Movatterモバイル変換

MOEADr package

Felipe CampeloDepartment of Computer Science
Aston University
Birmingham, UK

Lucas Batista
Operations Research and Complex Systems Laboratory - ORCS Lab
Universidade Federal de Minas Gerais
Belo Horizonte, Brazil

Claus Aranha
Faculty of Engineering, Information and Systems
University of Tsukuba
Tsukuba, Japan

R package containing a component-based, modularimplementation of the Multiobjective Evolutionary Algorithm withDecomposition (MOEA/D) framework.

The MOEA/D framework is seen as a combination of specific designdecisions regarding several independent modules:

• Decomposition strategy;
  
• Aggregation function;
  
• Objective scaling strategy;
  
• Neighborhood assignment strategy;
  
• Variation Stack;
  
• Update strategy;
  
• Constraint handling method;
  
• Termination criteria.

This package provides several options for each module, as explainedin the documentation of its main function,MOEADr::moead().The input structure of this function is also explained in itsdocumentation. More details on the component-based approach behind theMOEADr package are available in our paper,The MOEADrPackage - A Component-Based Framework for Multiobjective EvolutionaryAlgorithms Based on Decomposition, available on the ArXiv:https://arxiv.org/abs/1807.06731.

To install the current release version in your system, simplyuse:

install.packages("MOEADr")

For the most up-to-date development version, install the githubversion using:

# install.packages("devtools")devtools::install_github("fcampelo/MOEADr")

Example

As a simple example, we can reproduce the original MOEA/D (Zhang andLi, 2007) and run it on a 30-variable ZDT1 function:

 ## 1: prepare test problem library(smoof) ZDT1 <- make_vectorized_smoof(prob.name  = "ZDT1",                               dimensions = 30) ## 2: set input parameters problem   <- list(name       = "ZDT1",                   xmin       = rep(0, 30),                   xmax       = rep(1, 30),                   m          = 2) decomp    <- list(name       = "SLD", H = 99) neighbors <- list(name       = "lambda",                   T          = 20,                   delta.p    = 1) aggfun    <- list(name       = "wt") variation <- list(list(name  = "sbx",                        etax  = 20, pc = 1),                   list(name  = "polymut",                        etam  = 20, pm = 0.1),                   list(name  = "truncate")) update    <- list(name       = "standard",                    UseArchive = FALSE) scaling   <- list(name       = "none") constraint<- list(name       = "none") stopcrit  <- list(list(name  = "maxiter",                     maxiter  = 200)) showpars  <- list(show.iters = "dots",                   showevery  = 10) seed      <- NULL ## 3: run MOEA/D out1 <- moead(problem = problem,                decomp = decomp, aggfun = aggfun, neighbors = neighbors, variation = variation,                update = update, constraint = constraint, scaling = scaling, stopcrit = stopcrit,               showpars = showpars, seed = seed) ## 3.1: For your convenience, you can also use the preset_moead() function to reproduce the above setup,  ##      and only modify the desired parts:  out2 <- moead(problem = problem,               preset = preset_moead("original"),                stopcrit = list(list(name = "maxiter", maxiter = 1000)),               showpars = showpars, seed = 42) # 4: Plot output: plot(out1$Y[,1], out1$Y[,2], type = "p", pch = 20)

Have fun!
Felipe