best {aylmer} | R Documentation |
Uses simulated annealing to find the `best' permissible board, using any objective function
best(x, func = NULL, n = 100, ...)
x |
A board |
func |
The objective function, with default NULL meaning
to use -prob(x) |
n |
Maximum number of attempts (passed to candidate() ) |
... |
Further arguments passed to optim() |
The help page for optim()
gives an example of simulated
annealing being used to solve the travelling salesman problem and
best()
uses the same technique in which the gr
argument
specifies a function used to generate a new candidate point
(candidate()
).
Function randomprobs()
also takes a func
argument and (if
argument give.best
is TRUE
) will return the optimal
board. But these two functions are very different: best()
uses
optim()
which incorporates highly specific optimization
algorithms to find a global maximum, while randomprobs()
creates a Markov chain and reports the board with the most desirable
objective function.
Robin K. S. Hankin
a <- matrix(0,5,5) diag(a) <- NA a[cbind(1:5 , c(2:5,1))] <- 4 ## Not run: best(a,control=list(maxit=10)) ## Answer should be all ones except the diagonal ## End(Not run) # Now a non-default function; SANN should be able to get func(x) down to # zero pretty quickly: ## Not run: best(a,func=function(x){x[1,2]},control=list(maxit=100)) ## End(Not run) # The 'dontrun' is needed because sometimes the method needs a bigger n