The choice of errors to plot is controlled by errtype, and can be one
of four things: normal, corresponding to the regular standardised errors; eigen,
corresponding to the errors after reordering given by the eigendecomposition
of the emulator covariance matrix; chol, similarly deriving errors after Cholesky
decomposition; and cholpivot, deriving the errors after pivoted Cholesky decomposition.
What the errors are plotted with respect to is controlled by xtype. The options
are index, which plots them in their order in the validation set; em, which plots errors
with respect to the emulator prediction at that point; and any named parameter of the
model, which plots with respect to the values of that parameter.
Finally, the plot type is controlled by plottype: this can be one of normal,
which plots the errors; or qq, which produces a Q-Q plot of the errors.
The default output is to plot the standardised errors (with no decomposition)
against the ordering in the validation set; i.e. errtype = "normal",
xtype = "index", plottype = "normal".
Some combinations are not permitted, as the output would not be meaningful. Errors
arising from an eigendecomposition cannot be plotted against either emulator prediction
or a particular parameter (due to the transformation induced by the eigendecomposition);
Q-Q plots are not plotted for a non-decomposed set of errors, as the correlation
between errors makes it much harder to interpret.