Skip to content
Snippets Groups Projects

Optimisation Algorithm

Last edited by Bernhard J. Berger

Java

Creating a new optimisation is straightforward in EvoAl. First, you have to create a new class that implements OptimisationAlgorithm and annotate it properly:

package com.example;

import lombok.extern.slf4j.Slf4j;

import javax.enterprise.context.Dependent;

import de.evoal.core.api.optimisation.OptimisationAlgorithm;

@Slf4j
@Dependent
@Named("com.example.optimisation.new-algorithm")
public class NewAlgorithm implements OptimisationAlgorithm {
  @Override
  public OptimisationAlgorithm init(final Instance instance) {
    // do initialisation stuff

    return this;
  }

  @Override
  public void run() {
    // do the actual run
  }
}

DSL

For the DSL side, creating a corresponding .dl file is necessary to declare all possible configuration options. Therefore, you declare a new type that extends de.evoal.core.optimisation.algorithm. Thus, it inherits different attributes from the base type algorithm. By declaring additional attributes, you can add any configuration option you need.

import de.evoal.core.math;
import de.evoal.core.optimisation;

module com.example.optimisation {
  type 'new-algorithm' extends algorithm {
    option : real := 3 * PI;
  }
}

Looking up attributes

The example introduces a new attribute named option with a corresponding default value. There are two options to get the values from an .ol file. First, you can use CDI and second, programmatically:

public class NewAlgorithm implements OptimisationAlgorithm {
  // Option 1) getting value using CDI
  @Inject
  @ConfigurationValue(entry = CoreBlackboardEntries.OPTIMISATION_CONFIGURATION, access = "algorithm.option")
  private double option;

  // Option 2) Programmatic
  @Inject
  private LanguageHelper helper;

  @Override
  public OptimisationAlgorithm init(final Instance instance) {
    // 2) Programmatic lookup
    double option = helper.lookup(instance, "option");

    return this;
  }

  // actual run is missing
}

Now, everything is set up for using the newly implemented optimisation algorithm in a .dl file.

Using pre-implemented options

There are different options for your optimisation algorithm that you can use from the problem definition or the base type algorithm.

algorithm.initialisation

Helps you to to initialise your population for the optimisation algorithm. You can inject an InitialCandidatesProvider that you can ask for new candidates. These candidates are generated according to the configuration.

  @Inject @Named("initial")
  private InitialCandidatesProvider provider;

  void do() {
    provider.create()
            .stream()
            // do stuff
            ;
  }

Please be aware that creating a new provider whenever you need an instance may lead to a predictable order of individuals, for instance, when the individuals are read from a file.

algorithm.optimisation-function

Configures the optimisation (aka fitness) function and allows you to inject it via CDI:

  @Inject @Named("optimisation-function")
  private OptimisationFunction fitness;

  void do() {
    Properties individual = // get an individual, e.g. from the initial stream
    double [] values = fitness.evaluate(individual);
  }

algorithm.comparator

Configures the comparator of optimisation values and allows you to inject it via CDI. The comparator may implement different strategies of comparing individuals.

@Inject @Named("comparator")
private OptimisationValueComparator comparator;

void do() {
    Properties individual = // get an individual, e.g. from the initial stream
    double [] values = fitness.evaluate(individual);
    OptimisationValue value = comparator.apply(values);  
}

problem.optimisation-space and problem.search-space

These settings describe the optimisation and the search space. You can get the corresponding PropertiesSpecification via CDI:

@Inject @Named("search-space-specification")
private PropertiesSpecification searchSpaceSpecification;

@Inject @Named("optimisation-space-specification")
private PropertiesSpecification optimisationSpaceSpecification;

problem.maximise

If the problem is a maximise or minimise problem:

@Inject @Named("problem.maximise")
private boolean maximise;

Comments

    Please register or sign in to add a comment.