NeuralEstimators (version 0.1.2)

plotdistribution: Plot the empirical sampling distribution of an estimator.

Description

Plot the empirical sampling distribution of an estimator.

Usage

plotdistribution(
  df,
  type = c("box", "density", "scatter"),
  parameter_labels = NULL,
  estimator_labels = ggplot2::waiver(),
  truth_colour = "black",
  truth_size = 8,
  truth_line_size = NULL,
  pairs = FALSE,
  upper_triangle_plots = NULL,
  legend = TRUE,
  return_list = FALSE,
  flip = FALSE
)

Value

a list of 'ggplot' objects or, if pairs = TRUE, a single 'ggplot'.

Arguments

df

a long form data frame containing fields estimator, parameter, estimate, truth, and a column (e.g., replicate) to uniquely identify each observation.

type

string indicating whether to plot kernel density estimates for each individual parameter (type = "density") or scatter plots for all parameter pairs (type = "scatter").

parameter_labels

a named vector containing parameter labels.

estimator_labels

a named vector containing estimator labels.

truth_colour

the colour used to denote the true parameter value.

truth_size

the size of the point used to denote the true parameter value (applicable only for type = "scatter").

truth_line_size

the size of the cross-hairs used to denote the true parameter value. If NULL (default), the cross-hairs are not plotted. (applicable only for type = "scatter").

pairs

logical; should we combine the scatter plots into a single pairs plot (applicable only for type = "scatter")?

upper_triangle_plots

an optional list of plots to include in the uppertriangle of the pairs plot.

legend

Flag; should we include the legend (only applies when constructing a pairs plot)

return_list

Flag; should the parameters be split into a list?

flip

Flag; should the boxplots be "flipped" using coord_flip() (default FALSE)?

Examples

if (FALSE) {
# In the following, we have two estimators and, for each parameter, 50 estimates
# from each estimator.

estimators <- c("Estimator 1", "Estimator 2")
estimator_labels <- c("Estimator 1" = expression(hat(theta)[1]("·")),
                      "Estimator 2" = expression(hat(theta)[2]("·")))

# Single parameter:
df <- data.frame(
  estimator = estimators, truth = 0, parameter = "mu",
  estimate  = rnorm(2*50),
  replicate = rep(1:50, each = 2)
)
parameter_labels <- c("mu" = expression(mu))
plotdistribution(df)
plotdistribution(df, type = "density")
plotdistribution(df, parameter_labels = parameter_labels, estimator_labels = estimator_labels)

# Two parameters:
df <- rbind(df, data.frame(
  estimator = estimators, truth = 1, parameter = "sigma",
  estimate  = rgamma(2*50, shape = 1, rate = 1),
  replicate = rep(1:50, each = 2)
))
parameter_labels <- c(parameter_labels, "sigma" = expression(sigma))
plotdistribution(df, parameter_labels = parameter_labels)
plotdistribution(df, parameter_labels = parameter_labels, type = "density")
plotdistribution(df, parameter_labels = parameter_labels, type = "scatter")

# Three parameters:
df <- rbind(df, data.frame(
  estimator = estimators, truth = 0.25, parameter = "alpha",
  estimate  = 0.5 * runif(2*50),
  replicate = rep(1:50, each = 2)
))
parameter_labels <- c(parameter_labels, "alpha" = expression(alpha))
plotdistribution(df, parameter_labels = parameter_labels)
plotdistribution(df, parameter_labels = parameter_labels, type = "density")
plotdistribution(df, parameter_labels = parameter_labels, type = "scatter")
plotdistribution(df, parameter_labels = parameter_labels, type = "scatter", pairs = TRUE)

# Pairs plot with user-specified plots in the upper triangle:
upper_triangle_plots <- lapply(1:3, function(i) {
  x = rnorm(10)
  y = rnorm(10)
  shape = sample(c("Class 1", "Class 2"), 10, replace = TRUE)
  ggplot() +
    geom_point(aes(x = x, y = y, shape = shape)) + 
    labs(shape = "") +
    theme_bw()
})
plotdistribution(
    df, 
    parameter_labels = parameter_labels, estimator_labels = estimator_labels,
    type = "scatter", pairs = TRUE, upper_triangle_plots = upper_triangle_plots
    )}