screeplot: Screeplots of model results

Description

Draws screeplots of performance statistics for models of varying complexity.

Usage

# S3 method for mat
screeplot(x, k, restrict = 20,
          display = c("rmsep", "avg.bias",
                      "max.bias", "r.squared"),
          weighted = FALSE,  col = "red", xlab = NULL,
          ylab = NULL, main = NULL, sub = NULL, ...)
# S3 method for bootstrap.mat
screeplot(x, k, restrict = 20,
          display = c("rmsep","avg.bias","max.bias",
                      "r.squared"),
          legend = TRUE, loc.legend = "topright",
          col = c("red", "blue"),
          xlab = NULL, ylab = NULL,
          main = NULL, sub = NULL,
          ...,
          lty = c("solid","dashed"))

Arguments

x: object of class mat and bootstrap.mat.
k: number of analogues to use. If missing 'k' is chosen automatically as the 'k' that achieves lowest RMSE.
restrict: logical; restrict comparison of k-closest model to k \(<=\) restrict.
display: which aspect of x to plot? Partial match.
weighted: logical; should the analysis use weighted mean of env data of analogues as fitted/estimated values?
xlab, ylab: x- and y-axis labels respectively.
main, sub: main and subtitle for the plot.
legend: logical; should a legend be displayed on the figure?
loc.legend: character; a keyword for the location of the legend. See legend for details of allowed keywords.
col: Colours for lines drawn on the screeplot. Method for class "bootstrap.mat" takes a vector of two colours.
lty: vector detailing the line type to use in drawing the screeplot of the apparent and bootstrap statistics, respectively. Code currently assumes that length(lty) is 2.
...: arguments passed to other graphics functions.

Author

Gavin Simpson

Details

Screeplots are often used to graphically show the results of cross-validation or other estimate of model performance across a range of model complexity.

Four measures of model performance are currently available: i) root mean square error of prediction (RMSEP); ii) average bias --- the mean of the model residuals; iii) maximum bias --- the maximum average bias calculated for each of n sections of the gradient of the environmental variable; and v) model \(R^2\).

For the maximum bias statistic, the response (environmental) gradient is split into n = 10 sections.

For the bootstrap method, apparent and bootstrap versions of these statistics are available and plotted.

Examples

Run this code

## Imbrie and Kipp example
## load the example data
data(ImbrieKipp)
data(SumSST)
data(V12.122)

## merge training and test set on columns
dat <- join(ImbrieKipp, V12.122, verbose = TRUE)

## extract the merged data sets and convert to proportions
ImbrieKipp <- dat[[1]] / 100
V12.122 <- dat[[2]] / 100

## fit the MAT model using the chord distance measure
(ik.mat <- mat(ImbrieKipp, SumSST, method = "chord"))

screeplot(ik.mat)