source: branches/ClassificationEnsembleVoting/HeuristicLab.Problems.DataAnalysis/3.4/Implementation/Classification/WeightCalculators/MedianThresholdCalculator.cs @ 7562

#region License Information
/* HeuristicLab
 * Copyright (C) 2002-2012 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
 *
 * This file is part of HeuristicLab.
 *
 * HeuristicLab is free software: you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation, either version 3 of the License, or
 * (at your option) any later version.
 *
 * HeuristicLab is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with HeuristicLab. If not, see <http://www.gnu.org/licenses/>.
 */
#endregion

using System.Collections;
using System.Collections.Generic;
using System.Linq;
using HeuristicLab.Common;
using HeuristicLab.Core;
using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;

namespace HeuristicLab.Problems.DataAnalysis {
  [StorableClass]
  [Item("MedianThresholdCalculator", "")]
  public class MedianThresholdCalculator : DiscriminantClassificationWeightCalculator {

    public MedianThresholdCalculator()
      : base() {
    }
    [StorableConstructor]
    protected MedianThresholdCalculator(bool deserializing) : base(deserializing) { }
    protected MedianThresholdCalculator(MedianThresholdCalculator original, Cloner cloner)
      : base(original, cloner) {
    }
    public override IDeepCloneable Clone(Cloner cloner) {
      return new MedianThresholdCalculator(this, cloner);
    }

    protected double[] threshold;
    protected double[] classValues;

    /// <summary>
    /// 
    /// </summary>
    /// <param name="discriminantSolutions"></param>
    /// <returns>median instead of weights, because it doesn't use any weights</returns>
    protected override IEnumerable<double> DiscriminantCalculateWeights(IEnumerable<IDiscriminantFunctionClassificationSolution> discriminantSolutions) {
      List<List<double>> estimatedValues = new List<List<double>>();
      List<List<double>> estimatedClassValues = new List<List<double>>();

      List<IClassificationProblemData> solutionProblemData = discriminantSolutions.Select(sol => sol.ProblemData).ToList();
      Dataset dataSet = solutionProblemData[0].Dataset;
      IEnumerable<int> rows = Enumerable.Range(0, dataSet.Rows);
      foreach (var solution in discriminantSolutions) {
        estimatedValues.Add(solution.Model.GetEstimatedValues(dataSet, rows).ToList());
        estimatedClassValues.Add(solution.Model.GetEstimatedValues(dataSet, rows).ToList());
      }

      List<double> median = new List<double>();
      List<double> targetValues = dataSet.GetDoubleValues(solutionProblemData[0].TargetVariable).ToList();
      IList<double> curTrainingpoints = new List<double>();
      int removed = 0;
      int count = targetValues.Count;
      for (int point = 0; point < count; point++) {
        curTrainingpoints.Clear();
        for (int solutionPos = 0; solutionPos < solutionProblemData.Count; solutionPos++) {
          if (PointInTraining(solutionProblemData[solutionPos], point)) {
            curTrainingpoints.Add(estimatedValues[solutionPos][point]);
          }
        }
        if (curTrainingpoints.Count > 0)
          median.Add(GetMedian(curTrainingpoints.OrderBy(p => p).ToList()));
        else {
          //remove not used points
          targetValues.RemoveAt(point - removed);
          removed++;
        }
      }
      AccuracyMaximizationThresholdCalculator.CalculateThresholds(solutionProblemData[0], median, targetValues, out classValues, out threshold);
      return Enumerable.Repeat<double>(1, discriminantSolutions.Count());
    }

    protected override double DiscriminantAggregateEstimatedClassValues(IDictionary<IClassificationSolution, double> estimatedClassValues, IDictionary<IClassificationSolution, double> estimatedValues) {
      IList<double> values = estimatedValues.Select(x => x.Value).ToList();
      if (values.Count <= 0)
        return double.NaN;
      double median = GetMedian(values);
      return GetClassValueToMedian(median);
    }
    private double GetClassValueToMedian(double median) {
      double classValue = classValues.First();
      for (int i = 0; i < classValues.Count(); i++) {
        if (median > threshold[i])
          classValue = classValues[i];
        else
          break;
      }
      return classValue;
    }

    protected override double GetDiscriminantConfidence(IEnumerable<IDiscriminantFunctionClassificationSolution> solutions, int index, double estimatedClassValue) {
      // only works with binary classification
      if (!classValues.Count().Equals(2))
        return double.NaN;
      Dataset dataset = solutions.First().ProblemData.Dataset;
      IList<double> values = solutions.Select(s => s.Model.GetEstimatedValues(dataset, Enumerable.Repeat(index, 1)).First()).ToList();
      if (values.Count <= 0)
        return double.NaN;
      double median = GetMedian(values);
      if (estimatedClassValue.Equals(classValues[0])) {
        if (median < estimatedClassValue)
          return 1;
        else if (median >= threshold[1])
          return 0;
        else {
          double distance = threshold[1] - classValues[0];
          return (1 / distance) * (median - classValues[0]);
        }
      } else if (estimatedClassValue.Equals(classValues[1])) {
        if (median > estimatedClassValue)
          return 1;
        else if (median <= threshold[1])
          return 0;
        else {
          double distance = classValues[1] - threshold[1];
          return (1 / distance) * (classValues[1] - median);
        }
      } else
        return double.NaN;
    }

    private double GetMedian(IList<double> estimatedValues) {
      int count = estimatedValues.Count;
      if (count % 2 == 0)
        return 0.5 * (estimatedValues[count / 2 - 1] + estimatedValues[count / 2]);
      else
        return estimatedValues[count / 2];
    }
  }
}