collectResult: Collect Result Information

Description

Merge both log and log.newinds data for a complete data.frame with information about progress (both on training data and on holdout data) and ressource usage.

Usage

collectResult(
  ecr.object,
  aggregate.perresult = list(domHV = function(x) computeHV(x, ref.point)),
  aggregate.perobjective = list("min", "mean", "max"),
  ref.point = smoof::getRefPoint(ecr.object$control$task$fitness.fun),
  cor.fun = cor
)

Value

data.frame

Arguments

ecr.object: [MosmafsResult] slickEcr() result to analyse.
aggregate.perresult: [list] list of functions to apply to fitness and holdout fitness. Every entry must either be a character(1) naming the function to use, or a function, in which that entry must have a name. Each function must return exactly one numeric value when fed a fitness matrix of one generation. This is ignored for single-objective runs.
aggregate.perobjective: [list] list of functions to apply to fitness and holdout fitness matrix rows, formatted like aggregate.perresult. Each function must return exactly one numeric value when fed a fitness vector.
ref.point: [numeric] reference point to use for HV computation.
cor.fun: [function] function to use for calculation of correlation between objective and holdout. Must take two numeric arguments and return a numeric(1).

Examples

Run this code

library(mlrCPO)

# Setup of optimization problem 
ps.simple <- pSS(
 a: numeric [0, 10],
 selector.selection: logical^10)
 
mutator.simple <- combine.operators(ps.simple,
 a = mutGauss,
 selector.selection = mutBitflipCHW)
 
crossover.simple <- combine.operators(ps.simple,
 a = recSBX,
 selector.selection = recPCrossover)

initials <- sampleValues(ps.simple, 30, discrete.names = TRUE)

fitness.fun <- smoof::makeMultiObjectiveFunction(
 sprintf("simple test"),
 has.simple.signature = FALSE, par.set = ps.simple, n.objectives = 2, 
 noisy = TRUE,
 ref.point = c(10, 1),
 fn = function(args, fidelity = NULL, holdout = FALSE) {
   pfeat <- mean(args$selector.selection)
   c(perform = args$a, pfeat = pfeat)
 })
fitness.fun.single <- smoof::makeMultiObjectiveFunction(
 sprintf("simple test"),
 has.simple.signature = FALSE, par.set = ps.simple, n.objectives = 1, 
 noisy = TRUE,
 ref.point = c(10),
 fn = function(args, fidelity = NULL, holdout = FALSE) {
   propfeat <- mean(args$selector.selection)
   c(propfeat = propfeat)
 })

# Run NSGA-II
results <- slickEcr(fitness.fun = fitness.fun, lambda = 10, population = initials, 
 mutator = mutator.simple, recombinator = crossover.simple, generations = 10)

# Collect results
colres <- collectResult(results)
print(colres)

Run the code above in your browser using DataCamp Workspace