[R] How to speed up a for() loop

[Phil Spector]

Simona -
    I don't think preallocating your random variables would make
the code run any faster.
    A very simple change that would speed things up a little would
be to replace

  simPD.vec[i]=length(R.vec[R.vec<cutoff.])
with
  simPD.vec[i] = sum(R.vec < cutoff.)

Since you're using a different set of random numbers for each
element of systemic, I don't see an easy way to vectorize your
code (which doesn't mean that such a way doesn't exist).

 					- Phil Spector
 					 Statistical Computing Facility
 					 Department of Statistics
 					 UC Berkeley
 					 spector at stat.berkeley.edu

On Thu, 17 Feb 2011, simona.costanzo at ubs.com wrote:

>
> Dear all,
>
> Does anyone have any idea on how to speed up the for() loop below.
> Currently it takes approximately 2 minutes and 30 seconds.
>
> Because of the size of Nsim and N, simulating a multivariate normal
> (instead of simulating Nsim times a vector of N normal distributions)
> would require too much memory space.
>
> Many thanks for your kind help,
>
> Simona
>
>
> N=3000
> PD=runif(N,0,1)
> cutoff.=qnorm(PD)
>
> rho.=0.1
>
> Nsim=100000
> simPD.vec=0*(1:Nsim)
>
> systemic = rnorm(Nsim,0,1);
>
> for (i in 1:Nsim)
> {
> R.vec=sqrt(rho.)*systemic[i]+sqrt(1-rho.)*rnorm(N,0,1)
> simPD.vec[i]=length(R.vec[R.vec<cutoff.])
> }
>
>
>
>