SVM PSO optimization in R

Question

I am working on a PSO optimisation in R to optimize SVM parameters gamma, epsilon and C. I can optimise one variable but got this error when I use more than one variable.

Error in tune(svm, Train_Input_NonNorm, Train_Target_NonNorm, kernel = "radial", : argument "C" is missing, with no default

SVM_Nonnorm_Optim_f <- function(E,C,G)
{
set.seed(123)
SVM_NonNorm_Tune <-tune(svm,Train_Input_NonNorm,Train_Target_NonNorm, kernel="radial",ranges= 
list(epsilon = E, cost = C, gamma=G))

SVM_NonNorm_tuned <- SVM_NonNorm_Tune$best.model

Predict_SVM_NonNorm_tuned <- predict(SVM_NonNorm_tuned,Train_Input_NonNorm)

RMSE_SVM_NonNorm_tuned <- sqrt(mean((Train_Target_NonNorm - Predict_SVM_NonNorm_tuned)^2))

return(RMSE_SVM_NonNorm_tuned)

}

n <- 50 
m.l <- 50 
w <- 0.95 
c1 <- 2 
c2 <- 2 
xmin <- 0 
xmax <- 1000 
vmax <- c(4, 4) 

SVM_NonNorm_PSO <- psoptim(SVM_Nonnorm_Optim_f, n=n, max.loop=m.l, w=w, c1=c1, c2=c2,
               xmin=xmin, xmax=xmax, vmax=vmax, seed=5, anim=FALSE)`

Answer 1

I don't use psoptim, but very likely you need to pass a vector of parameters, i.e. something like

SVM_Nonnorm_Optim_f <- function(x) {
    E <- x[1]
    C <- x[2]
    G <- x[3]
    ## do computations
}

With only three parameters, there may be better methods, perhaps even a lowly grid search would do.