IDEMC: IDEMC Point Generation

In Tandethsquire/emulatorr: Emulation and History Matching Package

Description

Performs Implausibility-Driven Evolutionary Monte Carlo.

Usage

IDEMC(xsamp, ems, targets, s, sn, p, imp = 3, verbose = F, ...)

Arguments

xsamp	The initial sample of points, as a data.frame
ems	The emulators with which to evaluate implausibility
targets	The corresponding output targets
s	The number of points to generate in the burn-in phase
sn	The final number of points to generate
p	The proportion of points to keep in each new ladder rung
imp	The value of implausibility that is ultimately desired
verbose	Should logging messages be outputted? Default: F
...	Any additional parameters to pass to IDEMC_step

Details

Given a set of initial points (preferably space-filling across the space in question), the implausibility ladder is set up via a burn-in phase. Once the ladder of implausibilities has been determined, these are used to generate the full set of points.

This is a very computationally expensive procedure for generating points, and should only be used when it is strongly suspected that the target region is extremely small or has an interesting disconnected structure. For more mundane point generation, a more suitable approach is the functionality in generate_new_runs.

The burn-in phase starts with a rung defined over the full space (i.e. the implausibility for this rung is simply the maximum implausibility over the space). The implausibility of the next rung is chosen to be such that 30 A full complement of points are generated at this new level, and the process is repeated using these new points to find the next rung of the ladder. This iterates until the desired final implausibility, imp has been reached as a ladder rung.

Once the burn in has been performed, a full set of points are generated at each rung of the ladder (where it is assumed that the burn-in generated fewer points than required, so that sn<s).

Value

A list of data.frames, corresponding to the points generated.

References

Vernon, I. & Williamson, D. (2013) Efficient uniform designs for multi-wave computer experiments. arXiv:1309.3520

See Also

generate_new_runs for other point generation methods.

Examples

## Not run: 
ranges <- list(aSI = c(0.1, 0.8), aIR = c(0, 0.5), aSR = c(0, 0.05))
out_vars <- c('nS', 'nI', 'nR')
o_ems <- emulator_from_data(GillespieSIR, out_vars, ranges)
t_ems <- purrr::map(seq_along(o_ems), ~o_ems[[.]]$adjust(GillespieSIR, out_vars[[.]]))
z <- list(
 nS = list(val = 281, sigma = 10.43),
 nI = list(val = 30, sigma = 11.16),
 nR = list(val = 689, sigma = 14.32)
)
start_pts <- data.frame(
  aSI = runif(500, ranges$aSI[1], ranges$aSI[2]),
  aIR = runif(500, ranges$aIR[1], ranges$aIR[2]),
  aSR = runif(500, ranges$aSR[1], ranges$aSR[2])
)
result <- IDEMC(start_pts, t_ems, z, 50, 100, 0.3, imp = 2)

## End(Not run)

Tandethsquire/emulatorr documentation built on April 12, 2021, 1:08 a.m.