source: branches/EfficientGlobalOptimization/HeuristicLab.Algorithms.EGO/Operators/FitnessClusteringAnalyzer.cs @ 15976

#region License Information
/* HeuristicLab
 * Copyright (C) 2002-2016 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.Linq;
using System.Threading;
using HeuristicLab.Algorithms.DataAnalysis;
using HeuristicLab.Analysis;
using HeuristicLab.Common;
using HeuristicLab.Core;
using HeuristicLab.Data;
using HeuristicLab.Operators;
using HeuristicLab.Optimization;
using HeuristicLab.Parameters;
using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
using HeuristicLab.Problems.DataAnalysis;

namespace HeuristicLab.Algorithms.EGO {
  [Item("FitnessClusteringAnalyzer", "Analyzes the correlation between perdictions and actual fitness values")]
  [StorableClass]
  public class FitnessClusteringAnalyzer : SingleSuccessorOperator, IAnalyzer, IStochasticOperator, IResultsOperator {
    public override bool CanChangeName => true;
    public bool EnabledByDefault => false;

    public ILookupParameter<ModifiableDataset> DatasetParameter => (ILookupParameter<ModifiableDataset>)Parameters["Dataset"];
    public ILookupParameter<ResultCollection> ResultsParameter => (ILookupParameter<ResultCollection>)Parameters["Results"];
    public IFixedValueParameter<IntValue> KParameter => (IFixedValueParameter<IntValue>)Parameters["K"];
    public IFixedValueParameter<IntValue> K2Parameter => (IFixedValueParameter<IntValue>)Parameters["K2"];
    public ILookupParameter<IRandom> RandomParameter => (ILookupParameter<IRandom>)Parameters["Random"];

    private const string SolutionName = "FitnessClustering";
    private const string PlotName = "FitnessClusterPlot";

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

    protected FitnessClusteringAnalyzer(FitnessClusteringAnalyzer original, Cloner cloner) : base(original, cloner) { }

    public FitnessClusteringAnalyzer() {
      Parameters.Add(new LookupParameter<ModifiableDataset>("Dataset"));
      Parameters.Add(new LookupParameter<ResultCollection>("Results", "The collection to store the results in."));
      Parameters.Add(new FixedValueParameter<IntValue>("K", "The number of clusters.", new IntValue(3)));
      Parameters.Add(new FixedValueParameter<IntValue>("K2", "The number of clusters.", new IntValue(3)));
      Parameters.Add(new LookupParameter<IRandom>("Random"));
    }

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

    public sealed override IOperation Apply() {
      var dataset = DatasetParameter.ActualValue;
      var results = ResultsParameter.ActualValue;
      var random = RandomParameter.ActualValue;
      if (dataset.Rows < KParameter.Value.Value || dataset.Rows < 20) return base.Apply();

      var clustering = CreateClustering(dataset, random);
      if (!results.ContainsKey(SolutionName)) results.Add(new Result(SolutionName, clustering));
      results[SolutionName].Value = clustering;
      var plot = CreateTSNEPlot(clustering, dataset, random);
      if (!results.ContainsKey(PlotName)) results.Add(new Result(PlotName, plot));
      results[PlotName].Value = plot;

      return base.Apply();
    }


    private ScatterPlot CreateTSNEPlot(KMeansClusteringSolution clustering, ModifiableDataset data, IRandom random) {
      var clusteredData = (ModifiableDataset)data.Clone();
      clusteredData.AddVariable("cluster", clustering.ClusterValues.Select(x => (double)x).ToList());

      var prob = new RegressionProblem {
        ProblemData = new RegressionProblemData(clusteredData, new[] { "output" }, "cluster")
      };
      var tsne = new TSNEAlgorithm {
        Perplexity = data.Rows / 3 - 1,
        Problem = prob
      };
      tsne.ClassesNameParameter.Value = tsne.ClassesNameParameter.ValidValues.FirstOrDefault(x => x.Value.Equals("cluster"));
      var res = EgoUtilities.SyncRunSubAlgorithm(tsne, random.Next(), CancellationToken.None);
      return res.Select(r => r.Value).OfType<ScatterPlot>().First();
    }

    private KMeansClusteringSolution CreateClustering(ModifiableDataset dataset, IRandom random) {
      var pd = new ClusteringProblemData(dataset, new[] { "output" });
      pd.TestPartition.Start = dataset.Rows;
      pd.TestPartition.End = dataset.Rows;
      pd.TrainingPartition.Start = 0;
      pd.TrainingPartition.End = dataset.Rows;
      return KMeansClustering.CreateKMeansSolution(pd, KParameter.Value.Value, 1);
    }
    private double[] GetWeights(ModifiableDataset dataset) {
      var inputMatrix = dataset.ToArray(dataset.VariableNames.Where(x => x.StartsWith("input")), Enumerable.Range(0, dataset.Rows));
      if (inputMatrix.Cast<double>().Any(x => double.IsNaN(x) || double.IsInfinity(x)))
        throw new NotSupportedException("k-Means clustering does not support NaN or infinity values in the input dataset.");
      var indices = Enumerable.Range(0, inputMatrix.GetLength(0)).ToArray();
      return indices.Select(i =>
            K2Parameter.Value.Value > 0 ? indices.Where(j => j != i).Select(j =>
                    1 / Math.Sqrt(EuclideanSquared(inputMatrix, inputMatrix, i, j))
            ).OrderBy(x => x).Take(K2Parameter.Value.Value).Sum() : 1.0
         ).ToArray();
    }
    private static double EuclideanSquared(double[,] input, double[,] input2, int row1, int row2) {
      var sum = 0.0;
      for (var i = 0; i < input.GetLength(1); i++) {
        var d = input[row1, i] - input2[row2, i];
        sum += d * d;
      }
      return sum;
    }

  }
}