source: branches/Sliding Window GP/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/SlidingWindow/SlidingWindowBestSolutionsCollection.cs @ 12382

#region License Information
/* HeuristicLab
 * Copyright (C) 2002-2013 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.ComponentModel;
using System.Linq;
using HeuristicLab.Common;
using HeuristicLab.Core;
using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;

namespace HeuristicLab.Problems.DataAnalysis.Symbolic {
  [StorableClass]
  [Item("SlidingWindowBestSolutionsCollection", "An object holding a collection of the best sliding window solutions.")]
  public abstract class SlidingWindowBestSolutionsCollection : Item {
    [Storable]
    private List<SlidingWindowRange> slidingWindowRanges;

    public List<SlidingWindowRange> SlidingWindowRanges {
      get { return slidingWindowRanges; }
      private set { slidingWindowRanges = value; }
    }

    [Storable(AllowOneWay = true, Name = "bestSolutions")]
    private Dictionary<Tuple<int, int>, ISymbolicExpressionTree> StorableBestSolutions {
      set {
        var bestSolutions = value;
        var ranges = bestSolutions.Keys.OrderBy(x => x.Item1).ToList();
        slidingWindowRanges = ranges.Select(x => new SlidingWindowRange(x.Item1, x.Item2)).ToList();
        slidingWindowBestSolutions = new Dictionary<SlidingWindowRange, ISymbolicExpressionTree>();
        for (int i = 0; i < slidingWindowRanges.Count; ++i) {
          slidingWindowBestSolutions.Add(slidingWindowRanges[i], bestSolutions[ranges[i]]);
        }
      }
    }

    [Storable]
    private Dictionary<SlidingWindowRange, ISymbolicExpressionTree> slidingWindowBestSolutions;

    public Dictionary<SlidingWindowRange, ISymbolicExpressionTree> SlidingWindowBestSolutions {
      get { return slidingWindowBestSolutions; }
      set { slidingWindowBestSolutions = value; }
    }

    [Storable]
    private IDataAnalysisProblemData problemData;

    public IDataAnalysisProblemData ProblemData {
      get { return problemData; }
      set { problemData = value; }
    }

    [Storable]
    private ISymbolicDataAnalysisExpressionTreeInterpreter interpreter;

    public ISymbolicDataAnalysisExpressionTreeInterpreter Interpreter {
      get { return interpreter; }
      set { interpreter = value; }
    }

    [Storable]
    private bool applyLinearScaling;

    public bool ApplyLinearScaling {
      get { return applyLinearScaling; }
      set { applyLinearScaling = value; }
    }

    [StorableHook(HookType.AfterDeserialization)]
    private void AfterDeserialization() {
      if (bw == null) {
        bw = new BackgroundWorker();
        bw.WorkerSupportsCancellation = true;
        bw.WorkerReportsProgress = true;
        bw.DoWork += CalculateQualities;
      }
    }

    public double[,] SlidingWindowQualities { get; set; }

    private BackgroundWorker bw;

    public enum QualityMeasures { PEARSON, MSE };

    private QualityMeasures qualityMeasure;

    public QualityMeasures QualityMeasure {
      get { return qualityMeasure; }
      set {
        if (qualityMeasure != value) {
          qualityMeasure = value;
          CalculateQualities();
        }
      }
    }

    public bool QualitiesCalculationInProgress {
      get { return bw.IsBusy; }
    }

    public event ProgressChangedEventHandler QualitiesCalculationProgress {
      add { bw.ProgressChanged += value; }
      remove { bw.ProgressChanged -= value; }
    }

    public event RunWorkerCompletedEventHandler QualitiesCalculationCompleted {
      add { bw.RunWorkerCompleted += value; }
      remove { bw.RunWorkerCompleted -= value; }
    }

    public event EventHandler QualitiesCalculationStarted;

    private void OnQualitiesCalculationStarted(object sender, EventArgs e) {
      var started = QualitiesCalculationStarted;
      if (started != null) {
        started(sender, e);
      }
    }

    public event EventHandler QualitiesUpdated;
    private void OnQualitiesUpdated(object sender, EventArgs e) {
      var updated = QualitiesUpdated;
      if (updated != null) {
        updated(sender, e);
      }
    }

    [StorableConstructor]
    protected SlidingWindowBestSolutionsCollection(bool deserializing)
      : base(deserializing) {
    }

    protected SlidingWindowBestSolutionsCollection(SlidingWindowBestSolutionsCollection original, Cloner cloner)
      : base(original, cloner) {
      this.slidingWindowBestSolutions = original.slidingWindowBestSolutions;
      this.problemData = original.problemData;
      this.interpreter = original.interpreter;
      this.applyLinearScaling = original.ApplyLinearScaling;
    }

    protected SlidingWindowBestSolutionsCollection() {
      slidingWindowBestSolutions = new Dictionary<SlidingWindowRange, ISymbolicExpressionTree>();
      slidingWindowRanges = new List<SlidingWindowRange>();
      qualityMeasure = QualityMeasures.PEARSON;

      bw = new BackgroundWorker();
      bw.WorkerSupportsCancellation = true;
      bw.WorkerReportsProgress = true;

      bw.DoWork += CalculateQualities;
    }

    public bool ContainsKey(SlidingWindowRange key) {
      return slidingWindowBestSolutions.ContainsKey(key);
    }

    public ISymbolicExpressionTree this[SlidingWindowRange key] {
      get { return slidingWindowBestSolutions[key]; }
      set {
        AddSolution(key, value); // this should be fast so there's no need for a background worker
        OnQualitiesUpdated(this, EventArgs.Empty);
      }
    }

    public void Add(SlidingWindowRange range, ISymbolicExpressionTree solution) {
      if (!slidingWindowBestSolutions.ContainsKey(range)) {
        slidingWindowBestSolutions.Add(range, solution);
        slidingWindowRanges.Add(range);
      } else {
        slidingWindowBestSolutions[range] = solution;
      }
    }

    public void Clear() {
      if (slidingWindowBestSolutions != null) slidingWindowBestSolutions.Clear();
      if (slidingWindowRanges != null) slidingWindowRanges.Clear();
    }

    public abstract ISymbolicDataAnalysisModel CreateModel(ISymbolicExpressionTree tree,
      ISymbolicDataAnalysisExpressionTreeInterpreter interpreter,
      double lowerEstimationLimit = double.MinValue, double upperEstimationLimit = double.MaxValue);

    public abstract ISymbolicDataAnalysisSolution CreateSolution(ISymbolicDataAnalysisModel model,
      IDataAnalysisProblemData problemData);

    private void AddSolution(SlidingWindowRange range, ISymbolicExpressionTree solution) {
      Add(range, solution);

      var solutions = slidingWindowRanges.Select(x => slidingWindowBestSolutions[x]).ToList();

      var nRows = solutions.Count;
      var nCols = nRows + 1; // an extra column corresponding to the whole trainig partition

      var trainingIndices = problemData.TrainingIndices.ToList();
      var matrix = new double[nRows, nCols];

      // copy old qualities into the new matrix
      for (int i = 0; i < nRows - 1; ++i) {
        for (int j = 0; j < nCols - 1; ++j) {
          matrix[i, j] = SlidingWindowQualities[i, j];
        }
      }
      // copy qualities of new solution into the new matrix
      var rows = Enumerable.Range(slidingWindowRanges.First().Start, slidingWindowRanges.Last().End - slidingWindowRanges.First().Start).ToList();
      var estimatedValues = Interpreter.GetSymbolicExpressionTreeValues(solution, problemData.Dataset, rows).ToList();
      var originalValues = ProblemData.Dataset.GetDoubleValues(GetTargetVariable(ProblemData), rows).ToList();
      for (int i = 0; i < nCols; ++i) {
        if (i == nCols - 1) {
          matrix[nRows - 1, i] = CalculateQuality(solution, trainingIndices);
        } else {
          var indices = Enumerable.Range(slidingWindowRanges[i].Start, slidingWindowRanges[i].Size).ToList();
          var estimated = indices.Select(x => estimatedValues[x]);
          var original = indices.Select(x => originalValues[x]);
          matrix[nRows - 1, i] = CalculateQuality(estimated, original);
        }
      }
      // shift old training qualities one column to the right
      rows = Enumerable.Range(range.Start, range.Size).ToList();
      for (int i = 0; i < nRows; ++i) {
        matrix[i, nCols - 1] = matrix[i, nCols - 2];
        matrix[i, nCols - 2] = CalculateQuality(solutions[i], rows);
      }
      // replace old matrix with new matrix
      SlidingWindowQualities = matrix;
    }

    private void CalculateQualities(object sender, DoWorkEventArgs e) {
      var worker = sender as BackgroundWorker;
      if (worker == null) return;
      if (worker.CancellationPending) {
        e.Cancel = true;
        return;
      }

      OnQualitiesCalculationStarted(this, EventArgs.Empty);

      var ranges = SlidingWindowRanges;
      var solutions = ranges.Select(x => SlidingWindowBestSolutions[x]).ToList();

      int nRows = solutions.Count;
      int nCols = ranges.Count + 1;

      SlidingWindowQualities = new double[nRows, nCols];
      var rows = Enumerable.Range(ranges.First().Start, ranges.Last().End - ranges.First().Start).ToList();
      var originalValues = ProblemData.Dataset.GetDoubleValues(GetTargetVariable(ProblemData), rows).ToList();

      for (int i = 0; i < nRows; ++i) {
        if (worker.CancellationPending) {
          e.Cancel = true;
          return;
        }

        var solution = solutions[i];
        var estimatedValues = Interpreter.GetSymbolicExpressionTreeValues(solution, problemData.Dataset, rows).ToList();

        for (int j = 0; j < nCols; ++j) {
          double q;
          if (j == nCols - 1) {
            q = CalculateQuality(solution, problemData.TrainingIndices);
          } else {
            var range = ranges[j];
            var indices = Enumerable.Range(range.Start, range.Size).ToList();
            var estimated = indices.Select(x => estimatedValues[x]);
            var original = indices.Select(x => originalValues[x]);

            q = CalculateQuality(estimated, original);
          }

          SlidingWindowQualities[i, j] = q;
        }

        worker.ReportProgress((int)Math.Round(i * 100.0 / nRows));
      }
    }

    public void CalculateQualities() {
      bw.RunWorkerAsync();
    }

    private string GetTargetVariable(IDataAnalysisProblemData problemData) {
      var regressionProblemData = problemData as IRegressionProblemData;
      var classificationProblemData = problemData as IClassificationProblemData;
      if (regressionProblemData != null) return regressionProblemData.TargetVariable;
      if (classificationProblemData != null) return classificationProblemData.TargetVariable;
      throw new NotSupportedException();
    }

    private double CalculateQuality(IEnumerable<double> estimatedValues, IEnumerable<double> originalValues) {
      var errorState = OnlineCalculatorError.None;
      double quality = 0.0;
      switch (QualityMeasure) {
        case QualityMeasures.PEARSON:
          quality = OnlinePearsonsRSquaredCalculator.Calculate(estimatedValues, originalValues, out errorState);
          break;
        case QualityMeasures.MSE:
          quality = OnlineMeanSquaredErrorCalculator.Calculate(estimatedValues, originalValues, out errorState);
          break;
      }
      return errorState == OnlineCalculatorError.None ? quality : double.NaN;
    }

    private double CalculateQuality(ISymbolicExpressionTree tree, IEnumerable<int> rows) {
      var estimatedValues = Interpreter.GetSymbolicExpressionTreeValues(tree, ProblemData.Dataset, rows);
      var originalValues = ProblemData.Dataset.GetDoubleValues(GetTargetVariable(ProblemData), rows);
      double quality = 0;
      var errorState = new OnlineCalculatorError();
      switch (QualityMeasure) {
        case QualityMeasures.PEARSON:
          quality = OnlinePearsonsRSquaredCalculator.Calculate(estimatedValues, originalValues, out errorState);
          break;
        case QualityMeasures.MSE:
          quality = OnlineMeanSquaredErrorCalculator.Calculate(estimatedValues, originalValues, out errorState);
          break;
      }
      return errorState == OnlineCalculatorError.None ? quality : double.NaN;
    }
  }

  [StorableClass]
  public sealed class SlidingWindowRange : IEquatable<SlidingWindowRange> {
    [Storable]
    private readonly Tuple<int, int> tuple;

    public int Start { get { return tuple.Item1; } }
    public int End { get { return tuple.Item2; } }

    [StorableConstructor]
    private SlidingWindowRange(bool deserializable) { }
    private SlidingWindowRange() { }

    public SlidingWindowRange(int start, int end) {
      if (start > end) throw new ArgumentException("SlidingWindowRange: Start cannot be greater than End.");
      tuple = new Tuple<int, int>(start, end);
    }

    public bool Equals(SlidingWindowRange other) {
      return tuple.Equals(other.tuple);
    }

    public override int GetHashCode() {
      return tuple.GetHashCode();
    }

    public int Size {
      get { return End - Start; }
    }
  }
}