filter_for_Game {GPGame}

All-purpose filter

Description

Select candidate points for conditional simulations or for criterion evaluation, based on a "window" or a probability related to the equilibrium at hand.

Usage

filter_for_Game(
  n.s.target,
  model = NULL,
  predictions = NULL,
  type = "window",
  equilibrium = "NE",
  integcontrol,
  options = NULL,
  ncores = 1,
  random = TRUE,
  include.obs = FALSE,
  min.crit = 1e-12,
  nsamp = NULL,
  Nadir = NULL,
  Shadow = NULL,
  target = NULL
)

Arguments

n.s.target

scalar or vector of number of strategies (one value per player) to select. For NE, if n.s.target is a scalar then each player will have round(n.s.target^(1/nobj) strategies.

model

is a list of nobj nobj km objects

predictions

is a list of size nobj

type

either "window", "PND" or "Pnash", see details

equilibrium

either 'NE', 'KSE' or 'NKSE' for Nash/Kalai-Smoridinsky/Nash-Kalai-Smoridinsky equilibria

integcontrol

is a list containing: integ.pts, a [npts x dim] matrix defining the grid, expanded.indices a matrix containing the indices of the integ.pts on the grid and n.s, a nobj vector containting the number of strategies per player

options

a list containing either the window (matrix or target) or the parameters for Pnash: method ("simu" or "exact") and nsim

ncores

mclapply is used if > 1 for parallel evaluation

random

Boolean. If FALSE, the best points according to the filter criterion are chosen, otherwise the points are chosen by random sampling with weights proportional to the criterion.

include.obs

Boolean. If TRUE, the observations are included to the filtered set.

min.crit

Minimal value for the criterion, useful if random = TRUE.

nsamp

number of samples to estimate the probability of non-domination, useful when type=PND and nobj>3.

Nadir, Shadow

optional vectors of size nobj. Replaces the nadir or shadow point for KSE. If only a subset of values needs to be defined, the other coordinates can be set to Inf (resp. -Inf).

target

a vector of target values for the objectives to use the calibration mode

Details

If type == "windows", points are ranked based on their distance to option$window (when it is a target vector), or based on the probability that the response belongs to option$window. The other options, "PND" (probability of non-domination, i.e., of not being dominated by the current Pareto front) and "Pnash" (probability of realizing a Nash equilibrium) base the ranking of points on the associated probability.

Value

List with two elements: I indices selected and crit the filter metric at all candidate points