Jenetics is an Genetic Algorithm, respectively an Evolutionary Algorithm, library written in Java. It is designed with a clear separation of the several concepts of the algorithm, e.g. Gene
, Chromosome
, Genotype
, Phenotype
, Population
and fitness Function
. Jenetics allows you to minimize and maximize the given fitness function without tweaking it. In contrast to other GA implementations, the library uses the concept of an evolution stream (EvolutionStream
) for executing the evolution steps. Since the EvolutionStream
implements the Java Stream interface, it works smoothly with the rest of the Java Stream API.
The library is fully documented (javadoc) and comes with an user manual (pdf).
- JRE 8: Java runtime version 8 is needed for using the library, respectively for running the examples.
- JDK 8: The Java JDK 8 must be installed.
- Gradle 3.x: Gradle is used for building the library. (Gradle is download automatically, if you are using the Gradle Wrapper script
./gradlew
, located in the base directory, for building the library.)
- TestNG 6.11: Jenetics uses TestNG framework for unit tests.
- Apache Commons Math 3.6: Library is used for testing statistical collectors.
- Github: https://github.com/jenetics/jenetics/releases/download/v3.8.0/jenetics-3.8.0.zip
- Sourceforge: https://sourceforge.net/projects/jenetics/files/latest/download
- Maven:
io.jenetics:jenetics:3.8.0
on Maven Central
For building the Jenetics library from source, download the most recent, stable package version from Github (or Sourceforge) and extract it to some build directory.
$ unzip jenetics-<version>.zip -d <builddir>
<version>
denotes the actual Jenetics version and <builddir>
the actual build directory. Alternatively you can check out the master branch from Github.
$ git clone https://github.com/jenetics/jenetics.git <builddir>
Jenetics uses Gradle as build system and organizes the source into sub-projects (modules). Each sub-project is located in it’s own sub-directory:
- org.jenetics: This project contains the source code and tests for the Jenetics core-module.
- org.jenetics.example: This project contains example code for the core-module.
- org.jenetics.doc: Contains the code of the web-site and the manual.
For building the library change into the <builddir>
directory (or one of the module directory) and call one of the available tasks:
- compileJava: Compiles the Jenetics sources and copies the class files to the
<builddir>/<module-dir>/build/classes/main
directory. - jar: Compiles the sources and creates the JAR files. The artifacts are copied to the
<builddir>/<module-dir>/build/libs
directory. - javadoc: Generates the API documentation. The Javadoc is stored in the
<builddir>/<module-dir>/build/docs
directory - test: Compiles and executes the unit tests. The test results are printed onto the console and a test-report, created by TestNG, is written to
<builddir>/<module-dir>
directory. - clean: Deletes the
<builddir>/build/*
directories and removes all generated artifacts.
For building the library jar from the source call
$ cd <build-dir>
$ ./gradlew jar
IDE Integration
Gradle has tasks which creates the project file for Eclipse and IntelliJ IDEA. Call
$ ./gradlew [eclipse|idea]
for creating the project files for Eclipse or IntelliJ, respectively. Whereas the latest version of IntelliJ IDEA has decent native Gradle support.
The latest Eclipse version (4.4.2) has problems compiling some valid lambda expressions; e.g. the HelloWorld::eval
function in the example below. If you have such problems when trying to compile the library with Eclipse, you can fix this by adding an explicit cast to the method reference:
Engine
.builder((Function<Genotype<BitGene>, Integer>)HelloWorld::eval, gtf)
.build();
Or you are using IntelliJ instead.
The minimum evolution Engine setup needs a genotype factory, Factory<Genotype<?>>
, and a fitness Function
. The Genotype
implements the Factory
interface and can therefore be used as prototype for creating the initial Population
and for creating new random Genotypes
.
import org.jenetics.BitChromosome;
import org.jenetics.BitGene;
import org.jenetics.Genotype;
import org.jenetics.engine.Engine;
import org.jenetics.engine.EvolutionResult;
import org.jenetics.util.Factory;
public class HelloWorld {
// 2.) Definition of the fitness function.
private static Integer eval(Genotype<BitGene> gt) {
return gt.getChromosome()
.as(BitChromosome.class)
.bitCount();
}
public static void main(String[] args) {
// 1.) Define the genotype (factory) suitable
// for the problem.
Factory<Genotype<BitGene>> gtf =
Genotype.of(BitChromosome.of(10, 0.5));
// 3.) Create the execution environment.
Engine<BitGene, Integer> engine = Engine
.builder(HelloWorld::eval, gtf)
.build();
// 4.) Start the execution (evolution) and
// collect the result.
Genotype<BitGene> result = engine.stream()
.limit(100)
.collect(EvolutionResult.toBestGenotype());
System.out.println("Hello World:\n" + result);
}
}
In contrast to other GA implementations, the library uses the concept of an evolution stream (EvolutionStream
) for executing the evolution steps. Since the EvolutionStream
implements the Java Stream interface, it works smoothly with the rest of the Java streaming API. Now let's have a closer look at listing above and discuss this simple program step by step:
-
The probably most challenging part, when setting up a new evolution
Engine
, is to transform the problem domain into a appropriateGenotype
(factory) representation. In our example we want to count the number of ones of aBitChromosome
. Since we are counting only the ones of one chromosome, we are adding only oneBitChromosome
to ourGenotype
. In general, theGenotype
can be created with 1 to n chromosomes. -
Once this is done, the fitness function which should be maximized, can be defined. Utilizing the new language features introduced in Java 8, we simply write a private static method, which takes the genotype we defined and calculate it's fitness value. If we want to use the optimized bit-counting method,
bitCount()
, we have to cast theChromosome<BitGene>
class to the actual usedBitChromosome
class. Since we know for sure that we created the Genotype with aBitChromosome
, this can be done safely. A reference to the eval method is then used as fitness function and passed to theEngine.build
method. -
In the third step we are creating the evolution
Engine
, which is responsible for changing, respectively evolving, a given population. TheEngine
is highly configurable and takes parameters for controlling the evolutionary and the computational environment. For changing the evolutionary behavior, you can set different alterers and selectors. By changing the usedExecutor
service, you control the number of threads, the Engine is allowed to use. An newEngine
instance can only be created via its builder, which is created by calling theEngine.builder
method. -
In the last step, we can create a new
EvolutionStream
from ourEngine
. TheEvolutionStream
is the model or view of the evolutionary process. It serves as a »process handle« and also allows you, among other things, to control the termination of the evolution. In our example, we simply truncate the stream after 100 generations. If you don't limit the stream, theEvolutionStream
will not terminate and run forever. Since theEvolutionStream
extends thejava.util.stream.Stream
interface, it integrates smoothly with the rest of the Java Stream API. The final result, the bestGenotype
in our example, is then collected with one of the predefined collectors of theEvolutionResult
class.
This example tries to approximate a given image by semitransparent polygons. It comes with an Swing UI, where you can immediately start your own experiments. After compiling the sources with
$ ./gradlew compileTestJava
you can start the example by calling
$ ./jrun org.jenetics.example.image.EvolvingImages
The previous image shows the GUI after evolving the default image for about 4,000 generations. With the »Open« button it is possible to load other images for polygonization. The »Save« button allows to store polygonized images in PNG format to disk. At the button of the UI, you can change some of the GA parameters of the example.
- Bandaru, S. and Deb, K. Metaheuristic Techniques. Decision Sciences. CRC Press, pp. 693-750, Nov. 2016.
- Lyazid Toumi, Abdelouahab Moussaoui, and Ahmet Ugur. EMeD-Part: An Efficient Methodology for Horizontal Partitioning in Data Warehouses. Proceedings of the International Conference on Intelligent Information Processing, Security and Advanced Communication. Djallel Eddine Boubiche, Faouzi Hidoussi, and Homero Toral Cruz (Eds.). ACM, New York, NY, USA, Article 43, 7 pages, 2015.
- Andreas Holzinger (Editor), Igo Jurisica (Editor). Interactive Knowledge Discovery and Data Mining in Biomedical Informatics. Lecture Notes in Computer Science, Vol. 8401. Springer, 2014.
- Lyazid Toumi, Abdelouahab Moussaoui, Ahmet Ugur. Particle swarm optimization for bitmap join indexes selection problem in data warehouses. The Journal of Supercomputing, Volume 68, Issue 2, pp 672-708, May 2014.
- TANG Yi (Guangzhou Power Supply Bureau Limited, Guangzhou 511400, China) Study on Object-Oriented Reactive Compensation Allocation Optimization Algorithm for Distribution Networks, Oct. 2012.
- John M. Linebarger, Richard J. Detry, Robert J. Glass, Walter E. Beyeler, Arlo L. Ames, Patrick D. Finley, S. Louise Maffitt. Complex Adaptive Systems of Systems Engineering Environment Version 1.0. SAND REPORT, Feb. 2012.
- Introduction to Jenetics Library, by baeldung, April 11. 2017
- How to Solve Tough Problems Using Genetic Algorithms, by Tzofia Shiftan, April 6. 2017
- Genetic algorithms with Java, by William Antônio, January 10. 2017
- Jenetics 설치 및 예제, by JDM, May 8. 2015
- 유전 알고리즘 (Genetic Algorithms), by JDM, April 2. 2015
The library is licensed under the Apache License, Version 2.0.
Copyright 2007-2017 Franz Wilhelmstötter
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
- #157: Add
LineCrossover
class. - #158: Add
IntermediateCrossover
class. - #168: Remove dependency to
java.desktop
module. - #169: Describe how to configure (μ, λ) and (μ + λ) Evolution Strategies in manual.
- #177: Additional 'Seq' conversion functions:
- #182: Rename build script to default names. All build scripts are now named
build.gradle
. - #188: Additional
Engine.Builder
methods - #189:
TruncationSelector
is now able to globally limit best selected individuals. This is used for (μ, λ) and (μ + λ) Evolution Strategies. - #197: Improve CPU utilization for long running fitness functions. The original concurrent fitness function evaluation where assumed to be quite fast.