Clever Geek Handbook

R with functions as arguments, each with variable arguments

In response to the Cross Validated question, I wrote a simple function that used arbitrary quantile functions as its arguments

etacor=function(rho=0,nsim=1e4,fx=qnorm,fy=qnorm){
  #generate a bivariate correlated normal sample
  x1=rnorm(nsim);x2=rnorm(nsim)
  if (length(rho)==1){
    y=pnorm(cbind(x1,rho*x1+sqrt((1-rho^2))*x2))
    return(cor(fx(y[,1]),fy(y[,2])))
    }
  coeur=rho
  rho2=sqrt(1-rho^2)
  for (t in 1:length(rho)){
     y=pnorm(cbind(x1,rho[t]*x1+rho2[t]*x2))
     coeur[t]=cor(fx(y[,1]),fy(y[,2]))}
  return(coeur)
  }

However, fx

and fy

may require your own parameters. For example, when fx=qchisq

or when fy=qgamma

. As the default solution in my implementation, I used

fx=function(x) qchisq(x,df=3)

and

fy=function(x) qgamma(x,scale=.2)

but it takes a long time.

For example,

> rhos=seq(-1,1,.01)
> system.time(trancor<-etacor(rho=rhos,fx=qlnorm,fy=qexp))
utilisateur     système      écoulé 
      0.834       0.001       0.834

against

> system.time(trancor<-etacor(rho=rhos,fx=qlnorm,fy=function(x) qchisq(x,df=3)))
utilisateur     système      écoulé 
      8.673       0.006       8.675
+3


source to share


2 answers


Illustration of my comment above:

etacor1 <- function(rho = 0,
                    nsim = 1e4,
                    fx = qnorm,
                    fy = qnorm,
                    fx.args = formals(fx),
                    fy.args = formals(fy)){
    #generate a bivariate correlated normal sample
    x1 <- rnorm(nsim)
    x2 <- rnorm(nsim)

    fx.arg1 <- names(formals(fx))[1]
    fy.arg1 <- names(formals(fy))[1]

    if (length(rho) == 1){
        y <- pnorm(cbind(x1, rho * x1 + sqrt((1 - rho^2)) * x2))
        fx.args[[fx.arg1]] <- y[,1]
        fy.args[[fy.arg1]] <- y[,2]
        return(cor(do.call(fx,as.list(fx.args)),
                   do.call(fy,as.list(fy.args))))
    }

    coeur <- rho
    rho2 <- sqrt(1 - rho^2)

    for (t in 1:length(rho)){
        y <- pnorm(cbind(x1,rho[t]*x1+rho2[t]*x2))
        fx.args[[fx.arg1]] <- y[,1]
        fy.args[[fy.arg1]] <- y[,2]
        coeur[t] <- cor(do.call(fx,as.list(fx.args)),
                        do.call(fy,as.list(fy.args)))
    }

    return(coeur)
}


I am unhappy with the obvious need as.list

. I feel like I should know why this is so, but now it is eluding me.

When using this function, it is not necessary to pass all the arguments, but you do need to make sure that the list you pass to fx.args

or fy.args

is named.

+3


source


Thanks for the comments and response! I'm afraid the main problem is that, as pointed out by joran and Mr Flick , some quantile functions are much slower than others:

> system.time(etacor(rhos,fx=function(x) qexp(x)))
utilisateur     système      écoulé 
      1.182       0.000       1.182 
> system.time(etacor(rhos,fx=qexp))
utilisateur     système      écoulé 
      1.238       0.000       1.239

against



> system.time(etacor(rhos,fx=function(x) qchisq(x,df=3)))
utilisateur     système      écoulé 
      4.955       0.000       4.951 
> system.time(etacor(rhos,fx=function(x) qgamma(x,sha=.3)))
utilisateur     système      écoulé 
      4.316       0.000       4.314

So in the end, using a function definition when it requires parameters seems like a simple and easy solution. Thanks for all your inputs.

+1


source






More articles:


All Articles