   cat("Set S of observations that can be relocated:")
   print(Iuse)
   cat("\n")
  }
  ct <- 0 
  nostop <- 0 
  nb <- min(50,length(Iuse)) 
  
  if(verbose){
   cat("While loop starts: \n")
  } 
  while(length(Iuse)>0 & nostop<=nb){ 
    iter_while <- iter_while + 1
    if(length(Iuse)==1){ 
      iuse <- Iuse
      if(verbose){
       cat("Random subset E of S:")
       print(iuse)
      } 
    } 
    if(length(Iuse)>1){ 
      E <- round(runif(1,min=1,max=min(10,length(Iuse)))) 
      iuse <- sample(Iuse,E) 
      if(verbose){
       cat("Random subset E of S:")
       print(iuse)
      } 
    } 
    for(zz in (1:length(iuse))){ 
      NNusea[1,iuse[zz]] <- NNuse[2,iuse[zz]] 
      NNusea[2,iuse[zz]] <- NNuse[1,iuse[zz]] 
    } 
    DDia <- DDi-DDi 
    
    for(kz in (1:K)){ 
      Xmatr <- X[NNusea[1,]==kz,] 
      Nvecr <- Nvec[NNusea[1,]==kz] 
      DDia[,kz] <- apply(X,1,DDapply,Xmat=Xmatr) 
    } 
    
    DDa <- DDcalc2(DDia,NNusea,K,Km,norm) 
    pcca <- pamsil(X,NNusea[1,],K) 
    if(verbose){
     cat("Candidate partition:")
     print(table(NNusea[1,]))
     cat("\n")
    } 
    Kmataa <- pcca$sil * (1 - lambda) + lambda * DDa$DD #Value of new partition. 
    
    if((kl-1) == 0){
      Del <- mean(Kmataa) - mean(Kmata) 
    }else{
      Del <- mean(Kmataa) - NNold$Cost
    }
    
    if(Del == "NaN"){
     stop("This partition cannot be evaluated \n")  
     break
    }
    
    if((kl-1) == 0){
      if(verbose){
       cat("Value of the previous partition C(I_1^{K}):")
       print(mean(Kmata))
      } 
    }else{
      if(verbose){
       cat("Value of the previous partition C(I_1^{K}):")
       print(NNold$Cost)
      } 
    } 
    
    if(verbose){
     cat("Value of the new partition C(tilde{I}_1^{K}):")
     print(mean(Kmataa))
     cat("Current Delta value:")
     print(Del)
     cat("\n")
    }
    #Stopping criterion:
    if(Del > 0){
      if((Del / mean(Kmataa)) < 0.01){
        if(verbose){
         cat("STOPPING CRITERION FULFILLED")
         cat("\n")
        }
         change[iter_while] <- 0
        break         
      }
    } 
    
    pru <- runif(1)
    
    if(prT>0){
      PRT <- exp(Del/prT) 
      if(verbose){
       cat("Probability of mis-allocation:")
       print(PRT)
       cat("\n")
      }  
    } 
    if(prT==0){ 
      PRT <- 0 
      if(verbose){
       cat("Probability of mis-allocation:")
       print(PRT)
       cat("\n")
      } 
    } 

    if(Del>0 | (Del<=0 & pru==PRT)){ 
      NNuse <- NNusea
      Kmata <- Kmataa
      if(verbose){
       cat("The partition is accepted.\n")
       cat("New value of the partition:")
       print(mean(Kmata))
       cat("Partition:\n")
       print(table(NNuse[1,]))
      }
       NNold$Cost <- mean(Kmata)
      if(verbose){
       cat("\n")
      }
       improv[kl] <- kl - 1 
       change[iter_while] <- 1
    }else{
      NNold$Cost <- NNold$Cost  
      if(verbose){
       cat("The partition is rejected.\n")
       cat("Current value of the partition:")
       print(NNold$Cost)
       cat("\n")
      } 
      change[iter_while] <- 0
    } 
    nostop <- nostop+1 
  }
  if(verbose){
   cat("While loop ends: \n")
  }
  if(max(change) == 1){
    if(ct==0){ 
      stopnow <- 1 
    } 
    if(ct>0){ 
      stopnow <- 0 
    } 
    prs <- ct/length(NNuse[1,]) 
    NNnew <- NNuse 
    NN <- NNnew 
    Nuvec <- NNnew[1,] 
    
    DDia <- matrix(0,nrow=dim(X)[1],ncol=K)
    for(kz in (1:K)){ 
      Xmatr <- X[NNuse[1,]==kz,] 
      Nvecr <- Nvec[NNuse[1,]==kz] 
      DDia[,kz]<-apply(X,1,DDapply,Xmat=Xmatr) 
    } 
    
    DDa <- DDcalc2(DDia,NNuse,K,Km,norm) 
    pcc <- pamsil(X,NNuse[1,],K) 
    
    return(list(Iuse=Iuse,ct=ct,NN=NN,DDi=DDia,DD=DDa,pcc=pcc,Kmata=Kmata,Kmat0=Kmat0,Nuvec=Nuvec,Kmatb=Kmatb,
                prs=prs,stopnow=stopnow,Cost=NNold$Cost,trimmed=trimm,improv=improv))
  }else if(max(change) == 0 | is.null(change)){
    NNold_old <- NNold_old
    if((kl-1) == 0){
      NNold_old$Cost <- mean(Kmata)
    }else{
      NNold_old$Cost <- NNold$Cost
     } 
    NNold_old$ct <- ct
    NNold_old$trimmed <- trimm
    NNold_old$improv <- improv
    return(NNold_old)
   } 
} 


NNDDVQEstart <- function(X,MM,prT,lambda,NNold,Km,Th,Trimm,kl,NNold_old,verbose){ 
  #X is the data matrix. 
  #MM is the matrix of multivariate medians. 
  #Th is current temperature. 
   
  change <- c()
  iter_while <- 0
  improv <- c()
  norm <- 0 
  K <- dim(MM)[1] 
  
  if(K < 3){ 
    Km <- 2 
  } 
  
  
  aux <- sort(NNold$Kmata)
  
  tri <- Trimm*(dim(X)[1])
  if(tri < 1){
    tri1 <- ceiling(tri)  
  }else{
    tri1 <- round(tri)
  }
  
  phi <- aux[abs(tri1)]
  trimm <- which(NNold$Kmata <= phi)
  if(verbose){
   cat("Trimmed women: ")
   print(trimm)
   cat("\n")
  }
   X <- X[-trimm,]
  
  sil <- as.matrix(NNold$pcc$sil[-trimm,])
  ccl <- NNold$pcc$ccl[-trimm]
  clvec <- as.matrix(NNold$pcc$clvec[-trimm,])
  pcc_aux <- list(sil=sil,ccl=ccl,clvec=clvec)  
  
  pcc <- pcc_aux
  NNuse <- NNold$NN[,-trimm]
  NNuse[2:K,] <- t(pcc$clvec) 

  Nvec <- rep(1,dim(X)[1])
  DDi <- NNold$DDi[-trimm,] 
  DDw <- as.matrix(NNold$DD$DDw[-trimm,])
  DD <- as.matrix(NNold$DD$DD[-trimm,])
  DD <- list(DDw=DDw,DD=DD)
  
  Kmata <- pcc$sil * (1 - lambda) + lambda * DD$DD 
  
  if((kl-1) == 0){
    if(verbose){
     cat("Initial value of the partition:")
     print(mean(Kmata))
     cat("\n")
    } 
  }else{
    if(verbose){
     cat("Initial value of the partition:")
     print(NNold$Cost)
     cat("\n")
    }
  } 
  
  Th <- as.numeric(quantile(Kmata,0.25))
  
  Kmat0 <- Kmata 
  Kmatb <- Kmata
  Kmatb[Kmata > Th] <- 1 
  Kmatb[Kmata <= Th] <- 0 
  
  NNnew <- NNuse 
  NNusea <- NNuse 
  indvec <- seq(1,dim(X)[1]) 
  
  Iuse <- indvec[Kmatb==0] #Elements that can be moved. 
  Iuse <- setdiff(Iuse,trimm)
  if(verbose){
   cat("Set S of observations that can be relocated:")
   print(Iuse)
   cat("\n")
  }
  ct <- 0 
  fir <- 0 
  nostop <- 0 
  nb <- min(50,length(Iuse)) 
  
  if(verbose){
   cat("While loop starts:\n")
  }
  while(length(Iuse)>0 & nostop<=nb){
    iter_while <- iter_while + 1
    if(length(Iuse)==1){ 
      iuse <- Iuse 
      if(verbose){
       cat("Random subset E of S:")
       print(iuse)
      } 
    } 
    if(length(Iuse)>1){ 
      E <- round(runif(1,min=1,max=min(5,length(Iuse)))) 
      iuse <- sample(Iuse,E) 
      if(verbose){
       cat("Random subset E of S:")
       print(iuse)
      }  
     } 
    for(zz in (1:length(iuse))){ 
      NNusea[1,iuse[zz]] <- NNuse[2,iuse[zz]]
      NNusea[2,iuse[zz]] <- NNuse[1,iuse[zz]] 
    } 
    DDia <- DDi-DDi 
    
    for(kz in (1:K)){ 
      Xmatr <- X[NNusea[1,]==kz,] 
      Nvecr <- Nvec[NNusea[1,]==kz] 
      DDia[,kz]<-apply(X,1,DDapply,Xmat=Xmatr) 
    } 
    
    DDa <- DDcalc2(DDia,NNusea,K,Km,norm) 
    pcca <- pamsil(X,NNusea[1,],K) 
    if(verbose){
     cat("Candidate partition:")
     print(table(NNusea[1,]))
     cat("\n")
    }
     Kmataa <- pcca$sil * (1 - lambda) + lambda * DDa$DD 
    
    if((kl-1) == 0){
      Del <- mean(Kmataa) - mean(Kmata)
    }else{
      Del <- mean(Kmataa) - NNold$Cost
    }
    
    if(Del == "NaN"){
      stop("This partition cannot be evaluated \n")  
      break
    }
    
    if((kl-1) == 0){
      if(verbose){
       cat("Value of the previous partition C(I_1^{K}):")
       print(mean(Kmata))
      }  
    }else{
      if(verbose){
       cat("Value of the previous partition C(I_1^{K}):")
       print(NNold$Cost)
      } 
    } 
    
    if(verbose){
     cat("Value of the new partition C(tilde{I}_1^{K}):")
     print(mean(Kmataa))
     cat("Current Delta value:")
     print(Del)
     cat("\n")
    }
    #Stopping criterion:
    if(Del > 0){
      if((Del / mean(Kmataa)) < 0.01){
        if(verbose){
         cat("STOPPING CRITERION FULFILLED")
         cat("\n")
        }
        change[iter_while] <- 0
        break         
      }
    }
    
    pru <- runif(1) 
    if(fir==0 & prT>0){ 
      prT <- -abs(Th)/log(prT) 
      T0 <- prT 
      PRT <- exp(Del/prT) 
      fir <- 1 
      if(verbose){
       cat("Probability of mis-allocation:")
       print(PRT)
       cat("\n")
      }    
    } 
    if(fir!=0 & prT>0){ 
      PRT <- exp(Del/prT) 
      if(verbose){
       cat("Probability of mis-allocation:")
       print(PRT)
       cat("\n")
      } 
    } 
    if(prT==0){ 
      PRT <- 0 
      if(verbose){
       cat("Probability of mis-allocation:")
       print(PRT)
       cat("\n")
      } 
    } 
    if(fir==0){ 
      PRT <- -abs(Del)/log(prT) 
      if(verbose){
       cat("Probability of mis-allocation:")
       print(PRT)
       cat("\n")
      }
      T0 <- prT 
      fir <- 1 
    } 
    
    if(verbose){
     cat("Probability of mis-allocation:")
     print(PRT)
     cat("\n")
    }
    if(Del>0 | (Del<=0 & pru==PRT)){ 
      NNuse <- NNusea
      Kmata <- Kmataa
      if(verbose){
       cat("The partition is accepted.\n")
       cat("New value of the partition:")
       print(mean(Kmata))
       cat("Partition:\n")
       print(table(NNuse[1,]))
      }
       NNold$Cost <- mean(Kmata)
      if(verbose){
       cat("\n")
      }
       improv[kl] <- kl - 1 
      change[iter_while] <- 1
    }else{
      NNold$Cost <- NNold$Cost  
      if(verbose){
       cat("The partition is rejected.\n")
       cat("Current value of the partition:")
       print(NNold$Cost)
       cat("\n")
      }
       change[iter_while] <- 0
    } 
    nostop <- nostop+1 
  }
  if(verbose){
   cat("While loop ends: \n")  
  }
  if(max(change) == 1){
    if(ct==0){ 
      stopnow <- 1 
    } 
    if(ct>0){ 
      stopnow <- 0 
    } 
    prs <- ct/length(NNuse[1,]) 
    NNnew <- NNuse 
    NN <- NNnew 
    Nuvec <- NNnew[1,] 
    
    DDia <- matrix(0,nrow=dim(X)[1],ncol=K)
    for(kz in (1:K)){ 
      Xmatr <- X[NNuse[1,]==kz,] 
      Nvecr <- Nvec[NNuse[1,]==kz] 
      DDia[,kz]<-apply(X,1,DDapply,Xmat=Xmatr) 
    } 
    
    DDa <- DDcalc2(DDia,NNuse,K,Km,norm) 
    pcc <- pamsil(X,NNuse[1,],K) 
    
    return(list(Iuse=Iuse,ct=ct,NN=NN,DDi=DDia,DD=DDa,pcc=pcc,Kmata=Kmata,Kmat0=Kmat0,Nuvec=Nuvec,Kmatb=Kmatb,
                prs=prs,stopnow=stopnow,Cost=NNold$Cost,trimmed=trimm,improv=improv,T0=T0))
  }else if(max(change) == 0 | is.null(change)){
    NNold_old <- NNold_old
    if((kl-1) == 0){
      NNold_old$Cost <- mean(Kmata)
    }else{
      NNold_old$Cost <- NNold$Cost
    } 
    NNold_old$ct <- ct
    NNold_old$trimmed <- trimm
    NNold_old$improv <- improv
    NNold_old$T0 <- T0
    return(NNold_old)
  }
}

Try the Anthropometry package in your browser

Any scripts or data that you put into this service are public.

Anthropometry documentation built on March 7, 2023, 6:58 p.m.