ECJ

An Evolutionary Computation and Genetic Programming System by Sean Luke
Copyright 2000 by Sean Luke, All Rights Reserved
Version 3

ECJ is a research EC and GP system written in Java. It was designed to be highly flexible, with nearly all classes (and their settings) dynamically determined at runtime by a user-provided parameter file. Even so, the system was designed with an eye toward efficiency, and it is reasonably competitive with other GP systems in speed.

Features include:

• Platform-independent checkpointing and logging
• Hierarchical parameter files
• Multithreading
• Mersenne Twister Random Number Generators
• Reading populations from files
• True full strongly-typed genetic programming (the only system that I am aware of which provides this)
• Ephemeral Random Constants
• Automatically-Defined Functions and Automatically Defined Macros
• Customizable individuals
• Multiple tree forests
• Steady State and Generational evolution
• Multiobjective fitnesses
• Full suite of Koza-I and Koza-II-style runs
• Flexible breeding architecture
• Extensive set of breeding and selection operators
• Multiple subpopulations and species
• Five pre-done GP application problem domains (ant,regression,multiplexer,lawnmower,parity)
• Hooks for doing experiments with:
  • Inter-subpopulation breeding
  • Coevolution
  • Inter-process or inter-machine transfer of individuals (even across different platforms)
• Abstractions for doing a implementing a variety of EC forms, not just GP.

Before you use this software, you must first carefully read the license agreement. It's pretty lenient but it has a few importnant items in it which differ from the standard open-source license agreement. Be certain that you agree with them. Keep in mind that the Author is a pretty reasonable guy.

License Agreement This software is Copyright 2000 by Sean Luke All Rights Reserved By using the source code, binary code files, or related data included in this distribution, you agree to the following terms of usage for this software distribution. In this license the Author means Sean Luke, and the Agent means any person authorized by the Author to maintain the software at present. You can determine the Agent by referring to the software's present distribution site (which as of January 2000 is located at http://www.cs.umd.edu/users/seanl/ecj/ but is subject to change at any time without notice). The Author hereby grants you a world-wide, royalty-free, non-exclusive license, subject to third party intellectual property claims: to use, reproduce, modify, display, perform, sublicense and distribute all or any portion of the source code or binary form of this software or related data with or without mdifications, or as part of a larger work; and under patents now or hereafter owned or controlled by the Author, to make, have made, use and sell ("Utilize") all or any portion of the source code or binary form of this software or related data, but solely to the extent that any such patent is reasonably necessary to enable you to Utilize all or any portion of the source code or binary form of this software or related data, and not to any greater extent that may be necessary to Utilize further modifications or combinations. In return you agree to the following conditions: If you redistribute all or any portion of the source code of this software or related data, it must retain the above copyright notice and this license and disclaimer. If you redistribute all or any portion of this code in binary form, you must include the above copyright notice and this license and disclaimer in the documentation and/or other materials provided with the distribution. You must not use the Author's name to endorse or promote products derived from this software without the specific prior written permission of the Author. If you publish research results derived through the use of all or any portion of this software or related data, you must make a reasonable effort to inform the Agent and provide the Agent with a reference to your published results. If a reasonable effort fails, but the Author or Agent later requests this reference, you must provide it. You must also acknowledge the use of this software in print in the body of your publication, unless your publication is two pages in length or less in published form. If you distribute or publish publically-available modifications to all or any portion of this software or related data, or produce a product for sale or for license which includes all or any portion of this software, you must you must make a reasonable effort to inform the Agent and provide the Agent with a reference to your publically-available modifications or product. If a reasonable effort fails, but the Author or Agent later requests this reference, you must provide it. You must also acknowledge use of this software in a prominent, publically-accessible location in said modifications or product. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

Documentation

This distribution comes with several pieces of documentation:

• A high-level overview of the packages and classes in the system. Read this thoroughly first.
• A primer about parameters and parameter files.
• A lot of class documentation.
• A PDF file showing the ECJ class hierarchy.
• A PDF file detailing the Object state of ECJ when running Symbolic Regression.
• A table detailing the current development state of various ECJ packages.

Quick Examples

Some runs which result in the ideal individual being found (in most cases). First, cd to the ec directory. Then add the parent directory of the ec directory to your CLASSPATH. Then issue javac -O *.java */*.java */*/*.java */*/*/*.java

Now you're ready. Here are some problems to try.

Symbolic Regression

java ec.Evolve -file app/regression/noerc.params

Symbolic Regression, no ercs

java ec.Evolve -file app/regression/erc.params

Artificial Ant, Santa Fe Trail

java ec.Evolve -file app/ant/params

Artificial Ant, Los Altos Trail

java ec.Evolve -file app/ant/params -p eval.problem.file=app/ant/losaltos.trl

Boolean 3 Multiplexer (new fast form)

java ec.Evolve -file app/multiplexer/3.params

Boolean 6 Multiplexer (new fast form)

java ec.Evolve -file app/multiplexer/6.params

Boolean 11 Multiplexer (new fast form)

java ec.Evolve -file app/multiplexer/11.params

Boolean 3 Multiplexer (original form)

java ec.Evolve -file app/multiplexerslow/3.params

Boolean 6 Multiplexer (original form)

java ec.Evolve -file app/multiplexerslow/6.params

Boolean 11 Multiplexer (original form)

java ec.Evolve -file app/multiplexerslow/11.params

8x8 Lawnmower

java ec.Evolve -file app/lawnmower/noadf.params

8x8 Lawnmower with 2 ADFs

java ec.Evolve -file app/lawnmower/adf.params

Even 4-Parity

java ec.Evolve -file app/parity/params -p eval.problem.even=true

Odd 4-Parity

java ec.Evolve -file app/parity/params -p eval.problem.even=false

Even 3-Parity with 2 ADFs

java ec.Evolve -file app/parity/adf.params -p eval.problem.even=true

Odd 3-Parity with 2 ADFs

java ec.Evolve -file app/parity/adf.params -p eval.problem.even=false

After a run, various statistical information is written to the out.stat file. All examples use a population size of 1000, and tournament selection. All other parameters are Koza-I/II standard.

Contacting the author