Dear Friends,
I have been trying to learn how to use the derivative free optimization
algorithms implemented in the package RGENOUD by Mebane and Sekhon. However, it
does not seem to work for reasons best described as my total ignorance. If
anybody has experience using this package, it would be really helpful if you can
point out where I'm making a mistake.
Thanks in advance
Anup
Sample code attached
library(rgenoud)
nobs <- 5000
t.beta <- c(0,1,-1)
X <- as.matrix(cbind(rep(1, nobs), runif(nobs), runif(nobs))) # Creating the
design matrix
prodterm <- (X%*%t.beta)+rnorm(nrow(X))
Y <- as.matrix(ifelse(prodterm<0, 0, 1))
# Defining the likelihood function
log.like <- function(beta, Y, X)
{
term1 <- pnorm(X%*%beta)
term2 <- 1-term1
loglik <- (sum(Y*log(term1))+sum((1-Y)*log(term2))) # Likelihood function to
be maximized
}
stval <- c(0,0,0)
opt.output <- optim(stval,log.like,Y=Y[,1], X=X[,1:3],
hessian=T, method="BFGS", control=c(fnscale=-1,trace=1))
opt.output
### Now using GENOUD gives me errors
genoud.output <- genoud(log.like,beta=stval,X=X[,1:3], Y=Y[,1], nvars=3,
pop.size=3000, max=TRUE)
---------------------------------
[[alternative HTML version deleted]]
