Description.

parSim is an R package that provides convenient functionality to perform flexible simulations in parallel using parabar backends.

Installation

to install from CRAN run install.packages("parSim")
to install the latest version from GitHub run remotes::install_github("SachaEpskamp/parSim")

Example

# Load the package.
library(parSim)
# Determine a function to evaluate for each simulation condition.
bias <- function(x, y) {
    # Perform some computation.
    result <- abs(x - y)
    # Return the result.
    return(result)
}
# Run the simulation.
results <- parSim(
    # The simulation conditions.
    sample_size = c(50, 100, 250),
    beta = c(0, 0.5, 1),
    sigma = c(0.25, 0.5, 1),
    # The expression to evaluate for each simulation condition.
    expression = {
        # Generate the data.
        x <- rnorm(sample_size)
        y <- beta * x + rnorm(sample_size, sigma)
        # Fit the model.
        fit <- lm(y ~ x)
        # Compute the relevant quantities.
        beta_estimate <- coef(fit)[2]
        r_squared <- summary(fit)$r.squared
        bias <- bias(beta, beta_estimate)
        # Return in a compatible format.
        list(
            beta_estimate = beta_estimate,
            r_squared = r_squared,
            bias = bias
        )
    },
    # The number of replications.
    replications = 100,
    # The conditions to exclude.
    exclude = sample_size == 50 | beta <= 0.5,
    # The variables to export.
    export = c("bias"),
    # No packages are required for export.
    packages = NULL,
    # Do not save the results.
    write = FALSE,
    # Execute the simulation on a single core.
    nCores = 1,
    # Show the progress bar.
    progress = TRUE
)
# Print the head of the results.
head(results)

We can also use the configure_bar function (i.e., exported from the parabar package) to customize the progress bar.

# Configure the progress bar.
configure_bar(
    type = "modern",
    format = "[:bar] [:percent] [:elapsed]",
    show_after = 0.15
)

Then, we can proceed with running the simulation as before.

# Run the simulation again with more cores and the updated progress bar.
results <- parSim(
    # The simulation conditions.
    sample_size = c(50, 100, 250),
    beta = c(0, 0.5, 1),
    sigma = c(0.25, 0.5, 1),
    # The expression to evaluate for each simulation condition.
    expression = {
        # Generate the data.
        x <- rnorm(sample_size)
        y <- beta * x + rnorm(sample_size, sigma)
        # Fit the model.
        fit <- lm(y ~ x)
        # Compute the relevant quantities.
        beta_estimate <- coef(fit)[2]
        r_squared <- summary(fit)$r.squared
        bias <- bias(beta, beta_estimate)
        # Return in a compatible format.
        list(
            beta_estimate = beta_estimate,
            r_squared = r_squared,
            bias = bias
        )
    },
    # The number of replications.
    replications = 1000,
    # The conditions to exclude.
    exclude = sample_size == 50,
    # The variables to export.
    export = c("bias"),
    # No packages are required for export.
    packages = NULL,
    # Save the results to a temporary file.
    write = TRUE,
    # Execute the simulation in parallel.
    nCores = 4,
    # Show the progress bar.
    progress = TRUE
)
# Print the tail of the results.
tail(results)

Finally, we can also plot the results, for example, using the ggplot2 package as follows.

# Load relevant libraries.
library(ggplot2)
library(tidyr)
# Pre-process the results in long format for plotting.
results_long <- tidyr::gather(results, metric, value, beta_estimate:bias)
# Make factors with nice labels for plotting.
results_long$sigma_factor <- factor(
    x = results_long$sigma,
    levels = c(0.25, 0.5, 1),
    labels = c("Sigma: 0.25", "Sigma: 0.5", "Sigma: 1")
)
# Plot.
ggplot2::ggplot(
    results_long, ggplot2::aes(
        x = factor(sample_size), y = value, fill = factor(beta))
    ) +
    ggplot2::facet_grid(
        metric ~ sigma_factor, scales = "free"
    ) +
    ggplot2::geom_boxplot() +
    ggplot2::theme_bw() +
    ggplot2::xlab("Sample size") +
    ggplot2::ylab("") +
    ggplot2::scale_fill_discrete("Beta")