R/cv.ELMCoxBAR.R

In ELMSurv: Extreme Learning Machine for Survival Analysis

##'Cross-Validation for ELMCoxBAR to tune kernel parameters Based on log Likelihood
##' @title Cross-Validation for ELMCoxBAR 
##' @param x  The covariates(predictor variables) of training data.
##' @param y  Survival time and censored status of training data. Must be a Surv  \code{survival} object
##' @param Kernel_type Type of kernel matrix. Currently four options avaibable. "RBF_kernel",a RBF kernel;"lin_kernel" , a linear kernel;poly_kernel ,a polynomial kernel;sigmoid_kernel, a sigmoid kernel. Default is "lin_kernel".
##' @param Kernel_para Parameters for different types of kernels. A single value for RBF and linear kernels. A vector for polynomial and sigmoid kernels and progam stops if only a single value is supplied. However, if the vector of values is supplied in the cases of RBF and liner kernels, only the first value will be used. Default is a vector value "c(2,1)".
##' @param penality Currently, penality is defaulted to 0 to train an ELMCoxBAR model. 
##' @param maxiter Maximum values of iterations to update the CoxBAR estimator. Default is 5.
##' @param nfolds  Number of folds in cross validation.
##' @param ... Additional arguments for  glmnet.
##' @return Object of class \code{ELMCoxBAR} with elements
##'   \tabular{ll}{
##'       \code{elmcox}    \tab  A glmnet type model. See \code{glmnet} for details. \cr
##'       \code{trainx} \tab  Training data covariates. \cr
##'          \code{kerneltype} \tab  Type of kernel matrix used in training. kerneltype=1,a RBF kernel;kerneltype=2 , a linear kernel;kerneltype=3 ,a polynomial kernel;kerneltype=4, a sigmoid kernel. \cr
 ##'   \code{Kernel_para} \tab  Optimal Parameters returned by cross validation. A single value for kerneltype=1 or 2. A vector for kerneltype=3 or 4. \cr
##'   }
##' @author Hong Wang
##' @references
##' \itemize{
##'   \item Wang, H, Li, G. Extreme learning machine Cox model for high-dimensional survival analysis. Statistics in Medicine. 2019; 38:2139-2156.
##'  }
##' @examples
##' set.seed(123)
##' require(ELMSurv)
##' require(survival)
##' #Lung DATA
##' data(lung)
##' lung=na.omit(lung)
##' lung[,3]=lung[,3]-1
##' n=dim(lung)[1]
##' L=sample(1:n,ceiling(n*0.5))
##' trset<-lung[L,]
##' teset<-lung[-L,]
##' rii=c(2,3)
##' # Default with lin_kernel 
##' elmsurvmodel=cv.ELMCoxBAR(x=trset[,-rii],y=Surv(trset[,rii[1]], trset[,rii[2]]))
##' #The predicted linear predictor
##' bestpara=elmsurvmodel$Kernel_para
##' @export
cv.ELMCoxBAR <- function(x,y,Kernel_type="lin_kernel", Kernel_para=c(2,1),penality=0, maxiter=5,nfolds=2,...) {

if (!inherits(y, "Surv"))
    stop("Response must be a 'survival' object - use the 'Surv()' function")

  kplen=length(Kernel_para)

  if(Kernel_type=="RBF_kernel"){
    kerneltype=1
    if(kplen==0||kplen<1){
      stop("Error: Kernel Parameter for RBF_kernel Error!")
    }
  }else if(Kernel_type=="lin_kernel"){
    kerneltype=2
    if(kplen==0||kplen<1){
      stop("Error: Kernel Parameter for lin_kernel Error!")
    }
  }  else if(Kernel_type=="poly_kernel"){
    kerneltype=3
    if(kplen==0||kplen<2){
      stop("Error: Kernel Parameter for poly_kernel Error!")
    }
  }  else if(Kernel_type=="sigmoid_kernel"){
    kerneltype=4
    if(kplen==0||kplen<2){
      stop("Error: Kernel Parameter for sigmoid_kernel Error!")
    }
  }else{
    stop("Error:Unknow kernel types!")
  }
  maxc=Kernel_para[1];minc=maxc*0.01
  cseq <- seq(maxc,minc,length.out=10)

  nlambd=length(cseq)
  #
 


  folds <- cut(seq(1,nrow(x)),breaks=nfolds,labels=FALSE)
  perffolds=rep(0,nlambd)
   acc_temp=rep(0,nfolds)
  #Perform 2 fold cross validation
  
  for(k in 1:nfolds){
  	tryCatch({
    for (i in 1:nlambd){
    #Segement your data by fold using the which() function
    testIndexes <- which(folds==k,arr.ind=TRUE)
    tesetx <- x[testIndexes, ]
    trsetx <- x[-testIndexes, ]
    tesety <- y[testIndexes, ]
    trsety <- y[-testIndexes, ]
    #Use the test and train data partitions however you desire...
   
	mycvbar=ELMCoxBAR(x=trsetx,y=trsety,Kernel_type, Kernel_para=c(cseq[i],minc),penality=penality, maxiter=maxiter)  
   #
   perffolds[i]=mycvbar$elmcox$logLik   
    }
	 #best lambda
     acc_temp[k]=cseq[which.min(perffolds)]
	 }, error=function(e){cat("ERROR :",conditionMessage(e), "\n")})
	
	}	
	#message(perffolds)
	
 bestc=mean(acc_temp)

 
  fit <- list()
  tryCatch({
  fit = ELMCoxBAR(x,y, Kernel_type,Kernel_para=c(bestc,minc),penality, maxiter=5)
  }, error=function(e){cat("ERROR :",conditionMessage(e), "\n")})	
  fit$trainx = x
  fit$kerneltype = kerneltype
  fit$Kernel_para = c(bestc,minc)
  class(fit)="ELMCoxBAR"
  fit
  #best beta



}