/*
* RapidMiner
*
* Copyright (C) 2001-2011 by Rapid-I and the contributors
*
* Complete list of developers available at our web site:
*
* http://rapid-i.com
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU Affero General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU Affero General Public License for more details.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see http://www.gnu.org/licenses/.
*/
package com.rapidminer.operator.features.selection;
import java.util.LinkedList;
import java.util.List;
import com.rapidminer.example.Attribute;
import com.rapidminer.example.AttributeWeights;
import com.rapidminer.example.ExampleSet;
import com.rapidminer.operator.OperatorDescription;
import com.rapidminer.operator.OperatorException;
import com.rapidminer.operator.UserError;
import com.rapidminer.operator.features.Individual;
import com.rapidminer.operator.features.Population;
import com.rapidminer.operator.features.PopulationOperator;
import com.rapidminer.operator.ports.InputPort;
import com.rapidminer.operator.ports.metadata.CompatibilityLevel;
import com.rapidminer.operator.ports.metadata.ExampleSetMetaData;
import com.rapidminer.operator.ports.metadata.MetaData;
import com.rapidminer.operator.ports.metadata.SimplePrecondition;
import com.rapidminer.parameter.ParameterType;
import com.rapidminer.parameter.ParameterTypeBoolean;
import com.rapidminer.parameter.ParameterTypeCategory;
import com.rapidminer.parameter.ParameterTypeDouble;
import com.rapidminer.parameter.ParameterTypeInt;
import com.rapidminer.parameter.UndefinedParameterError;
import com.rapidminer.parameter.conditions.BooleanParameterCondition;
/**
* A genetic algorithm for feature selection (mutation=switch features on and off, crossover=interchange used features).
* Selection is done by roulette wheel. Genetic algorithms are general purpose optimization / search algorithms that are
* suitable in case of no or little problem knowledge. <br/>
*
* A genetic algorithm works as follows
* <ol>
* <li>Generate an initial population consisting of <code>population_size</code> individuals. Each attribute is switched
* on with probability <code>p_initialize</code></li>
* <li>For all individuals in the population
* <ul>
* <li>Perform mutation, i.e. set used attributes to unused with probability <code>p_mutation</code> and vice versa.</li>
* <li>Choose two individuals from the population and perform crossover with probability <code>p_crossover</code>. The
* type of crossover can be selected by <code>crossover_type</code>.</li>
* </ul>
* </li>
* <li>Perform selection, map all individuals to sections on a roulette wheel whose size is proportional to the
* individual's fitness and draw <code>population_size</code> individuals at random according to their probability.</li>
* <li>As long as the fitness improves, go to 2</li>
* </ol>
*
* If the example set contains value series attributes with blocknumbers, the whole block will be switched on and off.
*
* @author Ingo Mierswa, Simon Fischer
*/
public class GeneticAlgorithm extends AbstractGeneticAlgorithm {
/** The parameter name for "Initial probability for an attribute to be switched on." */
public static final String PARAMETER_P_INITIALIZE = "p_initialize";
/** The parameter name for "Probability for an attribute to be changed (-1: 1 / numberOfAtt)." */
public static final String PARAMETER_P_MUTATION = "p_mutation";
/** The parameter name for "Probability for an individual to be selected for crossover." */
public static final String PARAMETER_P_CROSSOVER = "p_crossover";
/** The parameter name for "Type of the crossover." */
public static final String PARAMETER_CROSSOVER_TYPE = "crossover_type";
public static final String PARAMETER_MAX_NUMBER_OF_ATTRIBUTES = "max_number_of_attributes";
public static final String PARAMETER_MIN_NUMBER_OF_ATTRIBUTES = "min_number_of_attributes";
public static final String PARAMETER_EXACT_NUMBER_OF_ATTRIBUTES = "exact_number_of_attributes";
public static final String PARAMETER_INITIALIZE_WITH_INPUT_WEIGHTS = "initialize_with_input_weights";
public static final String PARAMETER_USE_EXACT_NUMBER = "use_exact_number_of_attributes";
public static final String PARAMETER_RESTRICT_NUMBER = "restrict_maximum";
private InputPort attributeWeightsInput = getInputPorts().createPort("attribute weights in");
public GeneticAlgorithm(OperatorDescription description) {
super(description);
attributeWeightsInput.addPrecondition(new SimplePrecondition(attributeWeightsInput, new MetaData(AttributeWeights.class), false) {
@Override
public boolean isCompatible(MetaData input, CompatibilityLevel level) {
if (isParameterSet(PARAMETER_INITIALIZE_WITH_INPUT_WEIGHTS)) {
if (input.getObjectClass().equals(AttributeWeights.class))
return getParameterAsBoolean(PARAMETER_INITIALIZE_WITH_INPUT_WEIGHTS);
}
return false;
}
@Override
protected boolean isMandatory() {
if (isParameterSet(PARAMETER_INITIALIZE_WITH_INPUT_WEIGHTS)) {
return getParameterAsBoolean(PARAMETER_INITIALIZE_WITH_INPUT_WEIGHTS);
}
return false;
}
});
}
@Override
protected ExampleSetMetaData modifyInnerOutputExampleSet(ExampleSetMetaData metaData) {
metaData.attributesAreSubset();
return metaData;
}
@Override
protected ExampleSetMetaData modifyOutputExampleSet(ExampleSetMetaData metaData) {
metaData.attributesAreSubset();
return metaData;
}
/**
* Sets up a population of given size and creates ExampleSets with randomly selected attributes (the probability to
* be switched on is controlled by pInitialize).
*/
@Override
public Population createInitialPopulation(ExampleSet es) throws OperatorException {
int minNumber = 1;
int maxNumber = 1;
int exactNumber = 0;
boolean useExactNumber = false;
boolean restrictMaxNumber = false;
int numberOfAttributes = es.getAttributes().size();
if (getParameterAsBoolean(PARAMETER_USE_EXACT_NUMBER)) {
useExactNumber = true;
exactNumber = getParameterAsInt(PARAMETER_EXACT_NUMBER_OF_ATTRIBUTES);
logNote("Using exact number of features for feature selection (" + exactNumber + "), ignoring possibly defined range for the number of features and / or input attribute weights.");
if (exactNumber > numberOfAttributes) {
throw new UserError(this, 125, numberOfAttributes, exactNumber);
}
} else {
minNumber = getParameterAsInt(PARAMETER_MIN_NUMBER_OF_ATTRIBUTES);
if (getParameterAsBoolean(PARAMETER_RESTRICT_NUMBER)) {
maxNumber = getParameterAsInt(PARAMETER_MAX_NUMBER_OF_ATTRIBUTES);
restrictMaxNumber = true;
if (minNumber > maxNumber) {
throw new UserError(this, 210, PARAMETER_MAX_NUMBER_OF_ATTRIBUTES, PARAMETER_MIN_NUMBER_OF_ATTRIBUTES);
}
} else {
maxNumber = numberOfAttributes;
}
if (minNumber > numberOfAttributes) {
throw new UserError(this, 125, numberOfAttributes, minNumber);
}
}
Population initP = new Population();
double[] initialWeights = null;
if (attributeWeightsInput.isConnected()) {
AttributeWeights inputWeights = null;
inputWeights = attributeWeightsInput.getData();
initialWeights = new double[numberOfAttributes];
int index = 0;
for (Attribute attribute : es.getAttributes()) {
double weight = inputWeights.getWeight(attribute.getName());
if (Double.isNaN(weight)) {
weight = getRandom().nextDouble();
}
if (weight < 0.0d)
weight = 0.0d;
if (weight > 1.0d)
weight = 1.0d;
if ((weight > 0) && (weight < 1.0d)) {
if (weight < (1.0d - getParameterAsDouble(PARAMETER_P_INITIALIZE))) {
weight = 1.0d;
} else {
weight = 0.0d;
}
}
initialWeights[index++] = weight;
}
if (!useExactNumber) {
Individual individual = new Individual(initialWeights);
int numberOfFeatures = individual.getNumberOfUsedAttributes();
if (((!restrictMaxNumber) || (numberOfFeatures <= maxNumber)) && (numberOfFeatures >= minNumber)) {
initP.add(individual);
} else {
logWarning("Input attribute weights found but number of selected features do not match specified minimum and maximum number, ignoring input weights.");
initialWeights = null;
}
}
}
if (useExactNumber) { // exact feature number
while (initP.getNumberOfIndividuals() < getParameterAsInt(PARAMETER_POPULATION_SIZE)) {
double[] weights = new double[numberOfAttributes];
double prob = 1.0d / weights.length * exactNumber;
for (int i = 0; i < weights.length; i++) {
if (getRandom().nextDouble() < prob) {
weights[i] = 1.0d;
} else {
weights[i] = 0.0d;
}
}
// add result with exact number of features
Individual individual = new Individual(weights);
int numberOfFeatures = individual.getNumberOfUsedAttributes();
if (exactNumber == numberOfFeatures)
initP.add(individual);
}
} else { // within range
while (initP.getNumberOfIndividuals() < getParameterAsInt(PARAMETER_POPULATION_SIZE)) {
double[] weights = new double[numberOfAttributes];
if ((initialWeights != null) && (getRandom().nextBoolean())) {
for (int i = 0; i < weights.length; i++) {
if (getRandom().nextDouble() < 1.0d / initialWeights.length) {
if (initialWeights[i] > 0.0d)
weights[i] = 0.0d;
else
weights[i] = 1.0d;
} else {
weights[i] = initialWeights[i];
}
}
} else {
double p = getParameterAsDouble(PARAMETER_P_INITIALIZE);
for (int i = 0; i < weights.length; i++) {
if (getRandom().nextDouble() < (1.0d - p)) {
weights[i] = 1.0d;
}
}
}
Individual individual = new Individual(weights);
int numberOfSelectedAttributes = individual.getNumberOfUsedAttributes();
// increase number of selected if needed
while (numberOfSelectedAttributes < minNumber && numberOfAttributes >= minNumber) {
int random = getRandom().nextInt(numberOfAttributes);
if (weights[random] == 0) {
weights[random] = 1.0d;
numberOfSelectedAttributes++;
}
}
// deselect attributes if more than needed
if (restrictMaxNumber && maxNumber > minNumber) {
while (numberOfSelectedAttributes > maxNumber) {
// double probability to converge faster
double deSelectProb = ((numberOfSelectedAttributes - maxNumber)) / ((double) numberOfSelectedAttributes - 1);
for (int i = 0; i < numberOfAttributes; i++) {
if (weights[i] > 0 && getRandom().nextDouble() < deSelectProb) {
weights[i] = 0;
numberOfSelectedAttributes--;
}
}
}
}
if (((!restrictMaxNumber) || (numberOfSelectedAttributes <= maxNumber)) && (numberOfSelectedAttributes >= minNumber)) {
initP.add(individual);
}
}
}
return initP;
}
/**
* Returns an operator that performs the mutation. Can be overridden by subclasses.
*/
@Override
protected PopulationOperator getMutationPopulationOperator(ExampleSet eSet) throws UndefinedParameterError {
double pMutation = getParameterAsDouble(PARAMETER_P_MUTATION);
int minNumber = 1;
int maxNumber = 1;
int exactNumber = 0;
boolean useExactNumber = false;
boolean restrictMaxNumber = false;
if (getParameterAsBoolean(PARAMETER_USE_EXACT_NUMBER)) {
useExactNumber = true;
exactNumber = getParameterAsInt(PARAMETER_EXACT_NUMBER_OF_ATTRIBUTES);
logNote("Using exact number of features for feature selection (" + exactNumber + "), ignoring possibly defined range for the number of features and / or input attribute weights.");
} else {
minNumber = getParameterAsInt(PARAMETER_MIN_NUMBER_OF_ATTRIBUTES);
if (getParameterAsBoolean(PARAMETER_RESTRICT_NUMBER)) {
maxNumber = getParameterAsInt(PARAMETER_MAX_NUMBER_OF_ATTRIBUTES);
restrictMaxNumber = true;
} else
maxNumber = eSet.getAttributes().size();
}
return new SelectionMutation(pMutation, getRandom(), minNumber, restrictMaxNumber ? maxNumber : -1, useExactNumber ? exactNumber : -1);
}
/**
* Returns an operator that performs crossover. Can be overridden by subclasses.
*/
@Override
protected PopulationOperator getCrossoverPopulationOperator(ExampleSet eSet) throws UndefinedParameterError {
double pCrossover = getParameterAsDouble(PARAMETER_P_CROSSOVER);
int crossoverType = getParameterAsInt(PARAMETER_CROSSOVER_TYPE);
int minNumber = 1;
int maxNumber = 1;
int exactNumber = 0;
boolean useExactNumber = false;
boolean restrictMaxNumber = false;
if (getParameterAsBoolean(PARAMETER_USE_EXACT_NUMBER)) {
useExactNumber = true;
exactNumber = getParameterAsInt(PARAMETER_EXACT_NUMBER_OF_ATTRIBUTES);
logNote("Using exact number of features for feature selection (" + exactNumber + "), ignoring possibly defined range for the number of features and / or input attribute weights.");
} else {
minNumber = getParameterAsInt(PARAMETER_MIN_NUMBER_OF_ATTRIBUTES);
if (getParameterAsBoolean(PARAMETER_RESTRICT_NUMBER)) {
maxNumber = getParameterAsInt(PARAMETER_MAX_NUMBER_OF_ATTRIBUTES);
restrictMaxNumber = true;
} else
maxNumber = eSet.getAttributes().size();
}
return new SelectionCrossover(crossoverType, pCrossover, getRandom(), minNumber, restrictMaxNumber ? maxNumber : -1, useExactNumber ? exactNumber : -1);
}
@Override
public List<ParameterType> getParameterTypes() {
List<ParameterType> types = new LinkedList<ParameterType>();
ParameterType type = new ParameterTypeBoolean(PARAMETER_USE_EXACT_NUMBER, "Determines if only combinations containing this numbers of attributes should be tested.", false);
type.setExpert(false);
types.add(type);
type = new ParameterTypeBoolean(PARAMETER_RESTRICT_NUMBER, "If checked the maximal number of attributes might be restricted. Otherwise all combinations of all number of attributes are generated and tested.", false);
type.registerDependencyCondition(new BooleanParameterCondition(this, PARAMETER_USE_EXACT_NUMBER, false, false));
type.setExpert(false);
types.add(type);
type = new ParameterTypeInt(PARAMETER_MIN_NUMBER_OF_ATTRIBUTES, "Determines the minimum number of features used for the combinations.", 1, Integer.MAX_VALUE, 1);
type.registerDependencyCondition(new BooleanParameterCondition(this, PARAMETER_USE_EXACT_NUMBER, true, false));
type.setExpert(false);
types.add(type);
type = new ParameterTypeInt(PARAMETER_MAX_NUMBER_OF_ATTRIBUTES, "Determines the maximum number of features used for the combinations.", 1, Integer.MAX_VALUE, 1);
type.registerDependencyCondition(new BooleanParameterCondition(this, PARAMETER_RESTRICT_NUMBER, true, true));
type.registerDependencyCondition(new BooleanParameterCondition(this, PARAMETER_USE_EXACT_NUMBER, true, false));
type.setExpert(false);
types.add(type);
type = new ParameterTypeInt(PARAMETER_EXACT_NUMBER_OF_ATTRIBUTES, "Determines the exact number of features used for the combinations.", 1, Integer.MAX_VALUE, 1);
type.registerDependencyCondition(new BooleanParameterCondition(this, PARAMETER_USE_EXACT_NUMBER, true, true));
type.setExpert(false);
types.add(type);
type = (new ParameterTypeBoolean(PARAMETER_INITIALIZE_WITH_INPUT_WEIGHTS, "Indicates if this operator should look for attribute weights in the given input and use the input weights of all known attributes as starting point for the optimization.", false));
type.setDeprecated();
types.add(type);
types.addAll(super.getParameterTypes());
type = (new ParameterTypeDouble(PARAMETER_P_INITIALIZE, "Initial probability for an attribute to be switched on.", 0, 1, 0.5));
types.add(type);
type = new ParameterTypeDouble(PARAMETER_P_MUTATION, "Probability for an attribute to be changed (-1: 1 / numberOfAtt).", -1.0d, 1.0d, -1.0d);
types.add(type);
type = new ParameterTypeDouble(PARAMETER_P_CROSSOVER, "Probability for an individual to be selected for crossover.", 0.0d, 1.0d, 0.5d);
types.add(type);
type = (new ParameterTypeCategory(PARAMETER_CROSSOVER_TYPE, "Type of the crossover.", SelectionCrossover.CROSSOVER_TYPES, SelectionCrossover.UNIFORM));
type.setExpert(true);
types.add(type);
return types;
}
}