com.expleague.ml.loss.blockwise.BlockwiseL2 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.loss.blockwise;

import com.expleague.commons.math.MathTools;
import com.expleague.commons.math.vectors.Vec;
import com.expleague.commons.func.AdditiveStatistics;
import com.expleague.commons.func.Factory;
import com.expleague.ml.BlockwiseFuncC1;
import com.expleague.ml.TargetFunc;
import com.expleague.ml.data.set.DataSet;

/**
 * User: solar
 * Date: 21.12.2010
 * Time: 22:37:55
 */
public class BlockwiseL2 extends BlockwiseFuncC1.Stub implements BlockwiseStatBasedLoss, TargetFunc {
  public final Vec target;
  private final DataSet owner;

  public BlockwiseL2(final Vec target, final DataSet owner) {
    this.target = target;
    this.owner = owner;
  }

  @Override
  public int dim() {
    return target.dim();
  }

  @Override
  public void gradient(final Vec pointBlock, final Vec result, final int index) {
    final int blockSize = blockSize();
    for (int i = 0; i < blockSize; i++) {
      result.set(i, 2 * (pointBlock.get(i) - target.get(index * blockSize + i)));
    }
  }

  @Override
  public double value(final Vec pointBlock, final int index) {
    double result = 0.0;
    final int blockSize = blockSize();
    for (int i = 0; i < blockSize; i++) {
      final double val = pointBlock.get(i) - target.get(index * blockSize + i);
      result += val * val;
    }
    return result;
  }

  @Override
  public double transformResultValue(final double value) {
    return Math.sqrt(value / dim());
  }

  @Override
  public int blockSize() {
    return 1;
  }

  @Override
  public Factory statsFactory() {
    return new Factory() {
      @Override
      public MSEStats create() {
        return new MSEStats(target);
      }
    };
  }

  @Override
  public double value(final MSEStats stats) {
    return stats.sum2;
  }

  @Override
  public double score(final MSEStats stats) {
    return stats.weight > MathTools.EPSILON ? (stats.sum2 - stats.sum * stats.sum / stats.weight) : stats.sum2;
  }

  @Override
  public double bestIncrement(final MSEStats stats) {
    return stats.weight > MathTools.EPSILON ? stats.sum / stats.weight : 0;
  }

  public double get(final int i) {
    return target.get(i);
  }

  @Override
  public DataSet owner() {
    return owner;
  }

  public static class MSEStats implements AdditiveStatistics {
    public volatile double sum;
    public volatile double sum2;
    public volatile int weight;

    private final Vec targets;

    public MSEStats(final Vec target) {
      this.targets = target;
    }

    @Override
    public MSEStats remove(final int index, final int times) {
      final double v = targets.get(index);
      sum -= times * v;
      sum2 -= times * v * v;
      weight -= times;
      return this;
    }

    @Override
    public MSEStats remove(final AdditiveStatistics otheras) {
      final MSEStats other = (MSEStats) otheras;
      sum -= other.sum;
      sum2 -= other.sum2;
      weight -= other.weight;
      return this;
    }

    @Override
    public AdditiveStatistics append(int index, double weight) {
      return append(index, (int)weight);
    }

    @Override
    public AdditiveStatistics remove(int index, double weight) {
      return remove(index,(int)weight);
    }


    @Override
    public MSEStats append(final int index, final int times) {
      final double v = targets.get(index);
      final double v1 = times * v;
      sum += v1;
      sum2 += v1 * v;
      weight += times;
      return this;
    }


    @Override
    public MSEStats append(final AdditiveStatistics otheras) {
      final MSEStats other = (MSEStats) otheras;
      sum += other.sum;
      sum2 += other.sum2;
      weight += other.weight;
      return this;
    }

  }
}