Question

Here’s my code: a function to be optimized with DEoptim algorithm; the function is quite simple, indeed.

Reproducible code:

library(DEoptim)
library(sm)

tau.0 <- c(58.54620, 61.60164, 64.65708, 71.19507, 82.39836, 101.28953, 119.68789)
rate <- c(0.04594674, 0.01679026, 0.02706263, 0.04182605, 0.03753949, 0.04740187, 0.05235710)
Du <- c(4.27157210, -0.07481174, -0.10551229, 0.51753843, 1.51075420, 6.51483315, 7.35631500)
Co <- c(0.2364985, -6.2947479, -7.5422644, -1.2745887, -42.6203118, 55.7663196, 70.9541141)

h <- h.select(x = tau.0, y = rate, method = 'cv')
sm <- sm.regression(x = tau.0, y = rate, h = h)
ya <- sm$estimate
xa <- sm$eval.points
y <- approx(x = xa, y = ya, xout = tau.0, rule = 2)$y

besty <- function(x) {

    dtau.0 <- x
    xout <- seq(1, max(tau.0), dtau.0)
    ratem <- approx(x = tau.0, y = rate / 1, xout = xout)$y
    ym <- approx(x = tau.0, y = y / 1, xout = xout)$y
    Dum <- approx(x = tau.0, y = Du, xout = xout)$y
    Com <- approx(x = tau.0, y = Co, xout = xout)$y
    dy <- NULL

    for(i in 1:length(ym)) {

        dy[i] <- ratem[i] - ym[i-1]

    }

    dy[is.na(dy)] <- na.omit(dy)[1]
    Dum[is.na(Dum)] <- na.omit(Dum)[1]
    Com[is.na(Com)] <- na.omit(Com)[1]
    dP <- Dum * dy - .5 * Com * dy ^ 2
    xout.m <- xout / 12
    dcurve <- cbind(dP * 100, xout.m)
    PVBP <- dcurve[which(dP == max(dP)),1]
    Maty <- dcurve[which(dP == max(dP)),2]

    return(- PVBP / x)

}

DEoptim(fn = besty, lower = 1, upper = 120)

To me the last command returns

ERROR: unsupported objective function return value

What’s wrong with my code for which good DEoptim does not succeed in optimizing?

If I replace the last function’s command line

return(- PVBP / x)

with

return(as.numeric(- PVBP / x))

it seems DEoptim works fine till few iterations, then…

> DEoptim(fn = besty, lower = 1, upper = 12)
Iteration: 1 bestvalit: -0.898391 bestmemit:    1.186242
Iteration: 2 bestvalit: -0.903304 bestmemit:    1.185117
Iteration: 3 bestvalit: -0.999273 bestmemit:    1.043355
Iteration: 4 bestvalit: -0.999273 bestmemit:    1.043355
Error in DEoptim(fn = besty, lower = 1, upper = 12) : 
  unsupported objective function return value

Maybe something in function syntax?

Thanks, guys 🙂

Answer 1

Solved due to Aaron, Joshua Ulrich and csgillespie tips.

Two modifications are required in order the code worked properly:

...
PVBP <- dcurve[which(dP == max(dP)),1]
Maty <- dcurve[which(dP == max(dP)),2]
...

must be replaced with

...
PVBP <- dcurve[which(dP == max(dP)),1][1]
Maty <- dcurve[which(dP == max(dP)),2][1]
...

while

...
return(- PVBP / x)
...

must be replaced with

...
return(as.numeric(- PVBP / x))
...

and in order to avoid NAs in objective function it is required that boundaries are set to

DEoptim(fn = besty, lower = 1, upper = max(tau.0) / 12)

Thanks guys who helped me!