full_wave: Automatic Wave Calculation

In hmer: History Matching and Emulation Package

Automatic Wave Calculation

Description

Performs a full wave of emulation and history matching, given data.

Usage

full_wave(
  data,
  ranges,
  targets,
  old_emulators = NULL,
  prop_train = 0.7,
  cutoff = 3,
  nth = 1,
  verbose = interactive(),
  n_points = nrow(data),
  ...
)

Arguments

data	The data to train with.
ranges	The ranges of the input parameters
targets	The output targets to match to.
old_emulators	Any emulators from previous waves.
prop_train	What proportion of the data is used for training.
cutoff	The implausibility cutoff for point generation and diagnostics.
nth	The level of maximum implausibility to consider.
verbose	Should progress be printed to console?
n_points	The number of points to generate from generate_new_design.
...	Any arguments to be passed to emulator_from_data.

Details

This function uses all of the functionality from the package in a relatively conservative form. The function performs the following steps:

1) Split the data into a training set and a validation set, where prop_train indicates what proportion of the data is used to train.

2) Perform emulator training using emulator_from_data. If a more involved specification is desired, optional arguments can be passed to emulator_from_data using the ... argument.

3) Perform diagnostics on the trained emulators, removing emulators that do not display acceptable performance. Global emulator variance may also be modified to ensure that none of the emulators demonstrate misclassification errors (from classification_diag).

4) Ordering the remaining emulators from most restrictive to least restrictive on the dataset provided at this wave. Some point generation mechanisms terminate early if a point is ruled out by a single emulator, so the ordering ensures this happens earlier rather than later.

5) Generate the new points using the default method of generate_new_design, using the normal procedure (for details, see the description for generate_new_design). By default, it generates the same number of points as it was provided to train and validate on.

If the parameter old_emulators is provided, this should be a list of emulators used at all previous waves - for example if full_wave is used to do a second wave of history matching, then old_emulators would contain the list of first-wave emulators.

The function returns a list of two objects: emulators corresponding to this wave's emulators, and points corresponding to the new proposed points. The points can then be put into the simulator to generate runs for a subsequent wave.

Value

A list of two objects: points and emulators

Examples

 #excessive runtime
 ranges <- list(aSI = c(0.1, 0.8), aIR = c(0, 0.5), aSR = c(0, 0.05))
 default <- full_wave(do.call('rbind.data.frame', SIRSample), ranges,
  SIREmulators$targets)
 non_quad <- full_wave(do.call('rbind.data.frame', SIRSample), ranges,
  SIREmulators$targets, quadratic = FALSE)
 second <- full_wave(SIRMultiWaveData[[2]], ranges, SIREmulators$targets,
  old_emulators = SIRMultiWaveEmulators[[1]])
 

hmer documentation built on June 22, 2024, 9:22 a.m.