CODEWITHSUNDEEP
roptimization
J. Doe.
J. Doe.

Reputation: 1305

R optim vector to maximise user defined function

I have the following dataframes and I would like to find the global maximum of the optimised_vector

set.seed(1)
xxx<-data.frame(C1 = c("obs1","obs2","obs3","obs4","obs5","obs6"),
                X1 = round(runif(6, 1, 6),digits=1),
                X2 = round(runif(6, 1, 6),digits=1),
                X3 = round(runif(6, 1, 6),digits=1),
                X4 = round(runif(6, 1, 6),digits=1),
                X5 = round(runif(6, 1, 6),digits=1)
)

yyy<-data.frame(C1 = c("obs1","obs2","obs3","obs4","obs5","obs6"),
                YY = c(1,2,3,4,5,6)
)

#optimised_vector<-c(1,1,1,1,1,1)
optimised_vector<-c("coef1","coef2","coef3","coef4","coef5","coef6")

So I've built these functions to try to find the maximum correlation between my dummy data. But I would like to be able to impose some constraints, like non-negativity and max of 0.75 for each of the elements of the optimised_vector

inner_function <- function(xxx,optimised_vector) {

   scoring <- rowSums(xxx[,2:6] * optimised_vector)
   return (scoring)
}

maximiser <- function(optimised_vector) {   
   1-cor(
      yyy$YY, 
      inner_function(xxx,optimised_vector),
      method="kendall", use="pairwise")
   }

And then I run optim to get the results...

optim(par = c(1,1,1,1,1,1),
      fn  = maximiser)

Is this correct, does it even find the global convergence points and how can I impose constraints?

Upvotes: 1

Views: 732

Answers (1)

JdP
JdP

Reputation: 764

Using bounderies is possible with the method L-BFGS-B in 

optim(par     = c(1,1,1,1,1,1),
      fn      = maximiser,     
      method  = "L-BFGS-B", 
      lower   = rep(0, 4), 
      upper   = rep(0.75, 4))

Upvotes: 1

Related Questions