Design algorithm


Funz design ‘algorithm’ is the component that features funz to drive an external simulation software for an engineering objective. It mainly provides some methods in R language to define numerical simulations to perform, and returns information about engineering target.

algorithm cheatsheet

For now, following algorithms are available out-of-the-box:

  • Brent Funz-Brent
  • EGO Funz-EGO
  • GradientDescent Funz-GradientDescent
  • PSO Funz-PSO
  • RandomSampling Funz-RandomSampling
  • Sensitivity Funz-Sensitivity
  • NSGA2 Funz-NSGA2

For conveniency, a basic template is provided and should be used as a scratch project.


Once installed, Funz can integrate some more algorithms. Python and R wrappers provide install.Design('ALGORITHM') methods to add one of the previously bundled.

From scratch (when no bundle is available), you can add a new algorithm by hand just by copying file ‘MyAlgorithm.R’ inside ‘Funz/plugins/doe/’ directory.

Bundle implementation



These steps will guide you to build a basic algorithm, which is sufficient for most of your needs.

  1. Checkout the algorithm template:
  2. Edit the ‘build.xml’ file to replace MyAlgorithm by the name of your algorithm (usually, the name of the simulation software)
  3. Edit the ‘’ description file
  4. Rename and implement the ‘src/main/doe/MyAlgorithm.R’ file:
    • fill the header:
    • fill the constructor:
       #' constructor and initializer of algorithm
       #' @param options algorithm options
       #' @return algorithm object : environment options, status
       MyAlgorithm <- function(options) {
       algorithm = new.env()
    • fill the initalizer of experiments (typically with random points):
       #' first design building.
       #' @param algorithm object handling options, status, ...
       #' @param input the input variables and their properties (like min, max)
       #' @param output the output targets names
       #' @return matrix of first design step
       getInitialDesign <- function(algorithm,input,output) {
    • fill the iterator (considering previous calculations resutls):
       #' iterated design building.
       #' @param algorithm object handling options, status, ...
       #' @param X matrix of current doe variables
       #' @param Y matrix of current results
       #' @return matrix of next doe step
       getNextDesign <- function(algorithm,X,Y) {
    • fill the results renderer (with plots & string return):
       #' final analysis.
       #' @param algorithm object handling options, status, ...
       #' @param X matrix of current doe variables
       #' @param Y matrix of current results
       #' @return HTML string of analysis
       displayResults <- function(algorithm,X,Y) {
  5. Provide (at least) one test case ‘src/test/cases/MyTest.R’ containing R::testthat tests:
     ## This file should provide following objects, when loaded:
     # f : function
     # input.f : list of input dimensions, contains list of properties like lower & upper bounds of each dimensions
     # output.f : list of output dimensions
     # *.f : list of math properties. To be compared with algorithm results
     # [print.f] : method to print/plot the function for information
     f = function(X) { return(cos(pi*x)) }
     input.f = list(
     output.f = "cos_pi"
     root.f = 0.5
     test = function(algorithm_file) {
         results = run.algorithm(algorithm_file, options=NULL,fun=list(input=input.f,output=output.f,fun=f))
         test_that("cos_pi info",{expect_equal(as.numeric(results$info),root.f,tolerance = .0001)})
  6. Check all test cases using ant test, which returns a json report, and should not finished by FAILED

Improve this page