Helper functions for updating components of an Experiment.

Helper functions for updating DGPs, Methods, Evaluators, and Visualizers already added to an Experiment.

Usage

update_dgp(experiment, dgp, name, ...)

update_method(experiment, method, name, ...)

update_evaluator(experiment, evaluator, name, ...)

update_visualizer(experiment, visualizer, name, ...)

Arguments

experiment: An Experiment object.
dgp: A DGP object.
name: An existing name identifying the object to be updated.
...: Not used.
method: A Method object.
evaluator: An Evaluator object.
visualizer: A Visualizer object.

Value

The original Experiment object passed to update_*.

Examples

## create toy DGPs, Methods, Evaluators, and Visualizers

# generate data from normal distribution with 100 samples
dgp1 <- create_dgp(
  .dgp_fun = function(n) rnorm(n), .name = "DGP", n = 100
)
# generate data from normal distribution with 500 samples
dgp2 <- create_dgp(
  .dgp_fun = function(n) rnorm(n), .name = "DGP", n = 500
)

# compute mean of data
mean_method <- create_method(
  .method_fun = function(x) list(mean = mean(x)), .name = "Method"
)
# compute mean of data
median_method <- create_method(
  .method_fun = function(x) list(mean = median(x)), .name = "Method"
)

# evaluate SD of mean(x) across simulation replicates
sd_eval <- create_evaluator(
  .eval_fun = function(fit_results, vary_params = NULL) {
    group_vars <- c(".dgp_name", ".method_name", vary_params)
    fit_results %>%
      dplyr::group_by(dplyr::across(tidyselect::all_of(group_vars))) %>%
      dplyr::summarise(sd = sd(mean), .groups = "keep")
  },
  .name = "Evaluator"
)
# evaluate Variance of mean(x) across simulation replicates
var_eval <- create_evaluator(
  .eval_fun = function(fit_results, vary_params = NULL) {
    group_vars <- c(".dgp_name", ".method_name", vary_params)
    fit_results %>%
      dplyr::group_by(dplyr::across(tidyselect::all_of(group_vars))) %>%
      dplyr::summarise(var = var(mean), .groups = "keep")
  },
  .name = "Evaluator"
)

# plot SD of method results across simulation replicates
sd_plot <- create_visualizer(
  .viz_fun = function(fit_results, eval_results, vary_params = NULL,
                      eval_name = 1) {
    if (!is.null(vary_params)) {
      add_aes <- ggplot2::aes(
        x = .data[[unique(vary_params)]], y = sd, color = .dgp_name
      )
    } else {
      add_aes <- ggplot2::aes(x = .dgp_name, y = sd)
    }
    plt <- ggplot2::ggplot(eval_results[[eval_name]]) +
      add_aes +
      ggplot2::geom_point()
    if (!is.null(vary_params)) {
      plt <- plt + ggplot2::geom_line()
    }
    return(plt)
  },
  .name = "Visualizer"
)
# plot variance of method results across simulation replicates
var_plot <- create_visualizer(
  .viz_fun = function(fit_results, eval_results, vary_params = NULL,
                      eval_name = 1) {
    if (!is.null(vary_params)) {
      add_aes <- ggplot2::aes(
        x = .data[[unique(vary_params)]], y = var, color = .dgp_name
      )
    } else {
      add_aes <- ggplot2::aes(x = .dgp_name, y = var)
    }
    plt <- ggplot2::ggplot(eval_results[[eval_name]]) +
      add_aes +
      ggplot2::geom_point()
    if (!is.null(vary_params)) {
      plt <- plt + ggplot2::geom_line()
    }
    return(plt)
  },
  .name = "Visualizer"
)

# initialize experiment with toy DGPs, Methods, Evaluators, and Visualizers
# using piping %>% and add_* functions
experiment <- create_experiment(name = "Experiment Name") %>%
  add_dgp(dgp1) %>%
  add_method(mean_method) %>%
  add_evaluator(sd_eval) %>%
  add_visualizer(sd_plot)
print(experiment)
#> Experiment Name: Experiment Name 
#>    Saved results at: results/Experiment Name 
#>    DGPs: DGP 
#>    Methods: Method 
#>    Evaluators: Evaluator 
#>    Visualizers: Visualizer 
#>    Vary Across: None

# example usage of update_* functions
experiment <- experiment %>%
  update_dgp(dgp2, "DGP") %>%
  update_method(median_method, "Method") %>%
  update_evaluator(var_eval, "Evaluator") %>%
  update_visualizer(var_plot, "Visualizer")