com.expleague.ml.cli.gridsearch.GridSearch 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.cli.gridsearch;

import com.expleague.commons.math.vectors.Vec;
import com.expleague.commons.math.Func;
import com.expleague.ml.TargetFunc;
import com.expleague.commons.math.Trans;
import com.expleague.ml.cli.builders.methods.MethodsBuilder;
import com.expleague.ml.data.tools.DataTools;
import com.expleague.ml.data.tools.Pool;
import com.expleague.ml.methods.VecOptimization;

import java.util.Arrays;

/**
 * User: qdeee
 * Date: 25.03.15
 */
public class GridSearch {
  private Pool learn;
  private Pool test;
  private MethodsBuilder methodsBuilder;
  private TargetFunc loss;
  private Func[] metrics;

  public GridSearch(
      final Pool learn,
      final Pool test,
      final TargetFunc loss,
      final Func[] metrics,
      final MethodsBuilder methodsBuilder
      ) {
    this.learn = learn;
    this.test = test;
    this.methodsBuilder = methodsBuilder;
    this.loss = loss;
    this.metrics = metrics;
  }

  public OptimumHolder[] search(
      final String commonScheme,
      final Object[][] parametersSpace
  ) {
    final OptimumHolder[] optimumHolders = new OptimumHolder[metrics.length];
    final ParametersGridEnumerator enumerator = new ParametersGridEnumerator<>(parametersSpace);
    while (enumerator.advance()) {
      final Object[] parameters = enumerator.getParameters();
      System.out.println(Arrays.toString(parameters));
      final String concreteScheme = String.format(commonScheme, parameters);
      final VecOptimization method = methodsBuilder.create(concreteScheme);
      final Trans result = method.fit(learn.vecData(), loss);
      final double targetValue = loss.value(DataTools.calcAll(result, learn.vecData()));
      final double[] metricsValues = new double[this.metrics.length];
      final Vec testEvaluation = DataTools.calcAll(result, test.vecData());
      for (int i = 0; i < metricsValues.length; i++) {
        metricsValues[i] = metrics[i].value(testEvaluation);
      }
      final OptimumHolder currentHolder = new OptimumHolder(parameters, metricsValues, targetValue);
      for (int i = 0; i < optimumHolders.length; i++) {
        if (currentHolder.isBetterThan(optimumHolders[i], i)) {
          optimumHolders[i] = currentHolder;
        }
      }
    }
    return optimumHolders;
  }

}