/*
* RapidMiner
*
* Copyright (C) 2001-2008 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.transformation;
import java.util.Iterator;
import com.rapidminer.example.Attribute;
import com.rapidminer.example.AttributeWeights;
import com.rapidminer.example.Example;
import com.rapidminer.example.ExampleSet;
import com.rapidminer.example.table.AttributeFactory;
import com.rapidminer.operator.AbstractModel;
import com.rapidminer.operator.OperatorException;
import com.rapidminer.operator.UserError;
import com.rapidminer.tools.Ontology;
import com.rapidminer.tools.Tools;
import Jama.Matrix;
/**
* This is the transformation model of the <code>FastICA</code>. The number
* of independent components is initially specified by the <code>FastICA</code>.
* Additionally you can specify parameters in the <code>ModelApplier</code>.
* <ul>
* <li><b>keep_attributes</b> <i>true|false</i> If true, the original
* features are not removed.
* </ul>
*
* @author Daniel Hakenjos, Ingo Mierswa
* @version $Id: FastICAModel.java,v 1.7 2008/05/28 10:52:03 ingomierswa Exp $
* @see FastICA
*/
public class FastICAModel extends AbstractModel implements ComponentWeightsCreatable {
private static final long serialVersionUID = -6380202686511014212L;
private double[] means;
private boolean rowNorm;
private int numberOfComponents;
private Matrix K, W, A;
private String[] attributeNames;
// -------------------------------------------------------------------------------------
private int numberOfSamples, numberOfAttributes;
private boolean keepAttributes = false;
public FastICAModel(ExampleSet exampleSet, int numberOfComponents, double[] means, boolean rowNorm, Matrix K, Matrix W, Matrix A) {
super(exampleSet);
this.attributeNames = com.rapidminer.example.Tools.getRegularAttributeNames(exampleSet);
this.numberOfComponents = numberOfComponents;
this.means = means;
this.rowNorm = rowNorm;
this.K = K;
this.W = W;
this.A = A;
}
public ExampleSet apply(ExampleSet testSet) throws OperatorException {
testSet.recalculateAllAttributeStatistics();
numberOfSamples = testSet.size();
numberOfAttributes = testSet.getAttributes().size();
if (numberOfAttributes != means.length) {
throw new UserError(null, 133, means.length, numberOfAttributes);
}
// all attributes numerical
for (Attribute attribute : testSet.getAttributes()) {
if (!attribute.isNumerical()) {
throw new UserError(null, 104, new Object[] { "FastICA", attribute.getName() });
}
}
// get the centered data
double[][] data = new double[numberOfSamples][numberOfAttributes];
Iterator<Example> reader = testSet.iterator();
Example example;
for (int sample = 0; sample < numberOfSamples; sample++) {
example = reader.next();
int d = 0;
for (Attribute attribute : testSet.getAttributes()) {
data[sample][d] = example.getValue(attribute) - means[d];
d++;
}
}
// row normalization
// Scaling is done by dividing the rows of the data
// by their root-mean-square. The root-mean-square for a row
// is obtained by computing the
// square-root of the sum-of-squares of the values in the
// row divided by the number of values minus one.
if (rowNorm) {
log("Scaling the data now.");
double rmsRow;
for (int row = 0; row < numberOfSamples; row++) {
// compute root mean square for the row
rmsRow = 0.0d;
for (int d = 0; d < numberOfAttributes; d++) {
rmsRow += data[row][d] * data[row][d];
}
rmsRow = Math.sqrt(rmsRow) / Math.max(1, numberOfAttributes - 1);
for (int d = 0; d < numberOfAttributes; d++) {
data[row][d] = data[row][d] / rmsRow;
}
}
}
Matrix X = new Matrix(data);
Matrix S = X.times(K).times(W);
if (!keepAttributes) {
testSet.getAttributes().clearRegular();
}
Attribute[] icAttributes = new Attribute[numberOfComponents];
for (int i = 0; i < numberOfComponents; i++) {
icAttributes[i] = AttributeFactory.createAttribute("ic_" + (i + 1), Ontology.REAL);
testSet.getExampleTable().addAttribute(icAttributes[i]);
testSet.getAttributes().addRegular(icAttributes[i]);
}
double[][] finaldata = S.getArray();
reader = testSet.iterator();
for (int sample = 0; sample < testSet.size(); sample++) {
example = reader.next();
for (int d = 0; d < numberOfComponents; d++) {
example.setValue(icAttributes[d], finaldata[sample][d]);
}
}
return testSet;
}
public void setNumberOfComponents(int number) {
this.numberOfComponents = number;
}
public void setParameter(String name, Object object) throws OperatorException {
if (name.equals("number_of_components")) {
String value = (String)object;
try {
this.setNumberOfComponents(Integer.parseInt(value));
} catch (NumberFormatException error) {
super.setParameter(name, value);
}
} else if (name.equals("keep_attributes")) {
String value = (String)object;
keepAttributes = false;
if (value.equals("true")) {
keepAttributes = true;
}
} else {
super.setParameter(name, object);
}
}
public AttributeWeights getWeightsOfComponent(int component) throws OperatorException {
if (component < 1) {
component = 1;
}
if (component > numberOfComponents) {
logWarning("Creating weights of component " + numberOfComponents + "!");
component = numberOfComponents;
}
AttributeWeights attweights = new AttributeWeights();
for (int i = 0; i < attributeNames.length; i++) {
attweights.setWeight(attributeNames[i], A.get(component - 1, i));
}
return attweights;
}
public String toString() {
StringBuffer result = new StringBuffer();
result.append("Number of Components: " + numberOfComponents + Tools.getLineSeparator());
result.append("Resulting attribute weights (from first component):" + Tools.getLineSeparator());
for (int i = 0; i < attributeNames.length; i++) {
result.append(attributeNames[i] + ": " + Tools.formatNumber(A.get(1, i)) + Tools.getLineSeparator());
}
return result.toString();
}
}