/*
* 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.learner.meta;
import java.util.Iterator;
import java.util.List;
import com.rapidminer.example.Attribute;
import com.rapidminer.example.Attributes;
import com.rapidminer.example.Example;
import com.rapidminer.example.ExampleSet;
import com.rapidminer.example.Statistics;
import com.rapidminer.example.table.AttributeFactory;
import com.rapidminer.operator.Model;
import com.rapidminer.operator.OperatorCapability;
import com.rapidminer.operator.OperatorDescription;
import com.rapidminer.operator.OperatorException;
import com.rapidminer.operator.ProcessSetupError.Severity;
import com.rapidminer.operator.ports.PortPairExtender;
import com.rapidminer.operator.ports.metadata.AttributeMetaData;
import com.rapidminer.operator.ports.metadata.ExampleSetMetaData;
import com.rapidminer.operator.ports.metadata.MDReal;
import com.rapidminer.operator.ports.metadata.MetaData;
import com.rapidminer.operator.ports.metadata.SetRelation;
import com.rapidminer.operator.ports.metadata.SimpleMetaDataError;
import com.rapidminer.parameter.ParameterType;
import com.rapidminer.parameter.ParameterTypeBoolean;
import com.rapidminer.parameter.ParameterTypeCategory;
import com.rapidminer.parameter.conditions.EqualTypeCondition;
import com.rapidminer.tools.Ontology;
/**
* This meta learner applies a transformation on the label before the inner
* regression learner is applied.
* Please note, that the logistic link function will only work for probabilities or
* other labels with a range from 0 to 1 exclusive. If a value exceeds this range, it is set to the nearest
* possible element within this range.
*
* @author Stefan Rueping, Ingo Mierswa
* @author Stefan Rueping, Ingo Mierswa, Sebastian Land
*/
public class TransformedRegression extends AbstractMetaLearner {
/** The parameter name for "Type of transformation to use on the labels (log, exp, transform to mean 0 and variance 1, rank, or none)." */
public static final String PARAMETER_TRANSFORMATION_METHOD = "transformation_method";
/** The parameter name for "Scale transformed values to mean 0 and standard deviation 1?" */
public static final String PARAMETER_Z_SCALE = "z_scale";
/** The parameter name for "Interpolate prediction if predicted rank is not an integer?" */
public static final String PARAMETER_INTERPOLATE_RANK = "interpolate_rank";
private final PortPairExtender through = new PortPairExtender("through", getSubprocess(0).getInnerSinks(), getOutputPorts());
public TransformedRegression(OperatorDescription description) {
super(description);
through.start();
getTransformer().addRule(through.makePassThroughRule());
}
@Override
protected MetaData modifyExampleSetMetaData(ExampleSetMetaData unmodifiedMetaData) {
switch (unmodifiedMetaData.hasSpecial(Attributes.LABEL_NAME)) {
case NO:
getTrainingSetInputPort().addError(new SimpleMetaDataError(Severity.ERROR, getTrainingSetInputPort(), "special_missing", "label"));
return unmodifiedMetaData;
case UNKNOWN:
getTrainingSetInputPort().addError(new SimpleMetaDataError(Severity.WARNING, getTrainingSetInputPort(), "special_unknown", "label"));
return unmodifiedMetaData;
case YES:
AttributeMetaData labelMD = unmodifiedMetaData.getLabelMetaData();
unmodifiedMetaData.removeAttribute(labelMD);
AttributeMetaData transformedMD = labelMD.copy();
transformedMD.setName("transformation(" + labelMD.getName() + ")");
// TODO: Transform values instead of setting unkown
transformedMD.setValueSetRelation(SetRelation.UNKNOWN);
transformedMD.setMean(new MDReal());
unmodifiedMetaData.addAttribute(transformedMD);
return unmodifiedMetaData;
default:
return unmodifiedMetaData;
}
}
public Model learn(ExampleSet inputSet) throws OperatorException {
int method = getParameterAsInt(PARAMETER_TRANSFORMATION_METHOD);
double[] rank = null;
double mean = 0.0d;
double stddev = 1.0d;
Attribute label = inputSet.getAttributes().getLabel();
inputSet.recalculateAttributeStatistics(label);
ExampleSet eSet = (ExampleSet) inputSet.clone();
Attribute tempLabel = AttributeFactory.createAttribute("transformation(" + label.getName() + ")", Ontology.REAL);
eSet.getExampleTable().addAttribute(tempLabel);
// 1. Set new regression labels
Iterator<Example> r = eSet.iterator();
switch (method) {
case TransformedRegressionModel.LOG:
double offset = 1.0d - inputSet.getStatistics(label, Statistics.MINIMUM);
rank = new double[1];
rank[0] = offset;
while (r.hasNext()) {
Example e = r.next();
e.setValue(tempLabel, Math.log(offset + e.getValue(label)));
}
break;
case TransformedRegressionModel.LOG_LINK:
while (r.hasNext()) {
Example e = r.next();
double value = e.getValue(label);
if (value >= 1d)
value = 0.99999999999d;
if (value <= 0d)
value = 0.00000000001d;
e.setValue(tempLabel, Math.log(value / (1 - value)));
}
break;
case TransformedRegressionModel.EXP:
while (r.hasNext()) {
Example e = r.next();
e.setValue(tempLabel, Math.exp(e.getValue(label)));
}
break;
case TransformedRegressionModel.RANK:
double[] dummy = new double[eSet.size()];
int i = 0;
while (r.hasNext()) {
Example e = r.next();
dummy[i] = e.getValue(label);
i++;
}
java.util.Arrays.sort(dummy);
// remove double entries
i = 0;
for (int j = 0; j < dummy.length; j++) {
if (dummy[i] != dummy[j]) {
i++;
dummy[i] = dummy[j];
}
}
rank = new double[i + 1];
for (int j = 0; j < i + 1; j++) {
rank[j] = dummy[j];
}
r = eSet.iterator();
while (r.hasNext()) {
Example e = r.next();
e.setValue(tempLabel, java.util.Arrays.binarySearch(rank, e.getValue(label)));
}
// }
break;
case TransformedRegressionModel.NONE:
// just for convenience...
while (r.hasNext()) {
Example e = r.next();
e.setValue(tempLabel, e.getValue(label));
}
break;
default:
// cannot happen
break;
}
if (getParameterAsBoolean(PARAMETER_Z_SCALE)) {
eSet.recalculateAttributeStatistics(tempLabel);
mean = eSet.getStatistics(tempLabel, Statistics.AVERAGE);
stddev = eSet.getStatistics(tempLabel, Statistics.VARIANCE);
if (stddev <= 0.0d) {
// catch numerical errors
stddev = 1.0d;
};
r = eSet.iterator();
while (r.hasNext()) {
Example e = r.next();
e.setValue(tempLabel, (e.getValue(tempLabel) - mean) / stddev);
}
};
// 2. Apply learner
eSet.getAttributes().remove(label);
eSet.getAttributes().addRegular(tempLabel);
eSet.getAttributes().setLabel(tempLabel);
Model model = applyInnerLearner(eSet);
TransformedRegressionModel resultModel = new TransformedRegressionModel(inputSet, method, rank, model, getParameterAsBoolean(PARAMETER_Z_SCALE), mean, stddev, getParameterAsBoolean(PARAMETER_INTERPOLATE_RANK));
// passing inner data
through.passDataThrough();
return resultModel;
}
@Override
public List<ParameterType> getParameterTypes() {
List<ParameterType> types = super.getParameterTypes();
ParameterType type = new ParameterTypeCategory(PARAMETER_TRANSFORMATION_METHOD, "Type of transformation to use on the labels (log, exp, transform to mean 0 and variance 1, rank, or none).", TransformedRegressionModel.METHODS, TransformedRegressionModel.LOG);
type.setExpert(false);
types.add(type);
type = new ParameterTypeBoolean(PARAMETER_Z_SCALE, "If checked the values will be normalized to mean 0 and standard deviation 1.", false);
type.setExpert(true);
types.add(type);
type = new ParameterTypeBoolean(PARAMETER_INTERPOLATE_RANK, "If checked and predicted rank is not an integer, it will be interpolated.", true);
type.registerDependencyCondition(new EqualTypeCondition(this, PARAMETER_TRANSFORMATION_METHOD, TransformedRegressionModel.METHODS, false, TransformedRegressionModel.RANK));
type.setExpert(true);
types.add(type);
return types;
}
@Override
public boolean supportsCapability(OperatorCapability capability) {
switch (capability) {
case POLYNOMINAL_LABEL:
case BINOMINAL_LABEL:
case NO_LABEL:
case UPDATABLE:
case FORMULA_PROVIDER:
return false;
default:
return true;
}
}
}