source: branches/ClassificationModelComparison/HeuristicLab.Algorithms.DataAnalysis/3.4/Linear/OneR.cs @ 13086

#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;
using System.Collections.Generic;
using System.Linq;
using HeuristicLab.Common;
using HeuristicLab.Core;
using HeuristicLab.Data;
using HeuristicLab.Optimization;
using HeuristicLab.Parameters;
using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
using HeuristicLab.Problems.DataAnalysis;
using HeuristicLab.Random;

namespace HeuristicLab.Algorithms.DataAnalysis {
  /// <summary>
  /// 1R classification algorithm.
  /// </summary>
  [Item("OneR", "1R classification algorithm.")]
  [Creatable("Data Analysis")]
  [StorableClass]
  public sealed class OneR : FixedDataAnalysisAlgorithm<IClassificationProblem> {

    public IValueParameter<IntValue> MinBucketSizeParameter {
      get { return (IValueParameter<IntValue>)Parameters["MinBucketSize"]; }
    }
    public IValueParameter<IRandom> RandomParameter {
      get { return (IValueParameter<IRandom>)Parameters["Random"]; }
    }

    [StorableConstructor]
    private OneR(bool deserializing) : base(deserializing) { }
    private OneR(OneR original, Cloner cloner)
      : base(original, cloner) {
    }
    public OneR()
      : base() {
      Parameters.Add(new ValueParameter<IntValue>("MinBucketSize", "Minimum size of a bucket for numerical values. (Except for the rightmost bucket)", new IntValue(6)));
      Parameters.Add(new ValueParameter<IRandom>("Random", "Random number generator", new FastRandom()));
      Problem = new ClassificationProblem();
    }

    public override IDeepCloneable Clone(Cloner cloner) {
      return new OneR(this, cloner);
    }

    protected override void Run() {
      var startTime = DateTime.Now;
      var solution = CreateOneRSolution(Problem.ProblemData, MinBucketSizeParameter.Value.Value, RandomParameter.Value);
      Results.Add(new Result("OneR solution", "The 1R classifier.", solution));
      Results.Add(new Result("OneR Execution Time", "", new TimeSpanValue(DateTime.Now - startTime)));

      startTime = DateTime.Now;
      var solution3 = OneRTest.CreateOneRSolution(Problem.ProblemData, MinBucketSizeParameter.Value.Value);
      Results.Add(new Result("OneR Test2 solution", "The 1R classifier.", solution3));
      Results.Add(new Result("OneR Test2 Execution", "", new TimeSpanValue(DateTime.Now - startTime)));
    }

    public static IClassificationSolution CreateOneRSolution(IClassificationProblemData problemData, int minBucketSize, IRandom random) {
      var dataset = problemData.Dataset;
      var trainingIndices = problemData.TrainingIndices;
      int rowCount = trainingIndices.Count();
      string target = problemData.TargetVariable;
      var inputVariables = problemData.AllowedInputVariables.ToArray();
      var classValues = problemData.ClassValues.ToArray();
      double dominatingClass;

      string bestVariable = null;
      Dictionary<double, double> bestSplits = null;
      double missingValuesClass = double.NaN;
      int correctClassified = 0;

      for (int variable = 0; variable < inputVariables.Length; variable++) {
        var inputVariableValues = dataset.GetDoubleValues(inputVariables[variable], trainingIndices).ToArray();
        var classValuesInDataset = dataset.GetDoubleValues(target, trainingIndices).ToArray();

        int curCorrectClassified = 0;
        Dictionary<double, int> classCount = PrepareClassCountDictionary(classValues);
        Array.Sort(inputVariableValues, classValuesInDataset);
        double curSplit = Double.NegativeInfinity;
        Dictionary<double, double> splits = new Dictionary<double, double>();
        bool newBucket = true;
        bool done = false;
        int curRow = 0;

        if (curRow < inputVariableValues.Length && Double.IsNaN(inputVariableValues[curRow])) {
          while (curRow < inputVariableValues.Length && Double.IsNaN(inputVariableValues[curRow])) {
            classCount[classValuesInDataset[curRow]] += 1;
            curRow++;
          }
          if (ExistsDominatingClass(classCount, out dominatingClass)) {
            missingValuesClass = dominatingClass;
          } else {
            missingValuesClass = GetRandomMaxClass(classCount, random);
          }
          correctClassified += classCount[missingValuesClass];
          classCount = PrepareClassCountDictionary(classValues);
        }
        while (curRow < inputVariableValues.Length) {
          if (newBucket) {
            for (int i = 0; i < minBucketSize && curRow + i < inputVariableValues.Length; i++) {
              classCount[classValuesInDataset[curRow + i]] += 1;
            }
            curRow += minBucketSize;
            if (curRow >= inputVariableValues.Length) {
              break;
            }
            curSplit = inputVariableValues[curRow];
            curRow = SetCurRowToEndOfSplit(curRow, inputVariableValues, classValuesInDataset, classCount, curSplit);
            newBucket = false;
          }

          if (ExistsDominatingClass(classCount, out dominatingClass)) {
            while (curRow + 1 < classValuesInDataset.Length &&
              IsNextSplitStillDominatingClass(curRow, inputVariableValues, classValuesInDataset, curSplit, dominatingClass)) {
              curRow++;
              curSplit = inputVariableValues[curRow];
              classCount[classValuesInDataset[curRow]] += 1;
              curRow = SetCurRowToEndOfSplit(curRow, inputVariableValues, classValuesInDataset, classCount, curSplit);
            }

            curCorrectClassified += classCount[dominatingClass];
            done = curRow >= inputVariableValues.Length - 1;

            if (done) {
              curSplit = Double.PositiveInfinity;
              splits.Add(curSplit, dominatingClass);
              break;
            }

            curRow++;
            //intervals exclude end
            curSplit = inputVariableValues[curRow];
            splits.Add(curSplit, dominatingClass);

            //intervals include start
            curSplit = inputVariableValues[curRow];
            classCount = PrepareClassCountDictionary(classValues);
            newBucket = true;
          } else {
            curSplit = inputVariableValues[curRow];
            classCount[classValuesInDataset[curRow]] += 1;
            curRow = SetCurRowToEndOfSplit(curRow, inputVariableValues, classValuesInDataset, classCount, curSplit);
          }
        }

        if (!done) {
          curSplit = Double.PositiveInfinity;
          double randomClass = GetRandomMaxClass(classCount, random);
          splits.Add(curSplit, randomClass);

          curCorrectClassified += classCount[randomClass];
        }

        if (curCorrectClassified > correctClassified) {
          bestVariable = inputVariables[variable];
          bestSplits = splits;
        }
      }

      //merge intervals to simplify symbolic expression tree
      Dictionary<double, double> mergedSplits = MergeSplits(bestSplits);

      var model = new OneRClassificationModel(bestVariable, mergedSplits.Keys.ToArray(), mergedSplits.Values.ToArray(), missingValuesClass);
      var solution = new OneRClassificationSolution(model, (IClassificationProblemData)problemData.Clone());

      return solution;
    }

    private static double GetRandomMaxClass(Dictionary<double, int> classCount, IRandom random) {
      IList<double> possibleClasses = new List<double>();
      int max = 0;
      foreach (var item in classCount) {
        if (max < item.Value) {
          max = item.Value;
          possibleClasses = new List<double>();
          possibleClasses.Add(item.Key);
        } else if (max == item.Value) {
          possibleClasses.Add(item.Key);
        }
      }
      int classindex = random.Next(possibleClasses.Count);
      return possibleClasses[classindex];
    }

    private static bool IsNextSplitStillDominatingClass(int curRow, double[] inputVariableValues, double[] classValuesInDataset, double curSplit, double dominatingClass) {
      if (curRow >= classValuesInDataset.Length) {
        return false;
      }
      double nextSplit = inputVariableValues[curRow + 1];
      int i = 1;
      while (curRow + i < classValuesInDataset.Length
        && inputVariableValues[curRow + i] == nextSplit
        && classValuesInDataset[curRow + i] == dominatingClass) {
        i++;
      }
      if (curRow + i >= classValuesInDataset.Length) {
        return true;
      }
      if (inputVariableValues[curRow + i] != nextSplit) {
        return true;
      }
      // the next split would also contain values of a class which 
      // is not dominating (classValuesInDataset[curRow + i] != dominatingClass)
      return false;
    }

    // needed if variable contains the same value several times
    private static int SetCurRowToEndOfSplit(int curRow, double[] inputVariableValues, double[] classValuesInDataset, Dictionary<double, int> classCount, double curSplit) {
      while (curRow + 1 < inputVariableValues.Length && inputVariableValues[curRow + 1] == curSplit) {
        curRow++;
        classCount[classValuesInDataset[curRow]] += 1;
      }
      return curRow;
    }

    private static Dictionary<double, double> MergeSplits(Dictionary<double, double> bestSplits) {
      Dictionary<double, double> mergedSplits = new Dictionary<double, double>();
      double nextSplit, nextClass;
      nextSplit = nextClass = double.NaN;
      foreach (var item in bestSplits) {
        if (Double.IsNaN(nextSplit)) {
          nextSplit = item.Key;
          nextClass = item.Value;
        } else {
          if (nextClass == item.Value) {
            nextSplit = item.Key;
          } else {
            mergedSplits.Add(nextSplit, nextClass);
            nextSplit = item.Key;
            nextClass = item.Value;
          }
        }
      }
      mergedSplits.Add(nextSplit, nextClass);
      return mergedSplits;
    }

    private static bool ExistsDominatingClass(Dictionary<double, int> classCount, out double dominatingClass) {
      bool dominating = false;
      int count = 0;
      dominatingClass = double.NaN;
      foreach (var item in classCount) {
        if (item.Value > count) {
          dominatingClass = item.Key;
          count = item.Value;
          dominating = true;
        } else if (item.Value == count) {
          dominatingClass = double.NaN;
          dominating = false;
        }
      }
      return dominating;
    }

    private static Dictionary<double, int> PrepareClassCountDictionary(double[] classValues) {
      Dictionary<double, int> classCount = new Dictionary<double, int>();
      for (int i = 0; i < classValues.Length; i++) {
        classCount[classValues[i]] = 0;
      }
      return classCount;
    }
  }
}