Generate multiple datasets from a single DAG object

This function takes a single DAG object and generates a list of multiple datasets, possible using parallel processing

Usage

sim_n_datasets(dag, n_sim, n_repeats, n_cores=parallel::detectCores(),
               data_format="raw", data_format_args=list(),
               seed=stats::runif(1), progressbar=TRUE, ...)

Arguments

dag: A DAG object created using the empty_dag function with nodes added to it using the + syntax. See ?empty_dag or ?node for more details. If the dag contains time-varying nodes added using the node_td function, the sim_discrete_time function will be used to generate the data. Otherwise, the sim_from_dag function will be used.
n_sim: A single number specifying how many observations per dataset should be generated.
n_repeats: A single number specifying how many datasets should be generated.
n_cores: A single number specifying the amount of cores that should be used. If n_cores = 1, a simple for loop is used to generate the datasets with no parallel processing. If n_cores > 1 is used, the doSNOW package is used in conjunction with the doRNG package to generate the datasets in parallel. By using the doRNG package, the results are completely reproducible by setting a seed.
data_format: An optional character string specifying the output format of the generated datasets. If "raw" (default), the dataset will be returned as generated by the respective data generation function. If the dag contains time-varying nodes added using the node_td function and this argument is set to either "start_stop", "long" or "wide", the sim2data function will be called to transform the dataset into the defined format. If any other string is supplied, regardless of whether time-varying nodes are included in the dag or not, the function with the name given in the string is called to transform the data. This can be any function. The only requirement is that it has a named argument called data. Arguments to the function can be set using the data_format_args argument (see below).
data_format_args: An optional list of named arguments passed to the function specified by data_format. Set to list() to use no arguments. Ignored if data_format="raw".
seed: A seed for the random number generator. By supplying a value to this argument, the results will be replicable, even if parallel processing is used to generate the datasets (using n_cores > 1), thanks to the magic performed by the doRNG package.
progressbar: Either TRUE (default) or FALSE, specifying whether a progressbar should be used. Currently only works if n_cores > 1, ignored otherwise.
...: Further arguments passed to the sim_from_dag function (if the dag does not contain time-varying nodes) or the sim_discrete_time function (if the dag contains time-varying nodes).

Details

Generating a number of datasets from a single defined dag object is usually the first step when conducting monte-carlo simulation studies. This is simply a convenience function which automates this process using parallel processing (if specified).

Note that for more complex monte-carlo simulations this function may not be ideal, because it does not allow the user to vary aspects of the data-generation mechanism inside the main for loop, because it can only handle a single dag. For example, if the user wants to simulate n_repeats datasets with confounding and n_repeats datasets without confounding, he/she has to call this function twice. This is not optimal, because setting up the clusters for parallel processing takes some processing time. If many different dags should be used, it would make more sense to write a single function that generates the dag itself for each of the desired settings. This can sadly not be automated by us though.

Author

Robin Denz

Value

Returns a list of length n_repeats containing datasets generated according to the supplied dag object.

Examples

library(simDAG)

# some example DAG
dag <- empty_dag() +
  node("death", type="binomial", parents=c("age", "sex"), betas=c(1, 2),
       intercept=-10) +
  node("age", type="rnorm", mean=10, sd=2) +
  node("sex", parents="", type="rbernoulli", p=0.5) +
  node("smoking", parents=c("sex", "age"), type="binomial",
       betas=c(0.6, 0.2), intercept=-2)

# generate 10 datasets without parallel processing
out <- sim_n_datasets(dag, n_repeats=10, n_cores=1, n_sim=100)

if (requireNamespace("doSNOW") & requireNamespace("doRNG") &
    requireNamespace("foreach")) {

# generate 10 datasets with parallel processing
out <- sim_n_datasets(dag, n_repeats=10, n_cores=2, n_sim=100)
}
#> Loading required namespace: doSNOW
#> Loading required namespace: doRNG
#> 
  |                                                                            
  |                                                                      |   0%
#> Error in e$fun(obj, substitute(ex), parent.frame(), e$data): worker initialization failed: package ‘RcppZiggurat’ could not be loaded

# generate 10 datasets and transforming the output
# (using the sim2data function internally)
dag <- dag + node_td("CV", type="time_to_event", prob_fun=0.01)
out <- sim_n_datasets(dag, n_repeats=10, n_cores=1, n_sim=100,
                      max_t=20, data_format="start_stop")
#> Loading required namespace: Rfast
#> Error in dyn.load(file, DLLpath = DLLpath, ...): unable to load shared object '/home/runner/work/_temp/Library/RcppZiggurat/libs/RcppZiggurat.so':
#>   libgsl.so.27: cannot open shared object file: No such file or directory

Generate multiple datasets from a single `DAG` object