R/cvgrid.ald.R

In dirttee: Distributional Regression for Time to Event Data

cvgrid.ald <- function(yy,delta,B,quantile,DD,nb,constmat,types)
{
  las1 = seq(-1, 4, by = .75)
  glatterms = which(types != "parametric")
  #print(glatterms)
  #lambdas = matrix(las1,nrow=length(las1)*length(glatterms),ncol=1)
  #if(length(glatterms) > 1)
  #for(i in 2:length(glatterms))
  #  lambdas = cbind(lambdas,rep(las1,each=i,times=length(glatterms)-i+1))
  
  lambdas_list <- list()
  for(i in 1:length(glatterms)) {
    lambdas_list[[i]] <- las1
  }
  lambdas <- expand.grid(lambdas_list)
  
  score = rep(0, nrow(lambdas))
  
  lambdas = 10^lambdas
  
  penalty = rep(0,length(types))
  
  for(i in 1:nrow(lambdas))
  {
    penalty[glatterms] = unlist(lambdas[i,])
    aa <- asyregpen.aft(yy, delta, B, quantile, abs(penalty), DD, nb, constmat)
    
    score[i] =  mean(-(delta * log(dald(yy,B%*%aa$a,abs(aa$sigma)+0.0001,quantile)) + (1-delta) * log(1-pald(yy,B%*%aa$a,abs(aa$sigma)+0.0001,quantile)))/(1-aa$diag.hat.ma)^2,na.rm=TRUE)
    
  }
  
  penalty[glatterms] = lambdas[which.min(score),]
  
  penalty
  
}