test.all: To test an emulator using leave-one-out cross-validation

In scrim-network/Stilt-Rpkge: Separable Gaussian Process Interpolation (Emulation)

Description

To test an emulator using the leave-one-out cross validation analysis, whereby each ensemble run is first excluded from an emulator, and the emulator is then used to predict at the parameter settings of the excluded run. The process is repeated for all runs within the ensemble. The function also makes a plot of emulator predictions vs. actual model output, and a plot of normalized prediction errors for each excluded run.

Usage

test.all(final.emul, t.plot)

Arguments

final.emul	Standard emulator object, that can be generated, for example, by 'emulator' function
t.plot	Time index of the emulator to analyze

Details

For a perfect emulator, emulator predictions vs. actual model output will fall exactly on a 1:1 line. If a withheld run is at the corner of parameter space, prediction gives a warning. Such runs are omitted from the plot.

See Also

test.csv

Examples

# Cross-validate the 1D-parameter ensemble emulator at the first time index
data(emul.1D)
test.all(emul.1D, 1)

scrim-network/Stilt-Rpkge documentation built on May 29, 2019, 4:07 p.m.