GPareto-package: Package GPareto

Description Details Note Author(s) References See Also Examples

Description

Multi-objective optimization and quantification of uncertainty on Pareto fronts, using Gaussian process models.

Details

Important functions:
GParetoptim
easyGParetoptim
crit_optimizer
plotGPareto
CPF

Note

Part of this work has been conducted within the frame of the ReDice Consortium, gathering industrial (CEA, EDF, IFPEN, IRSN, Renault) and academic (Ecole des Mines de Saint-Etienne, INRIA, and the University of Bern) partners around advanced methods for Computer Experiments. (http://redice.emse.fr/).

The authors would like to thank Yves Deville for his precious advices in R programming and packaging, as well as Olivier Roustant and David Ginsbourger for testing and suggestions of improvements for this package. We would also like to thank Tobias Wagner for providing his Matlab codes for the SMS-EGO strategy.

Author(s)

Mickael Binois, Victor Picheny

References

M. Binois, D. Ginsbourger and O. Roustant (2015), Quantifying Uncertainty on Pareto Fronts with Gaussian process conditional simulations, European Journal of Operational Research, 243(2), 386-394.

O. Roustant, D. Ginsbourger and Yves Deville (2012), DiceKriging, DiceOptim: Two R Packages for the Analysis of Computer Experiments by Kriging-Based Metamodeling and Optimization, Journal of Statistical Software, 51(1), 1-55, https://www.jstatsoft.org/v51/i01/.

M. T. Emmerich, A. H. Deutz, J. W. Klinkenberg (2011), Hypervolume-based expected improvement: Monotonicity properties and exact computation, Evolutionary Computation (CEC), 2147-2154.

V. Picheny (2015), Multiobjective optimization using Gaussian process emulators via stepwise uncertainty reduction, Statistics and Computing, 25(6), 1265-1280.

T. Wagner, M. Emmerich, A. Deutz, W. Ponweiser (2010), On expected-improvement criteria for model-based multi-objective optimization, Parallel Problem Solving from Nature, 718-727, Springer, Berlin.

J. D. Svenson (2011), Computer Experiments: Multiobjective Optimization and Sensitivity Analysis, Ohio State University, PhD thesis.

C. Chevalier (2013), Fast uncertainty reduction strategies relying on Gaussian process models, University of Bern, PhD thesis.

M. Binois, V. Picheny (2019), GPareto: An R Package for Gaussian-Process-Based Multi-Objective Optimization and Analysis, Journal of Statistical Software, 89(8), 1–30.

See Also

DiceKriging-package, DiceOptim-package

Examples

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
## Not run: 
#------------------------------------------------------------
# Example 1 : Surrogate-based multi-objective Optimization with postprocessing
#------------------------------------------------------------
set.seed(25468)

d <- 2 
fname <- P2

plotParetoGrid(P2) # For comparison

# Optimization
budget <- 25 
lower <- rep(0, d) 
upper <- rep(1, d)
     
omEGO <- easyGParetoptim(fn = fname, budget = budget, lower = lower, upper = upper)

# Postprocessing
plotGPareto(omEGO, add= FALSE, UQ_PF = TRUE, UQ_PS = TRUE, UQ_dens = TRUE)


## End(Not run)
#------------------------------------------------------------
# Example 2 : Surrogate-based multi-objective Optimization including a cheap function
#------------------------------------------------------------
set.seed(42)
library(DiceDesign)

d <- 2 

fname <- P1
n.grid <- 19
test.grid <- expand.grid(seq(0, 1, length.out = n.grid), seq(0, 1, length.out = n.grid))
nappr <- 15 
design.grid <- maximinESE_LHS(lhsDesign(nappr, d, seed = 42)$design)$design
response.grid <- t(apply(design.grid, 1, fname))

mf1 <- km(~., design = design.grid, response = response.grid[,1])
mf2 <- km(~., design = design.grid, response = response.grid[,2])
model <- list(mf1, mf2)

nsteps <- 1 
lower <- rep(0, d)
upper <- rep(1, d)

# Optimization with fastfun: hypervolume with discrete search

optimcontrol <- list(method = "discrete", candidate.points = test.grid)
omEGO2 <- GParetoptim(model = model, fn = fname, cheapfn = branin, crit = "SMS",
                      nsteps = nsteps, lower = lower, upper = upper,
                      optimcontrol = optimcontrol)
print(omEGO2$par)
print(omEGO2$values)

## Not run:  
plotGPareto(omEGO2)

#------------------------------------------------------------
# Example 3 : Surrogate-based multi-objective Optimization (4 objectives)
#------------------------------------------------------------
set.seed(42)
library(DiceDesign)

d <- 5 

fname <- DTLZ3
nappr <- 25
design.grid <- maximinESE_LHS(lhsDesign(nappr, d, seed = 42)$design)$design
response.grid <- t(apply(design.grid, 1, fname, nobj = 4))
mf1 <- km(~., design = design.grid, response = response.grid[,1])
mf2 <- km(~., design = design.grid, response = response.grid[,2])
mf3 <- km(~., design = design.grid, response = response.grid[,3])
mf4 <- km(~., design = design.grid, response = response.grid[,4])

# Optimization
nsteps <- 5 
lower <- rep(0, d) 
upper <- rep(1, d)     
omEGO3 <- GParetoptim(model = list(mf1, mf2, mf3, mf4), fn = fname, crit = "EMI",
                      nsteps = nsteps, lower = lower, upper = upper, nobj = 4)
print(omEGO3$par)
print(omEGO3$values)
plotGPareto(omEGO3)

#------------------------------------------------------------
# Example 4 : quantification of uncertainty on Pareto front
#------------------------------------------------------------
library(DiceDesign)
set.seed(42)

nvar <- 2

# Test function P1
fname <- "P1"

# Initial design
nappr <- 10
design.grid <- maximinESE_LHS(lhsDesign(nappr, nvar, seed = 42)$design)$design
response.grid <- t(apply(design.grid, 1, fname))

PF <- t(nondominated_points(t(response.grid)))

# kriging models : matern5_2 covariance structure, linear trend, no nugget effect
mf1 <- km(~., design = design.grid, response = response.grid[,1])
mf2 <- km(~., design = design.grid, response = response.grid[,2])

# Conditional simulations generation with random sampling points 
nsim <- 100 # increase for better results
npointssim <- 1000 # increase for better results
Simu_f1 <- matrix(0, nrow = nsim, ncol = npointssim)
Simu_f2 <- matrix(0, nrow = nsim, ncol = npointssim)
design.sim <- array(0, dim = c(npointssim, nvar, nsim))

for(i in 1:nsim){
  design.sim[,,i] <- matrix(runif(nvar*npointssim), nrow = npointssim, ncol = nvar)
  Simu_f1[i,] <- simulate(mf1, nsim = 1, newdata = design.sim[,,i], cond = TRUE,
                         checkNames = FALSE, nugget.sim = 10^-8)
  Simu_f2[i,] <- simulate(mf2, nsim = 1, newdata = design.sim[,,i], cond = TRUE,
                         checkNames = FALSE, nugget.sim = 10^-8)
}

# Computation of the attainment function and Vorob'ev Expectation
CPF1 <- CPF(Simu_f1, Simu_f2, response.grid, paretoFront = PF)

summary(CPF1)

plot(CPF1)

# Display of the symmetric deviation function
plotSymDevFun(CPF1)

## End(Not run)

GPareto documentation built on May 31, 2021, 5:09 p.m.