R: Run a function iteratively using a grid search approach for...

grid_search {paramtest}

R Documentation

Run a function iteratively using a grid search approach for parameter values, with options for parallel processing.

Description

grid_search runs a user-defined function iteratively. Parameter values can be given to grid_search, which will fully cross all parameters so that each parameter value is tested at all other values of all parameters.

Usage

grid_search(func, params = NULL, n.iter = 1, output = c("list",
  "data.frame"), boot = FALSE, bootParams = NULL, parallel = c("no",
  "multicore", "snow"), ncpus = 1, cl = NULL, beep = NULL, ...)

Arguments

`func`	A user-defined function. The first argument to this function will be the iteration number.
`params`	A list of parameters to be passed to `func`. The parameters are fully crossed so that each parameter value is tested at all other values of all parameters. (For example, list(N=c(5, 10), x=c(1, 2)) will test four sets of parameters: N=5 and x=1, N=5 and x=2, N=10 and x=1, and N=10 and x=2.) Each set of parameters will then be passed to `func` in turn.
`n.iter`	Number of iterations (per set of params).
`output`	Specifies how `grid_search` provides the ultimate output from `func`: can return a "list" or a "data.frame". Note that if "data.frame" is specified, the supplied function must return a vector, matrix, or data frame, so it can be coerced into the data frame format. The "list" option will accept any type of output.
`boot`	Whether or not to use bootstrapped data to pass along to `func`. Using this option instead of bootstrapping within `func` is preferable to take advantage of parallelization.
`bootParams`	If `boot=TRUE`, then use `bootParams` to pass along a named list of arguments to the `boot` function. The statistic and R parameters will be filled automatically, but at minimum you will need to pass along data. Information about parallel processing will also be passed along automatically.
`parallel`	The type of parallel operation to be used (if any).
`ncpus`	Integer: the number of processes to be used in parallel operation.
`cl`	An optional `parallel` or `snow` cluster for use if `parallel = 'snow'`. If not supplied, a cluster on the local machine is created for the duration of the iterations.
`beep`	Include a numeric value or character vector indicating the sound you wish to play once the tests are done running. Requires the 'beepr' package, and information about supported values is available in the documentation for that package.
`...`	Additional arguments to be passed to `func`. If you do not need to vary certain parameters in your model, you can pass them to `func` here.

Value

Returns a list (by default) with one element per iteration. If output is specified as "data.frame", then func must return a (named) vector with the results you wish to capture.

Examples

lm_test <- function(iter, N, b0, b1) {
    x <- rnorm(N, 0, 1)
    y <- rnorm(N, b0 + b1*x, sqrt(1 - b1^2))
    data <- data.frame(y, x)
    model <- lm(y ~ x, data)

    # capture output from model summary
    est <- coef(summary(model))['x', 'Estimate']
    se <- coef(summary(model))['x', 'Std. Error']
    p <- coef(summary(model))['x', 'Pr(>|t|)']

    return(c(xm=mean(x), xsd=sd(x), ym=mean(y), ysd=sd(y), est=est, se=se, p=p,
        sig=est > 0 & p <= .05))
}

# test power for sample size N=200 and N=300, with 500 iterations for each
power_sim <- grid_search(lm_test, params=list(N=c(200, 300)), n.iter=500, b0=0, b1=.15)