com.expleague.ml.loss.L2 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;

import com.expleague.commons.func.AdditiveStatistics;
import com.expleague.commons.func.Factory;
import com.expleague.commons.math.FuncC1;
import com.expleague.commons.math.MathTools;
import com.expleague.commons.math.vectors.Vec;
import com.expleague.commons.math.vectors.VecTools;
import com.expleague.ml.data.set.DataSet;
import com.expleague.ml.TargetFunc;
import org.jetbrains.annotations.NotNull;

/**
 * User: solar
 * Date: 21.12.2010
 * Time: 22:37:55
 */
public class L2 extends FuncC1.Stub implements StatBasedLoss, TargetFunc {
  public final Vec target;
  private final DataSet owner;

  public L2(final Vec target, final DataSet owner) {
    this.target = target;
    this.owner = owner;
  }

  @NotNull
  @Override
  public Vec gradient(final Vec x) {
    final Vec result = VecTools.copy(x);
    VecTools.scale(result, -1);
    VecTools.append(result, target);
    VecTools.scale(result, -2);
    return result;
  }

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

  @Override
  public double value(final Vec point) {
    final Vec temp = VecTools.copy(point);
    VecTools.scale(temp, -1);
    VecTools.append(temp, target);
    return Math.sqrt(VecTools.sum2(temp) / temp.dim());
  }

  @Override
  public Factory statsFactory() {
    return () -> new Stat(target);
  }

  @Override
  public Vec target() {
    return target;
  }

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

  @Override
  public double score(final Stat stats) {
    return stats.weight > MathTools.EPSILON ? (- stats.sum * stats.sum / stats.weight)/* + 5 * stats.weight2*/: 0/*+ 5 * stats.weight2*/;
  }

  @Override
  public double bestIncrement(final Stat 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 Stat implements AdditiveStatistics {
    public double sum;
    public double sum2;
    public double weight;
    public double weight2;

    private final Vec targets;

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

    @Override
    public Stat remove(final int index, final int times) {
      final double v = targets.get(index);
      sum -= times * v;
      sum2 -= times * v * v;
      weight -= times;
      weight2 -= times * times;
      return this;
    }


    public Stat remove(final int index, final int times, final double p) {
      final double v = targets.get(index);
      sum -= p * times * v;
      sum2 -= p * times * v * v;
      weight -= p * times;
      weight2 -= p * times * p * times;
      return this;
    }

    @Override
    public Stat remove(final AdditiveStatistics otheras) {
      final Stat other = (Stat) otheras;
      sum -= other.sum;
      sum2 -= other.sum2;
      weight -= other.weight;
      weight2 -= other.weight2;
      return this;
    }

    @Override
    public Stat append(final int index, final int times) {
      final double v = targets.get(index);
      sum += times * v;
      sum2 += times * v * v;
      weight += times;
      weight2 += times * times;
      return this;
    }

    public Stat append(final int index, final int times, final double p) {
      final double v = targets.get(index);
      sum += p * times * v;
      sum2 += p * times * v * v;
      weight += p * times;
      weight2 += p * times * times;
      return this;
    }

    @Override
    public Stat append(final AdditiveStatistics otheras) {
      final Stat other = (Stat) otheras;
      sum += other.sum;
      sum2 += other.sum2;
      weight += other.weight;
      weight2 += other.weight2;
      return this;
    }

    @Override
    public Stat append(int index, double w) {
      final double v = targets.get(index);
      sum += w * v;
      sum2 += w * v * v;
      weight += w;
      weight2 += w * w;

      return this;
    }

    @Override
    public Stat remove(int index, double w) {
      final double v = targets.get(index);
      sum -= w * v;
      sum2 -= w * v * v;
      weight -= w;
      weight2 -= w * w;

      return this;
    }
  }


}