Introduction

This is a Python wrapping using the C++ Implementation of the test suite for the Special Session on Large Scale Global Optimization at 2013 IEEE Congress on Evolutionary Computation.

Note

If you are to use any part of this code, please cite the following publications: X. Li, K. Tang, M. Omidvar, Z. Yang and K. Qin, "Benchmark Functions for the CEC'2013 Special Session and Competition on Large Scale Global Optimization," Technical Report, Evolutionary Computation and Machine Learning Group, RMIT University, Australia, 2013. http://goanna.cs.rmit.edu.au/~xiaodong/cec13-lsgo/competition/

Requirements

GNU Make
GNU G++
Python
Cython

Testing Environment

Debian GNU/Linux jessie/sid
GNU Make 3.81
g++ (Debian 4.7.3-4) 4.7.3
Python 2.7 and Python 3.2
numpy 1.8.1
cython 0.20.1

Results with Travis-CI

Instalation

Very easy, pip install cec2013lsgo ;-).

You can also download from https://github.com/dmolina/cec2013lsgo, and do python setup.py install [--user]. (the option --user is for installing the package locally, as a normal user (interesting when you want to run the experiments in a cluster/server without administration permissions).

To compile the source code in C++

The source code in C++ is also available. If you want to compile only the C++ version type in 'make' in the root directory of source code.

There are two equivalents demo executables: demo and demo2.

REMEMBER: To run the C++ version the directory cdatafiles must be available in the working directory. In the python version, these files are included in the packages, so it is not needed.

Tests

The source code has tests to check the information about each function, and the results obtained with the C version using the solution np.zeros(1000) (a solution of zeros).

Quickstart

The package is very simple to use. There is a class Benchmark with two functions:

Give information for each function: their optimum, their dimensionality, the domain search, and the expected threshold to achieve the optima.
Give a fitness function to evaluate solutions. It expect that these solutions are numpy arrays (vectors) but it can also work with normal arrays.

These two functionalities are done with two methods in Benchmark class:

get_num_functions()

Return the number of functions in the benchmarks (15)
get_info(function_id)

Return an array with the following information, where /function_id/ is the identifier of the function, a int value between 1 and 15.
- lower, upper
  
  lower and upper boundaries of the domain search.
- best
  
  Optimum to achieve, it is always zero, thus it can be ignored.
- threshold
  
  Threshold to obtain, it is always zero, thus it can also be ignored.
- dimension
  
  Dimension for the function, it is always 1000.
It can be noticed that several data are the same for all functions. It is made for maintaining the same interface to other cec20xx competitions.
get_function(function_id)

function_id is the same parameter than in get_info, an integer value between 1 and 15.

It returns the fitness function to evaluate the solutions.

Examples of use

Obtain information about one function

>>> from cec2013lsgo.cec2013 import Benchmark
>>> bench = Benchmark()
>>> bench.get_info(1)
{'best': 0.0,
 'dimension': 1000,
 'lower': -100.0,
 'threshold': 0,
 'upper': 100.0}

Create random solution for the search

>>> from numpy.random import rand
>>> info = bench.get_info(1)
>>> dim = info['dimension']
>>> sol = info['lower']+rand(dim)*(info['upper']-info['lower'])

Evaluate a solution

>>> fun_fitness = bench.get_function(1)
>>> fun_fitness(sol)
464006824710.75995

Contact

Python package and C++ version: Daniel Molina @ Computer Science Deparment, University of Granada Please feel free to contact me at <dmolina@decsai.ugr.es> for any enquiries or suggestions.

Last Updated:

C++ version <2018-12-10>
Python wrapping <2018-01-08>