com.expleague.ml.clustering.impl.NearestNeighborDRAlgorithm Maven / Gradle / Ivy

Go to download

Show more of this group Show more artifacts with this name
Show all versions of jmll

Various ML methods implemented by myself and my students

There is a newer version: 1.4.9

package com.expleague.ml.clustering.impl;

import com.expleague.commons.math.metrics.Metric;
import com.expleague.commons.math.vectors.Vec;
import com.expleague.ml.clustering.ClusterizationAlgorithm;
import org.jetbrains.annotations.NotNull;

import java.util.Arrays;
import java.util.Collection;
import java.util.Collections;
import java.util.HashSet;
import java.util.function.Function;

/**
 * User: terry
 * Date: 16.01.2010
 */
public class NearestNeighborDRAlgorithm implements ClusterizationAlgorithm {
  private final Metric metric;
  private final double acceptanceDistance;
  private final double distanceRatio;

  public NearestNeighborDRAlgorithm(final Metric metric, final double acceptanceDistance, final double distanceRatio) {
    this.metric = metric;
    this.acceptanceDistance = acceptanceDistance;
    this.distanceRatio = distanceRatio;
  }

  @NotNull
  @Override
  public Collection> cluster(final Collection dataSet, final Function data2DVector) {
    final Collection> clusters = new HashSet<>();
    for (final X data : dataSet) {
      final Vec dataVector = data2DVector.apply(data);
      Collection nearestCluster = null;
      double nearestDistance = Double.MAX_VALUE;
      double nearest2Distance = Double.MAX_VALUE;
      for (final Collection cluster : clusters) {
        double minDistance = Double.MAX_VALUE;
        for (final X element : cluster) {
          final double candidateDistance = metric.distance(data2DVector.apply(element), dataVector);
          minDistance = Math.min(minDistance, candidateDistance);
        }

        if (minDistance < nearestDistance) {
          nearestDistance = minDistance;
          nearestCluster = cluster;
        }
        else if (minDistance < nearest2Distance) {
          nearest2Distance = minDistance;
        }
      }

      final boolean good =
        (nearestDistance < acceptanceDistance && (nearest2Distance == Double.MAX_VALUE || nearestDistance / nearest2Distance < distanceRatio));
      if (nearestCluster == null || !good) {
        clusters.add(new HashSet<>(Collections.singletonList(data)));
      } else {
        nearestCluster.add(data);
      }
    }
    return clusters;
  }
}