package jsat.regression;
import java.util.*;
import jsat.classifiers.DataPoint;
import jsat.classifiers.bayesian.BestClassDistribution;
import jsat.distributions.multivariate.MultivariateKDE;
import jsat.linear.Vec;
import jsat.linear.VecPaired;
import jsat.parameters.Parameter.ParameterHolder;
import jsat.parameters.Parameterized;
/**
* The Nadaraya-Watson regressor uses the {@link MultivariateKDE Kernel Density Estimator } to perform regression on a data set.
* Nadaraya-Watson can also be expressed as a classifier, and equivalent results can be obtained by combining a KDE with {@link BestClassDistribution}.
*
* @author Edward Raff
*/
public class NadarayaWatson implements Regressor, Parameterized
{
private static final long serialVersionUID = 8632599345930394763L;
@ParameterHolder
private MultivariateKDE kde;
public NadarayaWatson(MultivariateKDE kde)
{
this.kde = kde;
}
@Override
public double regress(DataPoint data)
{
List extends VecPaired, Double>> nearBy = kde.getNearby(data.getNumericalValues());
if(nearBy.isEmpty())
return 0;///hmmm... what should be retruned in this case?
double weightSum = 0;
double sum = 0;
for(VecPaired, Double> v : nearBy)
{
double weight = v.getPair();
double regressionValue = ( (VecPaired) v.getVector().getVector()).getPair();
weightSum += weight;
sum += weight*regressionValue;
}
return sum / weightSum;
}
@Override
public void train(RegressionDataSet dataSet, boolean parallel)
{
List> vectors = collectVectors(dataSet);
kde.setUsingData(vectors, parallel);
}
private List> collectVectors(RegressionDataSet dataSet)
{
List> vectors = new ArrayList<>(dataSet.getSampleSize());
for(int i = 0; i < dataSet.getSampleSize(); i++)
vectors.add(new VecPaired<>(dataSet.getDataPoint(i).getNumericalValues(), dataSet.getTargetValue(i)));
return vectors;
}
@Override
public boolean supportsWeightedData()
{
return true;
}
@Override
public NadarayaWatson clone()
{
return new NadarayaWatson((MultivariateKDE)kde.clone());
}
}
Ads help maintain this website.