plot_errorABC {poolABC}

Prediction error plots for ABC using a list

Description

Plots the prediction error computed from a leave-one-out cross validation for ABC parameter inference. This function takes as input a list created when performing cross validation and allows the user to select which ABC algorithm and point estimate statistic to plot.

Usage

plot_errorABC(
  x,
  method,
  statistic,
  index,
  transformation = "none",
  main = NULL
)

Arguments

x

is a list produced by a leave-one-out cross validation of ABC. This list should contain the prediction errors computed using the rejection and/or regression algorithm. For each of those methods, the prediction error obtained using three different point estimates of the posterior should be included in this list.

method

a character that can be either 'rej' or 'reg' indicating whether you wish to plot the prediction error computed with a rejection or regression based ABC algorithm.

statistic

a character that can be 'mode', 'median' or 'mean' indicating if you wish to plot the prediction error obtained using the mode, median or mean as the point estimate of the posterior.

index

an integer indicating which parameter to look at. It corresponds to a column on a matrix. So, to plot the first parameter, corresponding to the first column, select 1. To plot the second parameter, select 2 and so on.

transformation

default is none. It can also be 'log' if you wish to transform both the true and estimated values using a log10 scale.

main

is an optional character input. It will be used as the title of the plot. If NULL (default), then a generic title will be used instead.

Details

These plots help in visualizing the quality of the estimation and the effect of the chosen tolerance level or point estimate statistic.

Value

a plot of the estimated value of the parameter (in the y-axis) versus the true parameter value (in the x-axis). A line marking the perfect correspondence between the true and estimated values is also plotted. Thus, the closer the points are to that line, the lower the prediction error is.

Examples

# load the matrix with parameter values
data(params)
# load the matrix with simulated parameter values
data(sumstats)
# load the matrix with the prior limits
data(limits)

# perform a leave-one-out cross validation for ABC
mysim <- simulationABC(params = params, sumstats = sumstats, limits, nval = 10,
tol = 0.1, method = "regression")

# plot the prediction error for a given parameter
plot_errorABC(x = mysim, method = "reg", statistic = "median", index = 1)