Is there a way so that an already defined argument can appear as missing when using optim() function in R?

1
votes

I am trying to get the Maximum Likelihood Estimators of the log-likelihood of a Gumbel distribution for survival analysis(i say that so that you dont get astranged by the log-likelihood function, i think its correct). In order to do that i have to maximize the minus log-likelihood by using the optim function, i tried to do so but the console is giving me an Error in fn(par, ...) : argument "b" is missing, with no default.

I tried also to do it in a similar way as in the answer to this link: Solve for maximum likelihood with two parameters under constraints ,but the console game me the following: Error in optim(c(1, 1), function(x) log_lhood(x[1], x[2], d = lung$status, : objective function in optim evaluates to length 0 not 1. 

log_lhood <- function(m,b,d,t){                           
  sum<-0
  for (i in 1:length(lung)){ 
    if (d[i] == 1){
      sum<- sum - log(1-exp(-exp(-(t[i]-m)/b)))
    } else {
      sum<- sum - log((1/b)*exp(-(t[i]-m/b+exp(-(t[i]-m/b)))))
    }
  }
}
#a,b parameter optimization

optim(c(0,0), fn = log_lhood, d = lung$status, t = KM_fit$time) #1st way

optim(c(1, 1), function(x) log_lhood(x[1], x[2],d=lung$status,t=KM_fit_test$time)) #2nd way as in the link
1
rmaxanalysissurvivallog-likelihood
This is almost testable, please provide the minimum amount for someone to run and experience your error. Is lung from a package? – Evan Friedland
it is from the survival package. you are right im sorry. – nnisgia

1 Answers

2
votes

There are a few problems here. The first argument of the function is supposed to be a vector of parameters. Also you need nrow(lung) not length(lung), and it would be better to use length(d). Also you should not use a loop here, it's very inefficient, use ifelse() (in R we always try to vectorise everything). Also you need to check that the log likelihood can be calculated for all values of the parameters (e.g. b=0). Also you forgot to return(sum). Also sum is a useful function you shouldn't mask.

This runs.

library(reprex)

lung <- data.frame(status=c(0,0,1,1))
KM_fit <- data.frame(time=c(0,1,2,3))

log_lhood <- function(x,d,t){
    m <- x[1]
    b <- x[2]
    sum <-0
    for (i in 1:nrow(lung)){
        if (d[i] == 1){
            sum <- sum - log(1-exp(-exp(-(t[i]-m)/b)))
        } else {
            sum <- sum - log((1/b)*exp(-(t[i]-m/b+exp(-(t[i]-m/b)))))
        }
    }
    return(sum)
}
#a,b parameter optimization

optim(par=c(0,1), fn = log_lhood, d = lung$status, t = KM_fit$time)
$par
[1] 1.661373 1.811780

$value
[1] 5.318068

$counts
function gradient 
      63       NA 

$convergence
[1] 0

$message
NULL

You could rewrite your function like this.

log_lhood <- function(x,d,t){
    m <- x[1]
    b <- x[2]
    s <- ifelse(d==1,
                  -log(1-exp(-exp(-(t-m)/b))),
                  -log((1/b)*exp(-(t-m/b+exp(-(t-m/b)))))
                  )
    return(sum(s, na.rm=TRUE))
}