org.dllearner.algorithms.el
Class ELLearningAlgorithmDisjunctive

java.lang.Object
  org.dllearner.core.AbstractComponent
      org.dllearner.core.AbstractCELA
          org.dllearner.algorithms.el.ELLearningAlgorithmDisjunctive

All Implemented Interfaces:

ClassExpressionLearningAlgorithm, Component, LearningAlgorithm, StoppableLearningAlgorithm

public class ELLearningAlgorithmDisjunctive
extends AbstractCELA

A learning algorithm for EL, which will based on an ideal refinement operator. The algorithm learns disjunctions of EL trees as follows: - given pos. and neg. examples, noise in %, min coverage per tree x % - it searches for an EL tree, which covers at least x % of all positive examples and at most (coverage_on_positives * noise) negative examples - the covered examples are removed from the pos. and neg. examples - termination: all(?) positive examples covered ev. besser: feste Suchzeiten pro Baum => es wird dann jeweils der beste Baum gewählt => Terminierung, wenn alles gecovered ist oder kein Baum mit ausreichender Qualität in dem Zeitfenster gefunden wird In contrast to many other algorithms, only one solution is returned. Additionally, the algorithm is not really an anytime algorithm, since the solution is constructed stepwise as a set of trees. Parameter optimisation: - runtime per tree: 10 seconds - tradeoff pos/neg: 1.0 1.2 1.4 1.6. 1.8 2.0 - min score: 0 -2.5 -5 -7.5 -10 - tests: 30 - runtime per test: 200 seconds => 2000 seconds cross val => 60000 seconds overall Next idea: - reduce tradeoff for each tree added (start with 2.0 and reduce by 0.1) - for the last tress it is not very important to cover less negatives - minimum is something between 0 and -1 (ensures that in the worst case as many positives as negatives are covered) - only high impact parameter is runtime (and maybe start tradeoff)

Author:

Jens Lehmann

Field Summary

Fields inherited from class org.dllearner.core.AbstractCELA

MAX_NR_OF_RESULTS

Constructor Summary

ELLearningAlgorithmDisjunctive(PosNegLP problem, AbstractReasonerComponent reasoner)

Method Summary

static Collection<ConfigOption<?>>	createConfigOptions()
Configurator	getConfigurator() For each component, a configurator class is generated in package org.dllearner.core.configurators using the script { org.dllearner.scripts.ConfigJavaGenerator}.
Description	getCurrentlyBestDescription()
EvaluatedDescription	getCurrentlyBestEvaluatedDescription() Returns the best descriptions obtained so far.
static String	getName()
SearchTreeNode	getStartNode()
void	init() Method to be called after the component has been configured.
boolean	isRunning() Returns whether the learning algorithm is running.
void	start() Starts the algorithm.
void	stop() Stops the algorithm gracefully.
static Collection<Class<? extends AbstractLearningProblem>>	supportedLearningProblems()

Methods inherited from class org.dllearner.core.AbstractCELA

changeLearningProblem, changeReasonerComponent, getCurrentlyBestDescriptions, getCurrentlyBestDescriptions, getCurrentlyBestDescriptions, getCurrentlyBestEvaluatedDescriptions, getCurrentlyBestEvaluatedDescriptions, getCurrentlyBestEvaluatedDescriptions, getCurrentlyBestEvaluatedDescriptions, getCurrentlyBestMostGeneralEvaluatedDescriptions

Methods inherited from class java.lang.Object

equals, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

ELLearningAlgorithmDisjunctive

public ELLearningAlgorithmDisjunctive(PosNegLP problem,
                                      AbstractReasonerComponent reasoner)

Method Detail

getName

public static String getName()

supportedLearningProblems

public static Collection<Class<? extends AbstractLearningProblem>> supportedLearningProblems()

createConfigOptions

public static Collection<ConfigOption<?>> createConfigOptions()

getConfigurator

public Configurator getConfigurator()

Description copied from class: AbstractComponent

For each component, a configurator class is generated in package org.dllearner.core.configurators using the script { org.dllearner.scripts.ConfigJavaGenerator}. The configurator provides set and get methods for the configuration options of a component.

Specified by:

getConfigurator in class AbstractComponent

Returns:

An object allowing to configure this component.

init

public void init()
          throws ComponentInitException

Description copied from interface: Component

Method to be called after the component has been configured. Implementation of components can overwrite this method to perform setup and initialisation tasks for this component.

Throws:

ComponentInitException - This exception is thrown if any exceptions occur within the initialisation process of this component. As component developer, you are encouraged to rethrow occuring exception as ComponentInitException and giving an error message as well as the actualy exception by using the constructor ComponentInitException.ComponentInitException(String, Throwable).

start

public void start()

Description copied from interface: LearningAlgorithm

Starts the algorithm. It runs until paused, stopped, or a termination criterion has been reached.

stop

public void stop()

Description copied from interface: StoppableLearningAlgorithm

Stops the algorithm gracefully. A stopped algorithm cannot be resumed anymore. Use this method for cleanup and freeing memory.

isRunning

public boolean isRunning()

Description copied from interface: StoppableLearningAlgorithm

Returns whether the learning algorithm is running. Implementation should use a boolean status variable in their implementations of the start and resume methods.

Returns:

True if the algorithm is running, false otherwise.

getCurrentlyBestDescription

public Description getCurrentlyBestDescription()

Specified by:

getCurrentlyBestDescription in class AbstractCELA

Returns:

The best class description found by the learning algorithm so far.

See Also:

AbstractCELA.getCurrentlyBestEvaluatedDescription()

getCurrentlyBestEvaluatedDescription

public EvaluatedDescription getCurrentlyBestEvaluatedDescription()

Description copied from class: AbstractCELA

Returns the best descriptions obtained so far.

Specified by:

getCurrentlyBestEvaluatedDescription in class AbstractCELA

Returns:

Best class description found so far.

getStartNode

public SearchTreeNode getStartNode()

Returns:

the startNode