com.expleague.ml.methods.multiclass.spoc.impl.CodingMatrixLearningGreedy 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.methods.multiclass.spoc.impl;

import com.expleague.commons.math.vectors.MxTools;
import com.expleague.commons.math.vectors.Mx;
import com.expleague.commons.math.vectors.VecTools;
import com.expleague.commons.math.vectors.impl.mx.VecBasedMx;
import com.expleague.commons.util.Combinatorics;
import com.expleague.ml.methods.multiclass.spoc.AbstractCodingMatrixLearning;
import com.expleague.ml.methods.multiclass.spoc.CMLHelper;

/**
 * User: qdeee
 * Date: 07.05.14
 */
public class CodingMatrixLearningGreedy extends AbstractCodingMatrixLearning {
  public CodingMatrixLearningGreedy(final int k, final int l, final double lambdaC, final double lambdaR, final double lambda1) {
    super(k, l, lambdaC, lambdaR, lambda1);
  }

  protected double calcLoss(final Mx B, final Mx S) {
    double result = 0;
    final Mx mult = MxTools.multiply(B, MxTools.transpose(B));
    result -= MxTools.trace(MxTools.multiply(mult, S));
    result += lambdaR * VecTools.sum(mult);
    result += lambdaC * VecTools.sum2(B);
    result += lambda1 * VecTools.l1(B);
    return result;
  }

  @Override
  public Mx findMatrixB(final Mx S) {
    final Mx mxB = new VecBasedMx(k, l);
    for (int j = 0; j < l; j++) {
      final Combinatorics.PartialPermutations permutationsGenerator = new Combinatorics.PartialPermutations(2, k);
      int[] bestPerm = null;
      double bestLoss = Double.MAX_VALUE;
      while (permutationsGenerator.hasNext()) {
        final int[] perm = permutationsGenerator.next();
        for (int i = 0; i < k; i++) {
          mxB.set(i, j, 2 * perm[i] - 1);
        }
        final Mx sub = mxB.sub(0, 0, k, j + 1);
        if (CMLHelper.checkConstraints(sub) && CMLHelper.checkColumnsIndependence(sub)) {
          final double loss = calcLoss(sub, S);
          if (loss < bestLoss) {
            bestLoss = loss;
            bestPerm = perm;
          }
        }
      }
      if (bestPerm != null) {
        for (int i = 0; i < k; i++) {
          mxB.set(i, j, 2 * bestPerm[i] - 1);
        }
      }
      else
        throw new IllegalStateException("Not found appreciate column #" + j);
//      System.out.println("Column " + j + " is over!");
    }
    return mxB;
  }
}