R: Error in optim: 'list' object cannot be coerced to type 'double'

Question

I'm trying to get the parameters of a bi-exponential model by minimizing the Kullback-Leibler divergence using optim . The function I'm using have 3 parameters, but when I pass them to optim using par = par it throws the error "list object cannot be coerced to type 'double'", but I'm not even using lists.

Here is the code I'm using:

library(logKDE) # for kernel density of positive distributions
# Simulate rv of the bi-exponential
p <- 0.7
n <- 50
w <- 2
b <- 0.2
delt <- 0.01

biexp_data <- (p * rexp(n, 1/w) + (1 - p) * rexp(n, 1/b)) -  delt

# define kld to optimize
kld_optim <- function(x, par, from_a, to_b) {
  par <- unlist(par)
  w <- par[1]
  b <- par[2]
  p <- par[3]
  d <- 0.002
  
  integrand <- function(x, w, b, p, d, t) {
    
    denx <- logdensity(x, bw = 'logG', from = from_a, to = to_b)
    f.y <- approx(unlist(denx$x), unlist(denx$y), t)$y
    f.x <- p * dexp(t - d, rate = 1/w) + (1 - p) * dexp(t - d, rate = 1/b)
    tmpRatio <- f.x * (log2(f.x) - log2(f.y))
    # Return
    ifelse(is.infinite(tmpRatio), 0, ifelse(is.na(tmpRatio), 0, tmpRatio))

  }
  
  integrate(integrand, 
            from_a, to_b, 
            x = x,
            w = w, b = b, p = p, d = d)
  
}

optim(par = c(2, 0.1, 0.6),
      fn = kld_optim, 
      from_a = 0.01, 
      to_b = 20,
      x = biexp_data)

Why is this happening?

Thanks!

Answer 1

The function you pass to optim needs to return a scalar value, and your kld_optim function returns the result of a call to integrate() which according to the ?integrate help page, returns a list, not a numeric value. The "value" in contained in that list under the name "value". So change your integrate() call to

  integrate(integrand, 
            from_a, to_b, 
            x = x,
            w = w, b = b, p = p, d = d)$value