bimodal_emulator_from_data {hmer}

Bimodal Emulation

Description

Performs emulation of bimodal outputs and/or systems.

Usage

bimodal_emulator_from_data(
  data,
  output_names,
  ranges,
  input_names = names(ranges),
  verbose = interactive(),
  na.rm = FALSE,
  ...
)

Arguments

data

The data to train emulators on (as in variance_emulator_from_data)

output_names

The names of the outputs to emulate

ranges

The parameter ranges

input_names

The names of the parameters (by default inferred from ranges)

verbose

Should status updates be provided?

na.rm

Should NA values be removed before training?

...

Any other parameters to pass to emulator training

Details

This function is deprecated in favour of using emulator_from_data with argument emulator_type = "multistate". See the associated help file.

In many stochastic systems, particularly disease models, the outputs exhibit bimodality - a familiar example is where a disease either takes off or dies out. In these cases, it is not sensible to emulate the outputs based on all realisations, and instead we should emulate each mode separately.

This function first tries to identify bimodality. If detected, it determines which of the outputs in the data exhibits the bimodality: to these two separate emulators are trained, one to each mode. The emulators are provided with any data that is relevant to their training; for example, bimodality can exist in some regions of parameter space but not others. Points where bimodality is present have their realisations allocated between the two modes while points where no bimodality exists have their realisations provided to both modes. Targets that do not exhibit bimodality are trained as a normal stochastic output: that is, using the default of variance_emulator_from_data.

The function also estimates the proportion of realisations in each mode for the set of outputs. This value is also emulated as a deterministic emulator and included in the output.

The output of the function is a list, containing three objects: mode1, mode2, and prop. The first two objects have the form produced by variance_emulator_from_data while prop has the form of an emulator_from_data output.

Value

A list (mode1, mode2, prop) of emulator lists and objects.

Examples

  # Excessive runtime
  # Use the stochastic SIR dataset
  SIR_ranges <- list(aSI = c(0.1, 0.8), aIR = c(0, 0.5), aSR = c(0, 0.05))
  SIR_names <- c("I10", "I25", "I50", "R10", "R25", "R50")
  b_ems <- bimodal_emulator_from_data(SIR_stochastic$training, SIR_names, SIR_ranges)