run.sims: Run a simulation study

In dereksonderegger/Simulations: Keeps track of many simulation replications

Description

Conduct a simulation study where we have a matrix of parameter values that we are interested. For each row in the parameter matrix, the simulation function is called num.reps times and the output of each run is saved to the simulation directory.

Usage

run.sims(sim.function, param.matrix, num.reps,
  sim.directory = "sim_directory", prep.function = NULL, verbose = FALSE)

Arguments

sim.function	A function to run a single simulation.
num.reps	The number of replicate simulations for each parameter combination.
sim.directory	The local directory to store the simulation results.
prep.function	A function to be run before each new parameter combination is started.
verbose	If information about what parameter combination and replicate number is about to be run.
params	A data frame of parameter values. The order of the columns must match the order of the arguments to the sim.function.

Details

The simulation can be re-run with higher num.reps and the original simulations (still stored in the simulation directory) will be used and only the additional replications will be run. Likewise, if the simulation is interupted (say a power outage...), the simulation can be re-started and the previous iterations will not be lost.

Examples

# Simulation is creating data and fitting
# a regression model to estimate a slope.
Sim.Function <- function(N, alpha, beta, sigma){
  x <- runif(N, 0, 10)
  y <- alpha + beta*x + rnorm(N, 0, sigma)
  model <- lm(y~x)
  return(model)
}

# Create a matrix where each row represents a set of parameters
Params <- expand.grid(N=20, alpha=0, beta=c(0,1), sigma=c(.2, 1))

# Run the simulations.
run.sims(Sim.Function, Params, num.reps=10)

# Now that the simulations are run, we want to analyze
# them.  Create a function to be applied to each simulation
# that results in a single row.  Those rows will be concatenated
# together into a large data frame that summarizes the simulation.

# Because the output of the Sim.Function was a linear model output,
# that is the input for this function
Summary.Function <- function(model, params){
  beta.hat <- coef(model)[2]
  beta.lwr <- confint(model)[2,1]
  beta.upr <- confint(model)[2,2]
  return( data.frame(beta=beta.hat, lwr=beta.lwr, upr=beta.upr))
}

Sims <- summarize.sims(Summary.Function, Params)

dereksonderegger/Simulations documentation built on May 15, 2019, 3:58 a.m.