memisc (version 0.96-10)

Simulate: Support for Simulation Experiments

Description

Simulate is a function to simplify simulation studies. It can be used to conduct Monte Carlo studies of statistical estimators, discrete event, and agent based simulations.

Usage

Simulate(step,
    conditions = NULL,
    start = NULL,
    cleanup = NULL,
    ...,
    nsim = 1,
    seed = NULL,
    trace=0,
    keep.data=TRUE,
    keep.states=FALSE,
    keep.seed = !is.null(seed),
    restore.seed = !is.null(seed),
    bucket = default_bucket
    )
# signal an interrupt condition
interrupt(msg=NULL)

Arguments

step

an expression that produces simulation results for each replication; can be a function call, or a braced expression that ``returns'' a value like a function body.

conditions

an optional data frame or object coerceable into a data frame. Each row of this data frame defines an experimental condition.

start

either NULL or an expression that computes starting values for step.

cleanup

either NULL or an expression does some cleaning up after the exectution of all steps.

...

other substitutions for step, held fixed in the simulation experiment

nsim

an integer value; the number of replication in each experimental setting. If nsim is infinite or NA, step is replicated (in each setting) until either a user interrupt is signalled (CTRL-C is pressed) or

seed

either NULL or an integer value suitable for set.seed. Note that the random state before the call to Simulate is restored.

trace

an integer value determining the amount of information output during the simulation process. If trace equals zero nothing is reported during the simulation run. Otherwise, the replication number is output for each multiple of

keep.data

logical value; if TRUE, return values of the expression in step are collected into a data fame.

keep.states

logical value; if TRUE, a list of all variables defined in step (after execution of cleanup if present) is returned.

keep.seed

logical value; if TRUE, the state of the random number generator is saved in an attribute "seed" of the return value of Simulate.

restore.seed

logical value; if TRUE, the state of the random number generator is restored after conducting the simulations.

bucket

a function that returns a bucket object, in which simulation results are collected.

msg

a character string, the message shown if an interrupt condition is signalled.

Value

• A data frame that contains experimental conditions and simulation results.

Details

Simulate calls or evaluates its first argument, step, or, if a conditions argument is given, nsim times for each row of the conditions data frame.

Before repeatingly evaluating step, the expression start, if present, is evaluated, which may be used to create starting values for a simulatation of to setup up the scenery for an agent-based simulation. After repeatingly evaluating step, the expression cleanup, if present, is evaluated.

If restore.seed is given, the state of the random generator is saved before conducting the simulation and restored afterwards. Therefore step, start, or cleanup may call set.seed without affecting the generation of random numbers after a call to Simulate.

interrupt raises an interrupt condition, which acts like a user interrupt.

Note that if an interrupt condition is signalled during a (replicated) evaluation of step the results of previous replications are still saved and Simulate jumps to the next condition of the simulation experiment (if there is any). That is, if a simulation is interrupted by the user because it takes too long, the results so far produced by the simulation are not lost.

On the other hand, interrupt can be used to determine at run-time how often step is evaluated.

Examples

Normal.example <- function(mean=0,sd=1,n=10){
  x <- rnorm(n=n,mean=mean,sd=sd)
  c(
    Mean=mean(x),
    Median=median(x),
    Var=var(x)
  )
}

Normal.simres <- Simulate(
    Normal.example(mean,sd,n),
    expand.grid(
          mean=0,
          sd=c(1,10),
          n=c(10,100)
          ),
    nsim=200,
    trace=50)

genTable(sd(Median)~sd+n,data=Normal.simres)

expr.simres <- Simulate(
      median(rnorm(n,mean,sd)),
      expand.grid(
          n=c(10,100),
          mean=c(0,1),
          sd=c(1,10)
      ),
    nsim=200,
    trace=50
    )

genTable(c(mean(result),sd(result))~sd+n+mean,data=expr.simres)

## This takes a little bit longer
lm.example <- function(a=0,b=1,n=101,xrange=c(-1,1),serr=1){
  x <- seq(from=xrange[1],to=xrange[2],length=n)
  y <- a + b*x + rnorm(n,sd=serr)
  lm.res <- lm(y~x)
  coef <- lm.res$coef
  names(coef) <- c("a","b")
  coef
}

lm.simres <- Simulate(
      lm.example(n=n,serr=serr),
      expand.grid(
      serr=c(0.1,1,10),
      n=c(11,101,501)
      ),
      nsim=200,
      trace=50
    )
genTable(c(sd(a),sd(b))~serr+n,data=lm.simres)