source: trunk/HeuristicLab.Algorithms.DataAnalysis/3.4/CrossValidation.cs @ 18190

#region License Information
/* HeuristicLab
 * Copyright (C) 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.Drawing;
using System.Linq;
using System.Threading;
using System.Threading.Tasks;
using HeuristicLab.Collections;
using HeuristicLab.Common;
using HeuristicLab.Core;
using HeuristicLab.Data;
using HeuristicLab.Optimization;
using HEAL.Attic;
using HeuristicLab.Problems.DataAnalysis;
using HeuristicLab.Problems.DataAnalysis.Symbolic;
using HeuristicLab.Random;
using HeuristicLab.Problems.DataAnalysis.Symbolic.Regression;
using HeuristicLab.Problems.DataAnalysis.Symbolic.Classification;

namespace HeuristicLab.Algorithms.DataAnalysis {
  [Item("Cross Validation (CV)", "Cross-validation wrapper for data analysis algorithms.")]
  [Creatable(CreatableAttribute.Categories.DataAnalysis, Priority = 100)]
  [StorableType("1C622121-AE5B-42FD-831C-FCA8F8E0AF8D")]
  public sealed class CrossValidation : ParameterizedNamedItem, IAlgorithm, IStorableContent {
    [Storable]
    private int seed;

    private SemaphoreSlim availableWorkers; // limits the number of concurrent algorithm executions
    private ManualResetEventSlim allAlgorithmsFinished; // this indicates that all started algorithms have been paused or stopped

    public CrossValidation()
      : base() {
      name = ItemName;
      description = ItemDescription;

      executionState = ExecutionState.Stopped;
      runs = new RunCollection { OptimizerName = name };
      runsCounter = 0;

      algorithm = null;
      clonedAlgorithms = new ItemCollection<IAlgorithm>();
      results = new ResultCollection();

      folds = new IntValue(2);
      numberOfWorkers = new IntValue(1);
      samplesStart = new IntValue(0);
      samplesEnd = new IntValue(0);
      shuffleSamples = new BoolValue(false);
      storeAlgorithmInEachRun = false;

      RegisterEvents();
      if (Algorithm != null) RegisterAlgorithmEvents();
    }

    public string Filename { get; set; }

    #region persistence and cloning
    [StorableConstructor]
    private CrossValidation(StorableConstructorFlag _) : base(_) {
    }
    [StorableHook(HookType.AfterDeserialization)]
    private void AfterDeserialization() {
      // BackwardsCompatibility3.3
      #region Backwards compatible code, remove with 3.4
      if (shuffleSamples == null) shuffleSamples = new BoolValue(false);
      #endregion

      RegisterEvents();
      if (Algorithm != null) RegisterAlgorithmEvents();
    }

    private CrossValidation(CrossValidation original, Cloner cloner)
      : base(original, cloner) {
      executionState = original.executionState;
      storeAlgorithmInEachRun = original.storeAlgorithmInEachRun;
      runs = cloner.Clone(original.runs);
      runsCounter = original.runsCounter;
      algorithm = cloner.Clone(original.algorithm);
      clonedAlgorithms = cloner.Clone(original.clonedAlgorithms);
      results = cloner.Clone(original.results);

      folds = cloner.Clone(original.folds);
      numberOfWorkers = cloner.Clone(original.numberOfWorkers);
      samplesStart = cloner.Clone(original.samplesStart);
      samplesEnd = cloner.Clone(original.samplesEnd);
      shuffleSamples = cloner.Clone(original.shuffleSamples);
      seed = original.seed;

      RegisterEvents();
      if (Algorithm != null) RegisterAlgorithmEvents();
    }
    public override IDeepCloneable Clone(Cloner cloner) {
      return new CrossValidation(this, cloner);
    }

    #endregion

    #region properties
    [Storable]
    private IAlgorithm algorithm;
    public IAlgorithm Algorithm {
      get { return algorithm; }
      set {
        if (ExecutionState != ExecutionState.Prepared && ExecutionState != ExecutionState.Stopped)
          throw new InvalidOperationException("Changing the algorithm is only allowed if the CrossValidation is stopped or prepared.");
        if (algorithm != value) {
          if (value != null && value.Problem != null && !(value.Problem is IDataAnalysisProblem))
            throw new ArgumentException("Only algorithms with a DataAnalysisProblem could be used for the cross validation.");
          if (algorithm != null) DeregisterAlgorithmEvents();
          algorithm = value;
          Parameters.Clear();

          if (algorithm != null) {
            algorithm.StoreAlgorithmInEachRun = false;
            RegisterAlgorithmEvents();
            algorithm.Prepare(true);
            Parameters.AddRange(algorithm.Parameters);
          }
          OnAlgorithmChanged();
          Prepare();
        }
      }
    }


    [Storable]
    private IDataAnalysisProblem problem;
    public IDataAnalysisProblem Problem {
      get {
        if (algorithm == null)
          return null;
        return (IDataAnalysisProblem)algorithm.Problem;
      }
      set {
        if (ExecutionState != ExecutionState.Prepared && ExecutionState != ExecutionState.Stopped)
          throw new InvalidOperationException("Changing the problem is only allowed if the CrossValidation is stopped or prepared.");
        if (algorithm == null) throw new ArgumentNullException("Could not set a problem before an algorithm was set.");
        algorithm.Problem = value;
        problem = value;
      }
    }

    IProblem IAlgorithm.Problem {
      get { return Problem; }
      set {
        if (value != null && !ProblemType.IsInstanceOfType(value))
          throw new ArgumentException("Only DataAnalysisProblems could be used for the cross validation.");
        Problem = (IDataAnalysisProblem)value;
      }
    }
    public Type ProblemType {
      get { return typeof(IDataAnalysisProblem); }
    }

    [Storable]
    private ItemCollection<IAlgorithm> clonedAlgorithms;

    public IEnumerable<IOptimizer> NestedOptimizers {
      get {
        if (Algorithm == null) yield break;
        yield return Algorithm;
      }
    }

    [Storable]
    private ResultCollection results;
    public ResultCollection Results {
      get { return results; }
    }
    [Storable]
    private BoolValue shuffleSamples;
    public BoolValue ShuffleSamples {
      get { return shuffleSamples; }
    }
    [Storable]
    private IntValue folds;
    public IntValue Folds {
      get { return folds; }
    }
    [Storable]
    private IntValue samplesStart;
    public IntValue SamplesStart {
      get { return samplesStart; }
    }
    [Storable]
    private IntValue samplesEnd;
    public IntValue SamplesEnd {
      get { return samplesEnd; }
    }
    [Storable]
    private IntValue numberOfWorkers;
    public IntValue NumberOfWorkers {
      get { return numberOfWorkers; }
    }

    [Storable]
    private bool storeAlgorithmInEachRun;
    public bool StoreAlgorithmInEachRun {
      get { return storeAlgorithmInEachRun; }
      set {
        if (storeAlgorithmInEachRun != value) {
          storeAlgorithmInEachRun = value;
          OnStoreAlgorithmInEachRunChanged();
        }
      }
    }

    [Storable]
    private int runsCounter;
    [Storable]
    private RunCollection runs;
    public RunCollection Runs {
      get { return runs; }
    }

    [Storable]
    private ExecutionState executionState;
    public ExecutionState ExecutionState {
      get { return executionState; }
      private set {
        if (executionState != value) {
          executionState = value;
          OnExecutionStateChanged();
          OnItemImageChanged();
        }
      }
    }
    public static new Image StaticItemImage {
      get { return HeuristicLab.Common.Resources.VSImageLibrary.Event; }
    }
    public override Image ItemImage {
      get {
        if (ExecutionState == ExecutionState.Prepared) return HeuristicLab.Common.Resources.VSImageLibrary.ExecutablePrepared;
        else if (ExecutionState == ExecutionState.Started) return HeuristicLab.Common.Resources.VSImageLibrary.ExecutableStarted;
        else if (ExecutionState == ExecutionState.Paused) return HeuristicLab.Common.Resources.VSImageLibrary.ExecutablePaused;
        else if (ExecutionState == ExecutionState.Stopped) return HeuristicLab.Common.Resources.VSImageLibrary.ExecutableStopped;
        else return base.ItemImage;
      }
    }

    public TimeSpan ExecutionTime {
      get {
        if (ExecutionState != ExecutionState.Prepared)
          return TimeSpan.FromMilliseconds(clonedAlgorithms.Select(x => x.ExecutionTime.TotalMilliseconds).Sum());
        return TimeSpan.Zero;
      }
    }
    #endregion

    protected override void OnNameChanged() {
      base.OnNameChanged();
      Runs.OptimizerName = Name;
    }

    public void Prepare() {
      if (startPending) return;
      if (ExecutionState == ExecutionState.Started)
        throw new InvalidOperationException(string.Format("Prepare not allowed in execution state \"{0}\".", ExecutionState));
      results.Clear();
      clonedAlgorithms.Clear();
      if (Algorithm != null) {
        Algorithm.Prepare();
        if (Algorithm.ExecutionState == ExecutionState.Prepared) OnPrepared();
      }
    }
    public void Prepare(bool clearRuns) {
      if (clearRuns) runs.Clear();
      Prepare();
    }

    private bool startPending;
    public void Start() {
      Start(CancellationToken.None);
    }
    public void Start(CancellationToken cancellationToken) {
      lock (locker) {
        if (startPending) return;
        startPending = true;
      }

      try {
        if ((ExecutionState != ExecutionState.Prepared) && (ExecutionState != ExecutionState.Paused))
          throw new InvalidOperationException(string.Format("Start not allowed in execution state \"{0}\".", ExecutionState));
        seed = RandomSeedGenerator.GetSeed();

        if (Algorithm == null) return;
        //create cloned algorithms
        if (clonedAlgorithms.Count == 0) {
          int testSamplesCount = (SamplesEnd.Value - SamplesStart.Value) / Folds.Value;
          IDataset shuffledDataset = null;
          for (int i = 0; i < Folds.Value; i++) {
            var cloner = new Cloner();
            if (ShuffleSamples.Value) {
              var random = new FastRandom(seed);
              var dataAnalysisProblem = (IDataAnalysisProblem)algorithm.Problem;
              var dataset = (Dataset)dataAnalysisProblem.ProblemData.Dataset;
              shuffledDataset = shuffledDataset ?? dataset.Shuffle(random);
              cloner.RegisterClonedObject(dataset, shuffledDataset);
            }
            IAlgorithm clonedAlgorithm = cloner.Clone(Algorithm);
            clonedAlgorithm.Name = algorithm.Name + " Fold " + i;
            IDataAnalysisProblem problem = clonedAlgorithm.Problem as IDataAnalysisProblem;
            ISymbolicDataAnalysisProblem symbolicProblem = problem as ISymbolicDataAnalysisProblem;

            int testStart = (i * testSamplesCount) + SamplesStart.Value;
            int testEnd = (i + 1) == Folds.Value ? SamplesEnd.Value : (i + 1) * testSamplesCount + SamplesStart.Value;

            problem.ProblemData.TrainingPartition.Start = SamplesStart.Value;
            problem.ProblemData.TrainingPartition.End = SamplesEnd.Value;
            problem.ProblemData.TestPartition.Start = testStart;
            problem.ProblemData.TestPartition.End = testEnd;
            DataAnalysisProblemData problemData = problem.ProblemData as DataAnalysisProblemData;
            if (problemData != null) {
              problemData.TrainingPartitionParameter.Hidden = false;
              problemData.TestPartitionParameter.Hidden = false;
            }

            if (symbolicProblem != null) {
              symbolicProblem.FitnessCalculationPartition.Start = SamplesStart.Value;
              symbolicProblem.FitnessCalculationPartition.End = SamplesEnd.Value;
            }

            // We need to set the estimation limits because they are recalculated by the problem
            // whenever the data partitions change.
            // Instead of explicitly handling all types we could also check the parameters-collection 
            // for a parameter with name "EstimationLimits".
            SetEstimationLimits(problem, new[] { typeof(SymbolicRegressionSingleObjectiveProblem),
                                                 typeof(SymbolicRegressionMultiObjectiveProblem),
                                                 typeof(SymbolicClassificationSingleObjectiveProblem),
                                                 typeof(SymbolicClassificationMultiObjectiveProblem) });

            clonedAlgorithm.Prepare();
            clonedAlgorithms.Add(clonedAlgorithm);
          }
        }

        OnStarted();
      } finally {
        if (startPending) startPending = false;
      }

      availableWorkers = new SemaphoreSlim(NumberOfWorkers.Value, NumberOfWorkers.Value);
      allAlgorithmsFinished = new ManualResetEventSlim(false);

      var startedTasks = new List<Task>(clonedAlgorithms.Count);

      //start prepared or paused cloned algorithms
      foreach (IAlgorithm clonedAlgorithm in clonedAlgorithms) {
        if (pausePending || stopPending || ExecutionState != ExecutionState.Started) break;
        if (clonedAlgorithm.ExecutionState == ExecutionState.Prepared ||
            clonedAlgorithm.ExecutionState == ExecutionState.Paused) {
          availableWorkers.Wait();
          lock (locker) {
            if (pausePending || stopPending || ExecutionState != ExecutionState.Started) break;
            var task = clonedAlgorithm.StartAsync(cancellationToken);
            startedTasks.Add(task);
          }
        }
      }

      allAlgorithmsFinished.Wait();

      Task.WaitAll(startedTasks.ToArray()); // to get exceptions not handled within the tasks
    }

    public async Task StartAsync() { await StartAsync(CancellationToken.None); }
    public async Task StartAsync(CancellationToken cancellationToken) {
      await AsyncHelper.DoAsync(Start, cancellationToken);
    }

    private bool pausePending;
    public void Pause() {
      if (startPending) return;
      if (ExecutionState != ExecutionState.Started)
        throw new InvalidOperationException(string.Format("Pause not allowed in execution state \"{0}\".", ExecutionState));
      if (!pausePending) {
        pausePending = true;
        lock (locker) {
          var toPause = clonedAlgorithms.Where(x => x.ExecutionState == ExecutionState.Started).ToList();
          foreach (var optimizer in toPause) {
            // a race-condition may occur when the optimizer has changed the state by itself in the meantime
            try { optimizer.Pause(); } catch (InvalidOperationException) { }
          }
        }
      }
    }

    private bool stopPending;
    public void Stop() {
      if (startPending) return;
      if ((ExecutionState != ExecutionState.Started) && (ExecutionState != ExecutionState.Paused))
        throw new InvalidOperationException(string.Format("Stop not allowed in execution state \"{0}\".",
                                                          ExecutionState));
      if (!stopPending) {
        stopPending = true;
        lock (locker) {
          var toStop = clonedAlgorithms.Where(x => x.ExecutionState == ExecutionState.Started || x.ExecutionState == ExecutionState.Paused).ToList();
          foreach (var optimizer in toStop) {
            // a race-condition may occur when the optimizer has changed the state by itself in the meantime
            try { optimizer.Stop(); } catch (InvalidOperationException) { }
          }
        }
      }
    }

    #region collect parameters and results
    public override void CollectParameterValues(IDictionary<string, IItem> values) {
      values.Add("Algorithm Name", new StringValue(Name));
      values.Add("Algorithm Type", new StringValue(GetType().GetPrettyName()));
      values.Add("Folds", new IntValue(Folds.Value));

      if (algorithm != null) {
        values.Add("CrossValidation Algorithm Name", new StringValue(Algorithm.Name));
        values.Add("CrossValidation Algorithm Type", new StringValue(Algorithm.GetType().GetPrettyName()));
        base.CollectParameterValues(values);
      }
      if (Problem != null) {
        values.Add("Problem Name", new StringValue(Problem.Name));
        values.Add("Problem Type", new StringValue(Problem.GetType().GetPrettyName()));
        Problem.CollectParameterValues(values);
      }
    }

    public void CollectResultValues(IDictionary<string, IItem> results) {
      var clonedResults = (ResultCollection)this.results.Clone();
      foreach (var result in clonedResults) {
        results.Add(result.Name, result.Value);
      }
    }

    private void AggregateResultValues(IDictionary<string, IItem> results) {
      IEnumerable<IRun> runs = clonedAlgorithms.Select(alg => alg.Runs.FirstOrDefault()).Where(run => run != null);
      IEnumerable<KeyValuePair<string, IItem>> resultCollections = runs.Where(x => x != null).SelectMany(x => x.Results).ToList();

      foreach (IResult result in ExtractAndAggregateResults<IntValue>(resultCollections))
        results.Add(result.Name, result.Value);
      foreach (IResult result in ExtractAndAggregateResults<DoubleValue>(resultCollections))
        results.Add(result.Name, result.Value);
      foreach (IResult result in ExtractAndAggregateResults<PercentValue>(resultCollections))
        results.Add(result.Name, result.Value);
      foreach (IResult result in ExtractAndAggregateRegressionSolutions(resultCollections)) {
        results.Add(result.Name, result.Value);
      }
      foreach (IResult result in ExtractAndAggregateClassificationSolutions(resultCollections)) {
        results.Add(result.Name, result.Value);
      }
      results.Add("Execution Time", new TimeSpanValue(this.ExecutionTime));
      results.Add("CrossValidation Folds", new RunCollection(runs));
    }

    private IEnumerable<IResult> ExtractAndAggregateRegressionSolutions(IEnumerable<KeyValuePair<string, IItem>> resultCollections) {
      Dictionary<string, List<IRegressionSolution>> resultSolutions = new Dictionary<string, List<IRegressionSolution>>();
      foreach (var result in resultCollections) {
        var regressionSolution = result.Value as IRegressionSolution;
        if (regressionSolution != null) {
          if (resultSolutions.ContainsKey(result.Key)) {
            resultSolutions[result.Key].Add(regressionSolution);
          } else {
            resultSolutions.Add(result.Key, new List<IRegressionSolution>() { regressionSolution });
          }
        }
      }
      List<IResult> aggregatedResults = new List<IResult>();
      foreach (KeyValuePair<string, List<IRegressionSolution>> solutions in resultSolutions) {
        // clone manually to correctly clone references between cloned root objects 
        Cloner cloner = new Cloner();
        if (ShuffleSamples.Value) {
          var dataset = (Dataset)Problem.ProblemData.Dataset;
          var random = new FastRandom(seed);
          var shuffledDataset = dataset.Shuffle(random);
          cloner.RegisterClonedObject(dataset, shuffledDataset);
        }
        var problemDataClone = (IRegressionProblemData)cloner.Clone(Problem.ProblemData);
        // set partitions of problem data clone correctly
        problemDataClone.TrainingPartition.Start = SamplesStart.Value; problemDataClone.TrainingPartition.End = SamplesEnd.Value;
        problemDataClone.TestPartition.Start = SamplesStart.Value; problemDataClone.TestPartition.End = SamplesEnd.Value;
        // clone models
        var ensembleSolution = new RegressionEnsembleSolution(problemDataClone);
        ensembleSolution.AddRegressionSolutions(solutions.Value);

        aggregatedResults.Add(new Result(solutions.Key + " (ensemble)", ensembleSolution));
      }
      List<IResult> flattenedResults = new List<IResult>();
      CollectResultsRecursively("", aggregatedResults, flattenedResults);
      return flattenedResults;
    }

    private IEnumerable<IResult> ExtractAndAggregateClassificationSolutions(IEnumerable<KeyValuePair<string, IItem>> resultCollections) {
      Dictionary<string, List<IClassificationSolution>> resultSolutions = new Dictionary<string, List<IClassificationSolution>>();
      foreach (var result in resultCollections) {
        var classificationSolution = result.Value as IClassificationSolution;
        if (classificationSolution != null) {
          if (resultSolutions.ContainsKey(result.Key)) {
            resultSolutions[result.Key].Add(classificationSolution);
          } else {
            resultSolutions.Add(result.Key, new List<IClassificationSolution>() { classificationSolution });
          }
        }
      }
      var aggregatedResults = new List<IResult>();
      foreach (KeyValuePair<string, List<IClassificationSolution>> solutions in resultSolutions) {
        // at least one algorithm (GBT with logistic regression loss) produces a classification solution even though the original problem is a regression problem.
        var dataset = (Dataset)Problem.ProblemData.Dataset;
        if (ShuffleSamples.Value) {
          var random = new FastRandom(seed);
          dataset = dataset.Shuffle(random);
        }
        var problemData = (IClassificationProblemData)Problem.ProblemData;
        var problemDataClone = new ClassificationProblemData(dataset, problemData.AllowedInputVariables, problemData.TargetVariable, problemData.ClassNames, problemData.PositiveClass);
        // set partitions of problem data clone correctly
        problemDataClone.TrainingPartition.Start = SamplesStart.Value; problemDataClone.TrainingPartition.End = SamplesEnd.Value;
        problemDataClone.TestPartition.Start = SamplesStart.Value; problemDataClone.TestPartition.End = SamplesEnd.Value;
        // clone models
        var ensembleSolution = new ClassificationEnsembleSolution(problemDataClone);
        ensembleSolution.AddClassificationSolutions(solutions.Value);

        aggregatedResults.Add(new Result(solutions.Key + " (ensemble)", ensembleSolution));
      }
      List<IResult> flattenedResults = new List<IResult>();
      CollectResultsRecursively("", aggregatedResults, flattenedResults);
      return flattenedResults;
    }

    private void CollectResultsRecursively(string path, IEnumerable<IResult> results, IList<IResult> flattenedResults) {
      foreach (IResult result in results) {
        flattenedResults.Add(new Result(path + result.Name, result.Value));
        ResultCollection childCollection = result.Value as ResultCollection;
        if (childCollection != null) {
          CollectResultsRecursively(path + result.Name + ".", childCollection, flattenedResults);
        }
      }
    }

    private static IEnumerable<IResult> ExtractAndAggregateResults<T>(IEnumerable<KeyValuePair<string, IItem>> results)
  where T : class, IItem, new() {
      Dictionary<string, List<double>> resultValues = new Dictionary<string, List<double>>();
      foreach (var resultValue in results.Where(r => r.Value.GetType() == typeof(T))) {
        if (!resultValues.ContainsKey(resultValue.Key))
          resultValues[resultValue.Key] = new List<double>();
        resultValues[resultValue.Key].Add(ConvertToDouble(resultValue.Value));
      }

      DoubleValue doubleValue;
      if (typeof(T) == typeof(PercentValue))
        doubleValue = new PercentValue();
      else if (typeof(T) == typeof(DoubleValue))
        doubleValue = new DoubleValue();
      else if (typeof(T) == typeof(IntValue))
        doubleValue = new DoubleValue();
      else
        throw new NotSupportedException();

      List<IResult> aggregatedResults = new List<IResult>();
      foreach (KeyValuePair<string, List<double>> resultValue in resultValues) {
        doubleValue.Value = resultValue.Value.Average();
        aggregatedResults.Add(new Result(resultValue.Key + " (average)", (IItem)doubleValue.Clone()));
        doubleValue.Value = resultValue.Value.StandardDeviation();
        aggregatedResults.Add(new Result(resultValue.Key + " (std.dev.)", (IItem)doubleValue.Clone()));
      }
      return aggregatedResults;
    }

    private static double ConvertToDouble(IItem item) {
      if (item is DoubleValue) return ((DoubleValue)item).Value;
      else if (item is IntValue) return ((IntValue)item).Value;
      else throw new NotSupportedException("Could not convert any item type to double");
    }
    #endregion

    #region events
    private void RegisterEvents() {
      Folds.ValueChanged += new EventHandler(Folds_ValueChanged);
      RegisterClonedAlgorithmsEvents();
    }
    private void Folds_ValueChanged(object sender, EventArgs e) {
      if (ExecutionState != ExecutionState.Prepared)
        throw new InvalidOperationException("Can not change number of folds if the execution state is not prepared.");
    }


    #region template algorithms events
    public event EventHandler AlgorithmChanged;
    private void OnAlgorithmChanged() {
      EventHandler handler = AlgorithmChanged;
      if (handler != null) handler(this, EventArgs.Empty);
      OnProblemChanged();
      if (Problem == null) ExecutionState = ExecutionState.Stopped;
    }
    private void RegisterAlgorithmEvents() {
      algorithm.ProblemChanged += new EventHandler(Algorithm_ProblemChanged);
      algorithm.ExecutionStateChanged += new EventHandler(Algorithm_ExecutionStateChanged);
      if (Problem != null) {
        Problem.Reset += new EventHandler(Problem_Reset);
      }
    }
    private void DeregisterAlgorithmEvents() {
      algorithm.ProblemChanged -= new EventHandler(Algorithm_ProblemChanged);
      algorithm.ExecutionStateChanged -= new EventHandler(Algorithm_ExecutionStateChanged);
      if (Problem != null) {
        Problem.Reset -= new EventHandler(Problem_Reset);
      }
    }
    private void Algorithm_ProblemChanged(object sender, EventArgs e) {
      if (algorithm.Problem != null && !(algorithm.Problem is IDataAnalysisProblem)) {
        algorithm.Problem = problem;
        throw new ArgumentException("A cross validation algorithm can only contain DataAnalysisProblems.");
      }
      if (problem != null) problem.Reset -= new EventHandler(Problem_Reset);
      problem = (IDataAnalysisProblem)algorithm.Problem;
      if (problem != null) problem.Reset += new EventHandler(Problem_Reset);
      OnProblemChanged();
    }
    public event EventHandler ProblemChanged;
    private void OnProblemChanged() {
      EventHandler handler = ProblemChanged;
      if (handler != null) handler(this, EventArgs.Empty);
      ConfigureProblem();
    }
    private void Problem_Reset(object sender, EventArgs e) {
      ConfigureProblem();
    }
    private void ConfigureProblem() {
      SamplesStart.Value = 0;
      if (Problem != null) {
        SamplesEnd.Value = Problem.ProblemData.Dataset.Rows;

        DataAnalysisProblemData problemData = Problem.ProblemData as DataAnalysisProblemData;
        if (problemData != null) {
          problemData.TrainingPartitionParameter.Hidden = true;
          problemData.TestPartitionParameter.Hidden = true;
        }
        ISymbolicDataAnalysisProblem symbolicProblem = Problem as ISymbolicDataAnalysisProblem;
        if (symbolicProblem != null) {
          symbolicProblem.FitnessCalculationPartitionParameter.Hidden = true;
          symbolicProblem.FitnessCalculationPartition.Start = SamplesStart.Value;
          symbolicProblem.FitnessCalculationPartition.End = SamplesEnd.Value;
          symbolicProblem.ValidationPartitionParameter.Hidden = true;
          symbolicProblem.ValidationPartition.Start = 0;
          symbolicProblem.ValidationPartition.End = 0;
        }
      } else
        SamplesEnd.Value = 0;
    }

    private void Algorithm_ExecutionStateChanged(object sender, EventArgs e) {
      switch (Algorithm.ExecutionState) {
        case ExecutionState.Prepared:
          OnPrepared();
          break;
        case ExecutionState.Started: throw new InvalidOperationException("Algorithm template can not be started.");
        case ExecutionState.Paused: throw new InvalidOperationException("Algorithm template can not be paused.");
        case ExecutionState.Stopped:
          OnStopped();
          break;
      }
    }
    #endregion

    #region clonedAlgorithms events
    private void RegisterClonedAlgorithmsEvents() {
      clonedAlgorithms.ItemsAdded += new CollectionItemsChangedEventHandler<IAlgorithm>(ClonedAlgorithms_ItemsAdded);
      clonedAlgorithms.ItemsRemoved += new CollectionItemsChangedEventHandler<IAlgorithm>(ClonedAlgorithms_ItemsRemoved);
      clonedAlgorithms.CollectionReset += new CollectionItemsChangedEventHandler<IAlgorithm>(ClonedAlgorithms_CollectionReset);
      foreach (IAlgorithm algorithm in clonedAlgorithms)
        RegisterClonedAlgorithmEvents(algorithm);
    }
    private void DeregisterClonedAlgorithmsEvents() {
      clonedAlgorithms.ItemsAdded -= new CollectionItemsChangedEventHandler<IAlgorithm>(ClonedAlgorithms_ItemsAdded);
      clonedAlgorithms.ItemsRemoved -= new CollectionItemsChangedEventHandler<IAlgorithm>(ClonedAlgorithms_ItemsRemoved);
      clonedAlgorithms.CollectionReset -= new CollectionItemsChangedEventHandler<IAlgorithm>(ClonedAlgorithms_CollectionReset);
      foreach (IAlgorithm algorithm in clonedAlgorithms)
        DeregisterClonedAlgorithmEvents(algorithm);
    }
    private void ClonedAlgorithms_ItemsAdded(object sender, CollectionItemsChangedEventArgs<IAlgorithm> e) {
      foreach (IAlgorithm algorithm in e.Items)
        RegisterClonedAlgorithmEvents(algorithm);
    }
    private void ClonedAlgorithms_ItemsRemoved(object sender, CollectionItemsChangedEventArgs<IAlgorithm> e) {
      foreach (IAlgorithm algorithm in e.Items)
        DeregisterClonedAlgorithmEvents(algorithm);
    }
    private void ClonedAlgorithms_CollectionReset(object sender, CollectionItemsChangedEventArgs<IAlgorithm> e) {
      foreach (IAlgorithm algorithm in e.OldItems)
        DeregisterClonedAlgorithmEvents(algorithm);
      foreach (IAlgorithm algorithm in e.Items)
        RegisterClonedAlgorithmEvents(algorithm);
    }
    private void RegisterClonedAlgorithmEvents(IAlgorithm algorithm) {
      algorithm.ExceptionOccurred += new EventHandler<EventArgs<Exception>>(ClonedAlgorithm_ExceptionOccurred);
      algorithm.ExecutionTimeChanged += new EventHandler(ClonedAlgorithm_ExecutionTimeChanged);
      algorithm.Started += new EventHandler(ClonedAlgorithm_Started);
      algorithm.Paused += new EventHandler(ClonedAlgorithm_Paused);
      algorithm.Stopped += new EventHandler(ClonedAlgorithm_Stopped);
    }
    private void DeregisterClonedAlgorithmEvents(IAlgorithm algorithm) {
      algorithm.ExceptionOccurred -= new EventHandler<EventArgs<Exception>>(ClonedAlgorithm_ExceptionOccurred);
      algorithm.ExecutionTimeChanged -= new EventHandler(ClonedAlgorithm_ExecutionTimeChanged);
      algorithm.Started -= new EventHandler(ClonedAlgorithm_Started);
      algorithm.Paused -= new EventHandler(ClonedAlgorithm_Paused);
      algorithm.Stopped -= new EventHandler(ClonedAlgorithm_Stopped);
    }
    private void ClonedAlgorithm_ExceptionOccurred(object sender, EventArgs<Exception> e) {
      Pause();
      OnExceptionOccurred(e.Value);
    }
    private void ClonedAlgorithm_ExecutionTimeChanged(object sender, EventArgs e) {
      OnExecutionTimeChanged();
    }

    private readonly object locker = new object();
    private readonly object resultLocker = new object();
    private void ClonedAlgorithm_Started(object sender, EventArgs e) {
      IAlgorithm algorithm = sender as IAlgorithm;
      lock (resultLocker) {
        if (algorithm != null && !results.ContainsKey(algorithm.Name))
          results.Add(new Result(algorithm.Name, "Contains results for the specific fold.", algorithm.Results));
      }
    }

    private void ClonedAlgorithm_Paused(object sender, EventArgs e) {
      lock (locker) {
        availableWorkers.Release();
        if (clonedAlgorithms.All(alg => alg.ExecutionState != ExecutionState.Started)) {
          OnPaused();
          allAlgorithmsFinished.Set();
        }
      }
    }

    private void ClonedAlgorithm_Stopped(object sender, EventArgs e) {
      lock (locker) {
        // if the algorithm was in paused state, its worker has already been released
        if (availableWorkers.CurrentCount < NumberOfWorkers.Value)
          availableWorkers.Release();
        if (clonedAlgorithms.All(alg => alg.ExecutionState == ExecutionState.Stopped)) {
          OnStopped();
          allAlgorithmsFinished.Set();
        } else if (stopPending && clonedAlgorithms.All(alg => alg.ExecutionState == ExecutionState.Prepared || alg.ExecutionState == ExecutionState.Stopped)) {
          OnStopped();
          allAlgorithmsFinished.Set();
        }
      }
    }
    #endregion
    #endregion

    #region event firing
    public event EventHandler ExecutionStateChanged;
    private void OnExecutionStateChanged() {
      EventHandler handler = ExecutionStateChanged;
      if (handler != null) handler(this, EventArgs.Empty);
    }
    public event EventHandler ExecutionTimeChanged;
    private void OnExecutionTimeChanged() {
      EventHandler handler = ExecutionTimeChanged;
      if (handler != null) handler(this, EventArgs.Empty);
    }
    public event EventHandler Prepared;
    private void OnPrepared() {
      ExecutionState = ExecutionState.Prepared;
      EventHandler handler = Prepared;
      if (handler != null) handler(this, EventArgs.Empty);
      OnExecutionTimeChanged();
    }
    public event EventHandler Started;
    private void OnStarted() {
      startPending = false;
      ExecutionState = ExecutionState.Started;
      EventHandler handler = Started;
      if (handler != null) handler(this, EventArgs.Empty);
    }
    public event EventHandler Paused;
    private void OnPaused() {
      pausePending = false;
      ExecutionState = ExecutionState.Paused;
      EventHandler handler = Paused;
      if (handler != null) handler(this, EventArgs.Empty);
    }
    public event EventHandler Stopped;
    private void OnStopped() {
      stopPending = false;
      Dictionary<string, IItem> collectedResults = new Dictionary<string, IItem>();
      AggregateResultValues(collectedResults);
      results.AddRange(collectedResults.Select(x => new Result(x.Key, x.Value)).Cast<IResult>().ToArray());
      clonedAlgorithms.Clear();
      runsCounter++;
      runs.Add(new Run(string.Format("{0} Run {1}", Name, runsCounter), this));
      ExecutionState = ExecutionState.Stopped;
      EventHandler handler = Stopped;
      if (handler != null) handler(this, EventArgs.Empty);
    }
    public event EventHandler<EventArgs<Exception>> ExceptionOccurred;
    private void OnExceptionOccurred(Exception exception) {
      EventHandler<EventArgs<Exception>> handler = ExceptionOccurred;
      if (handler != null) handler(this, new EventArgs<Exception>(exception));
    }
    public event EventHandler StoreAlgorithmInEachRunChanged;
    private void OnStoreAlgorithmInEachRunChanged() {
      EventHandler handler = StoreAlgorithmInEachRunChanged;
      if (handler != null) handler(this, EventArgs.Empty);
    }
    #endregion

    #region helper

    private void SetEstimationLimits(IDataAnalysisProblem problem, Type[] types) {
      foreach (var type in types) {
        if (type.IsAssignableFrom(problem.GetType())) {
          var originalLimits = (DoubleLimit)Problem.Parameters["EstimationLimits"].ActualValue;  // problem is a clone of Problem
          var limits = (DoubleLimit)problem.Parameters["EstimationLimits"].ActualValue;
          limits.Lower = originalLimits.Lower;
          limits.Upper = originalLimits.Upper;
        }
      }
    }

    #endregion 
  }
}