space_removal {hmer} | R Documentation |
Percentage of Space Removed
Description
For a wave of emulators, estimates the proportion of space removed at this wave.
Usage
space_removal(
ems,
targets,
points = NULL,
ppd = NULL,
cutoff = 3,
individual = TRUE
)
Arguments
ems |
The emulators to compute over, as a list |
targets |
The output target values |
points |
The points to test against |
ppd |
If no points are provided and uniform grid is wanted, the number of points per parameter dimension. |
cutoff |
The cutoff value for implausibility |
individual |
If true, gives emulator-by-emulator results; otherwise works with maximum implausibility |
Details
Given a collection of emulators corresponding to a wave, we can look at an estimate of the proportion of points from previous waves that will be accepted at this wave, either on an emulator-by-emulator basis (to see which outputs are most restrictive) or as an all-wave determination.
Naturally, such a statement will be an estimate of the restriction on the full space (which will become more unreliable for higher dimensions), but it can give an order-of-magnitude statement, or useful comparators between different emulators in a wave.
If no points are provided, the training points for the emulators are used. For best results, a good number of points should be given: typically one should consider using as many points as one knows to be in the NROY space (including any validation points, if accessible).
Value
A numeric corresponding to the proportions of points removed.
See Also
space_removed
for a visualisation of the space removal.
Examples
space_removal(SIREmulators$ems, SIREmulators$targets,
rbind(SIRSample$training, SIRSample$validation))
space_removal(SIREmulators$ems, SIREmulators$targets,
rbind(SIRSample$training, SIRSample$validation), individual = FALSE)