com.expleague.ml.data.tools.BinModelBuilder 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.data.tools;

import com.expleague.commons.util.ArrayTools;
import com.expleague.ml.BFGrid;
import com.expleague.ml.impl.BinaryFeatureImpl;
import com.expleague.ml.models.ObliviousTree;

/**
 * Created by noxoomo on 31/07/14.
 */
public class BinModelBuilder {
  private final FullMatrixClassifierInfo result;
  private final int[] rowStarts;

  public BinModelBuilder(final BFGrid grid) {
    rowStarts = new int[grid.rows()];
    int gridSize = 0;
    for (int i = 0; i < grid.rows(); ++i)
      gridSize += grid.row(i).size();
    result = new FullMatrixClassifierInfo(gridSize);
    int currentIndex = 0;
    for (int rowIndex = 0; rowIndex < grid.rows(); ++rowIndex) {
      rowStarts[rowIndex] = currentIndex;
      final BFGrid.Row row = grid.row(rowIndex);
      for (int bin = 0; bin < row.size(); ++bin) {
        final BFGrid.Feature bf = row.bf(bin);
        result.binFeatures[currentIndex] = new BinaryFeatureStat(rowIndex, bf.condition());
        ++currentIndex;
      }
    }
  }


  public void append(final ObliviousTree tree, final Double weight) {
    final int treeDepth = tree.features().size();
    final int[] conditions = new int[treeDepth];
    final double[][] values;

    final BinaryFeatureImpl[] features = (BinaryFeatureImpl[]) tree.features().toArray();
    for (int i = 0; i < treeDepth; ++i) {
      final int depth = treeDepth - i - 1;
      conditions[i] = rowStarts[features[depth].findex] + features[depth].binNo;
    }
    values = new double[1][1 << conditions.length];
    final double[] leaveValues = tree.values();

    System.arraycopy(leaveValues, 0, values[0], 0, (1 << conditions.length));
    ArrayTools.mul(values[0], 0, values.length, weight);
    result.trees.add(new TreeStat(conditions, values));
  }

  public FullMatrixClassifierInfo build() {
    return result;
  }
}