Home

µGP wiki

Notez Bien: after the Great Hosted Apps Retirement of 2014, the µGP collective started porting the documentation to the new wiki system. This process is not finished, yet.

µGP is a computational approach for autonomously pursuing a goal defined by the user. It is a versatile optimizer, and it was demonstrated able to outperform both human experts and conventional heuristics in finding the optimal solution of hard problems.

µGP has been devised in 2002 in the CAD Group at Politecnico di Torino, and subsequently developed thanks to the effort of several people. It is distributed under the GNU Public License; Sourceforge is kindly hosting the project on http://ugp3.sf.net/.

Version 3.1 of the toolkit has been thoroughly described in a book in 2010, but the µGP collective never stopped working. Significant changes and improvements have been implemented ever since. The goal of this wiki is to provide instructions and updated documentation for the users. It aims to be complete and comprehensive, rather than concise and clean.

Project Admins:

• Alberto Tonda
• Giovanni Squillero
• Ernesto Sanchez

---

Before you start

• What exactly is µGP?
• What is an evolutionary algorithm?
• Success stories
• Legal notice
• Version history
• Donations
• Credits

Using the toolkit

• Compiling and installing

Defining the goal

• Standard optimization TODO
• Multi-objective optimization
• Group optimization TODO
• Writing a compliant evaluator
• Writing a compliant evaluator with Lua

Tweaking the optimization process

• Main Settings
• Population's Settings and Genetic Operators
• Population's Constraints and Macros
• ugp3's command line options
• Islands and migration
• Output files
• Special files

Tutorials

• A first experiment: OneMax
• More fun with OneMax
• Even more fun with OneMax
• Generating real assembly programs: yet another OneMax
• An example of writing constraints: assembly programs
• Multi-objective evolution: the Polar testbench
• Group Evolution: the Lamps testbench TODO
• Graphical User Interface: Constraints Editor

Student projects and external contributions

• Genetic Improvement, by Francesco Marino

Hacking the toolkit (i.e., welcome aboard!)

• A brief summary of µGP's structure
• The Evolutionary Core
• Structure of individuals
• Operators selection

• Evaluator

• Adding new parameter types

• Adding new evolutionary operators TO BE COMPLETED
• Adding new populations TODO
• Adding new individual selectors
• Adding new operator selectors TODO
• Adding new front-ends TODO
• Benchmarking and spotting performance regressions

Additional info

• Premature convergence
• Wrong fitness function
• Diversity preservation TO BE COMPLETED
• Books