fun.load: Stochastic Search Helper Functions

Description

Functions that assign values and functions needed by MSS.snow

Usage

fun.load.hals.a()
fun.load.hals.fill()
fun.load.widals.a()
fun.load.widals.fill()

Arguments

Value

Nothing. The central role of these functions is the creation of four functions required by MSS.snow: FUN.MH, FUN.GP, FUN.I, and FUN.EXIT. These four functions are assigned to the Global Environment. This fun.load suite of functions also passes needed objects (out-of-scope) to snowfall threads if the global user-made variable run.parallel is set to TRUE.

Details

Please see MSS.snow and examples.

Examples

Run this code




### Here's an itty bitty example:
### we use stochastic search to find the minimum number in a vector
### GP isn't used here, and hence neither are p.ndx.ls nor f.d
### however, we still need to create them since MSS.snow requires their existence

if (FALSE) {

fun.load.simpleExample <- function() {

   if( run.parallel ) {
         sfExport("xx")
    }
    
    p.ndx.ls <- list( c(1) )
    
    p.ndx.ls <<- p.ndx.ls
    
    f.d <- list( dlog.norm )

    f.d <<- f.d
    
    FUN.MH <- function(jj, GP.mx, X) {
        our.cost <- sample(xx, 1)
    }

    FUN.MH <<- FUN.MH
    
    
    FUN.GP <- NULL
    FUN.GP <<- FUN.GP
    
    
    FUN.I <- function(envmh, X) {
        cat( "Hello, I have found an even smaller number in xx ---> ", envmh$current.best, "\n" )
    }
    FUN.I <<- FUN.I
    
    FUN.EXIT <- function(envmh, X) {
        cat( "Done",   "\n" )
    }

    FUN.EXIT <<- FUN.EXIT
    
}

xx <- 1:600

GP <- c(1)

run.parallel <- TRUE
sfInit(TRUE, 2)

MH.source <- fun.load.simpleExample
MH.source()

MSS.snow(MH.source, Inf, p.ndx.ls, f.d, matrix(1, nrow=28), 28, 7)
sfStop()




### Here's another itty bitty example:
### we use stochastic search to find the mean of a vector
### i.e., the argmin? of sum ( x - ? )^2

fun.load.simpleExample2 <- function() {

   if( run.parallel ) {
         sfExport("xx")
    }
    
    p.ndx.ls <- list( c(1) )
    p.ndx.ls <<- p.ndx.ls
    
    f.d <- list( unif.mh )
    f.d <<- f.d
    
    FUN.MH <- function(jj, GP.mx, X) {
        our.cost <- sum( ( xx - GP.mx[jj, 1] )^2 )
        return(our.cost)
    }
    FUN.MH <<- FUN.MH
    
    FUN.GP <- NULL
    FUN.GP <<-  FUN.GP
    
    FUN.I <- function(envmh, X) {
        cat( "Improvement ---> ", envmh$current.best, " ---- " , envmh$GP, "\n" )
    }
    FUN.I <<- FUN.I
    
    FUN.EXIT <- function(envmh, X) {
        our.cost <- envmh$current.best
        GP <- envmh$GP
        cat( "Done",   "\n" )
        cat( envmh$GP, our.cost, "\n" )
    }
    FUN.EXIT <<- FUN.EXIT
    
}

##set.seed(99999)
xx <- rnorm(300, 5, 10)

GP <- c(1)

run.parallel <- TRUE
sfInit(TRUE, 2)

MH.source <- fun.load.simpleExample2
MH.source()

MSS.snow(MH.source, Inf, p.ndx.ls, f.d, matrix(1/10, nrow=140, ncol=length(GP)), 140, 14)
sfStop()

##### in fact:
mean(xx)

}