Changeset 14666 for branches/PerformanceComparison/HeuristicLab.Algorithms.MemPR/3.3/MemPRContext.cs

Timestamp:

02/11/17 01:06:37 (7 years ago)

Author:

abeham

Message:

#2457: copied MemPR algorithm from its branch to this branch

Location:

branches/PerformanceComparison/HeuristicLab.Algorithms.MemPR

Files:

• . (copied) (copied from branches/MemPRAlgorithm/HeuristicLab.Algorithms.MemPR)
• 3.3/MemPRContext.cs (modified) (12 diffs)

branches/PerformanceComparison/HeuristicLab.Algorithms.MemPR/3.3/MemPRContext.cs

@@ -191 +191 @@
       get { return scope.SubScopes.Count; }
     }
-
-    [Storable]
-    private IConfidenceRegressionModel breedingPerformanceModel;
-    public IConfidenceRegressionModel BreedingPerformanceModel {
-      get { return breedingPerformanceModel; }
-    }
+    
     [Storable]
     private List<Tuple<double, double, double, double>> breedingStat;
@@ -203 +198 @@
     }
     [Storable]
-    private IConfidenceRegressionModel relinkingPerformanceModel;
-    public IConfidenceRegressionModel RelinkingPerformanceModel {
-      get { return relinkingPerformanceModel; }
-    }
-    [Storable]
     private List<Tuple<double, double, double, double>> relinkingStat;
     public IEnumerable<Tuple<double, double, double, double>> RelinkingStat {
@@ -213 +203 @@
     }
     [Storable]
-    private IConfidenceRegressionModel delinkingPerformanceModel;
-    public IConfidenceRegressionModel DelinkingPerformanceModel {
-      get { return delinkingPerformanceModel; }
-    }
-    [Storable]
     private List<Tuple<double, double, double, double>> delinkingStat;
     public IEnumerable<Tuple<double, double, double, double>> DelinkingStat {
@@ -223 +208 @@
     }
     [Storable]
-    private IConfidenceRegressionModel samplingPerformanceModel;
-    public IConfidenceRegressionModel SamplingPerformanceModel {
-      get { return samplingPerformanceModel; }
-    }
-    [Storable]
     private List<Tuple<double, double>> samplingStat;
     public IEnumerable<Tuple<double, double>> SamplingStat {
@@ -233 +213 @@
     }
     [Storable]
-    private IConfidenceRegressionModel hillclimbingPerformanceModel;
-    public IConfidenceRegressionModel HillclimbingPerformanceModel {
-      get { return hillclimbingPerformanceModel; }
-    }
-    [Storable]
     private List<Tuple<double, double>> hillclimbingStat;
     public IEnumerable<Tuple<double, double>> HillclimbingStat {
       get { return hillclimbingStat; }
-    }
-    [Storable]
-    private IConfidenceRegressionModel adaptiveWalkPerformanceModel;
-    public IConfidenceRegressionModel AdaptiveWalkPerformanceModel {
-      get { return adaptiveWalkPerformanceModel; }
     }
     [Storable]
@@ -276 +246 @@
       byAdaptivewalking = cloner.Clone(original.byAdaptivewalking);
       random = cloner.Clone(original.random);
-      breedingPerformanceModel = cloner.Clone(original.breedingPerformanceModel);
       breedingStat = original.breedingStat.Select(x => Tuple.Create(x.Item1, x.Item2, x.Item3, x.Item4)).ToList();
-      relinkingPerformanceModel = cloner.Clone(original.relinkingPerformanceModel);
       relinkingStat = original.relinkingStat.Select(x => Tuple.Create(x.Item1, x.Item2, x.Item3, x.Item4)).ToList();
-      delinkingPerformanceModel = cloner.Clone(original.delinkingPerformanceModel);
       delinkingStat = original.delinkingStat.Select(x => Tuple.Create(x.Item1, x.Item2, x.Item3, x.Item4)).ToList();
-      samplingPerformanceModel = cloner.Clone(original.samplingPerformanceModel);
       samplingStat = original.samplingStat.Select(x => Tuple.Create(x.Item1, x.Item2)).ToList();
-      hillclimbingPerformanceModel = cloner.Clone(original.hillclimbingPerformanceModel);
       hillclimbingStat = original.hillclimbingStat.Select(x => Tuple.Create(x.Item1, x.Item2)).ToList();
-      adaptiveWalkPerformanceModel = cloner.Clone(original.adaptiveWalkPerformanceModel);
       adaptivewalkingStat = original.adaptivewalkingStat.Select(x => Tuple.Create(x.Item1, x.Item2)).ToList();
       
@@ -349 +313 @@
         breedingStat.Add(Tuple.Create(a.Fitness, b.Fitness, parentDist, child.Fitness));
       else breedingStat.Add(Tuple.Create(b.Fitness, a.Fitness, parentDist, child.Fitness));
-      if (breedingStat.Count % 10 == 0) RelearnBreedingPerformanceModel();
-    }
-    public void RelearnBreedingPerformanceModel() {
-      breedingPerformanceModel = RunRegression(PrepareRegression(ToListRow(breedingStat)), breedingPerformanceModel).Model;
     }
     public bool BreedingSuited(ISingleObjectiveSolutionScope<TSolution> p1, ISingleObjectiveSolutionScope<TSolution> p2, double dist) {
-      if (breedingPerformanceModel == null) return true;
-      double minI1 = double.MaxValue, minI2 = double.MaxValue, maxI1 = double.MinValue, maxI2 = double.MinValue;
-      foreach (var d in BreedingStat) {
-        if (d.Item1 < minI1) minI1 = d.Item1;
-        if (d.Item1 > maxI1) maxI1 = d.Item1;
-        if (d.Item2 < minI2) minI2 = d.Item2;
-        if (d.Item2 > maxI2) maxI2 = d.Item2;
-      }
-      if (p1.Fitness < minI1 || p1.Fitness > maxI1 || p2.Fitness < minI2 || p2.Fitness > maxI2)
-        return true;
-      
-      return Random.NextDouble() < ProbabilityAcceptAbsolutePerformanceModel(new List<double> { p1.Fitness, p2.Fitness, dist }, breedingPerformanceModel);
+      return true;
     }
     #endregion
@@ -375 +324 @@
         relinkingStat.Add(Tuple.Create(a.Fitness, b.Fitness, parentDist, Maximization ? child.Fitness - a.Fitness : a.Fitness - child.Fitness));
       else relinkingStat.Add(Tuple.Create(a.Fitness, b.Fitness, parentDist, Maximization ? child.Fitness - b.Fitness : b.Fitness - child.Fitness));
-      if (relinkingStat.Count % 10 == 0) RelearnRelinkingPerformanceModel();
-    }
-    public void RelearnRelinkingPerformanceModel() {
-      relinkingPerformanceModel = RunRegression(PrepareRegression(ToListRow(relinkingStat)), relinkingPerformanceModel).Model;
     }
     public bool RelinkSuited(ISingleObjectiveSolutionScope<TSolution> p1, ISingleObjectiveSolutionScope<TSolution> p2, double dist) {
-      if (relinkingPerformanceModel == null) return true;
-      double minI1 = double.MaxValue, minI2 = double.MaxValue, maxI1 = double.MinValue, maxI2 = double.MinValue;
-      foreach (var d in RelinkingStat) {
-        if (d.Item1 < minI1) minI1 = d.Item1;
-        if (d.Item1 > maxI1) maxI1 = d.Item1;
-        if (d.Item2 < minI2) minI2 = d.Item2;
-        if (d.Item2 > maxI2) maxI2 = d.Item2;
-      }
-      if (p1.Fitness < minI1 || p1.Fitness > maxI1 || p2.Fitness < minI2 || p2.Fitness > maxI2)
-        return true;
-
-      if (IsBetter(p1, p2)) {
-        return Random.NextDouble() < ProbabilityAcceptRelativePerformanceModel(p1.Fitness, new List<double> { p1.Fitness, p2.Fitness, dist }, relinkingPerformanceModel);
-      }
-      return Random.NextDouble() < ProbabilityAcceptRelativePerformanceModel(p2.Fitness, new List<double> { p1.Fitness, p2.Fitness, dist }, relinkingPerformanceModel);
+      return true;
     }
     #endregion
@@ -404 +335 @@
         delinkingStat.Add(Tuple.Create(a.Fitness, b.Fitness, parentDist, Maximization ? child.Fitness - a.Fitness : a.Fitness - child.Fitness));
       else delinkingStat.Add(Tuple.Create(a.Fitness, b.Fitness, parentDist, Maximization ? child.Fitness - b.Fitness : b.Fitness - child.Fitness));
-      if (delinkingStat.Count % 10 == 0) RelearnDelinkingPerformanceModel();
-    }
-    public void RelearnDelinkingPerformanceModel() {
-      delinkingPerformanceModel = RunRegression(PrepareRegression(ToListRow(delinkingStat)), delinkingPerformanceModel).Model;
     }
     public bool DelinkSuited(ISingleObjectiveSolutionScope<TSolution> p1, ISingleObjectiveSolutionScope<TSolution> p2, double dist) {
-      if (delinkingPerformanceModel == null) return true;
-      double minI1 = double.MaxValue, minI2 = double.MaxValue, maxI1 = double.MinValue, maxI2 = double.MinValue;
-      foreach (var d in DelinkingStat) {
-        if (d.Item1 < minI1) minI1 = d.Item1;
-        if (d.Item1 > maxI1) maxI1 = d.Item1;
-        if (d.Item2 < minI2) minI2 = d.Item2;
-        if (d.Item2 > maxI2) maxI2 = d.Item2;
-      }
-      if (p1.Fitness < minI1 || p1.Fitness > maxI1 || p2.Fitness < minI2 || p2.Fitness > maxI2)
-        return true;
-      if (IsBetter(p1, p2)) {
-        return Random.NextDouble() < ProbabilityAcceptRelativePerformanceModel(p1.Fitness, new List<double> { p1.Fitness, p2.Fitness, dist }, delinkingPerformanceModel);
-      }
-      return Random.NextDouble() < ProbabilityAcceptRelativePerformanceModel(p2.Fitness, new List<double> { p1.Fitness, p2.Fitness, dist }, delinkingPerformanceModel);
+      return true;
     }
     #endregion
@@ -430 +344 @@
     public void AddSamplingResult(ISingleObjectiveSolutionScope<TSolution> sample, double avgDist) {
       samplingStat.Add(Tuple.Create(avgDist, sample.Fitness));
-      if (samplingStat.Count % 10 == 0) RelearnSamplingPerformanceModel();
-    }
-    public void RelearnSamplingPerformanceModel() {
-      samplingPerformanceModel = RunRegression(PrepareRegression(ToListRow(samplingStat)), samplingPerformanceModel).Model;
     }
     public bool SamplingSuited(double avgDist) {
-      if (samplingPerformanceModel == null) return true;
-      if (avgDist < samplingStat.Min(x => x.Item1) || avgDist > samplingStat.Max(x => x.Item1)) return true;
-      return Random.NextDouble() < ProbabilityAcceptAbsolutePerformanceModel(new List<double> { avgDist }, samplingPerformanceModel);
+      return true;
     }
     #endregion
@@ -445 +353 @@
     public void AddHillclimbingResult(ISingleObjectiveSolutionScope<TSolution> input, ISingleObjectiveSolutionScope<TSolution> outcome) {
       hillclimbingStat.Add(Tuple.Create(input.Fitness, Maximization ? outcome.Fitness - input.Fitness : input.Fitness - outcome.Fitness));
-      if (hillclimbingStat.Count % 10 == 0) RelearnHillclimbingPerformanceModel();
-    }
-    public void RelearnHillclimbingPerformanceModel() {
-      hillclimbingPerformanceModel = RunRegression(PrepareRegression(ToListRow(hillclimbingStat)), hillclimbingPerformanceModel).Model;
     }
     public bool HillclimbingSuited(double startingFitness) {
-      if (hillclimbingPerformanceModel == null) return true;
-      if (startingFitness < HillclimbingStat.Min(x => x.Item1) || startingFitness > HillclimbingStat.Max(x => x.Item1))
-        return true;
-      return Random.NextDouble() < ProbabilityAcceptRelativePerformanceModel(startingFitness, new List<double> { startingFitness }, hillclimbingPerformanceModel);
+      return true;
     }
     #endregion
@@ -461 +362 @@
     public void AddAdaptivewalkingResult(ISingleObjectiveSolutionScope<TSolution> input, ISingleObjectiveSolutionScope<TSolution> outcome) {
       adaptivewalkingStat.Add(Tuple.Create(input.Fitness, Maximization ? outcome.Fitness - input.Fitness : input.Fitness - outcome.Fitness));
-      if (adaptivewalkingStat.Count % 10 == 0) RelearnAdaptiveWalkPerformanceModel();
-    }
-    public void RelearnAdaptiveWalkPerformanceModel() {
-      adaptiveWalkPerformanceModel = RunRegression(PrepareRegression(ToListRow(adaptivewalkingStat)), adaptiveWalkPerformanceModel).Model;
     }
     public bool AdaptivewalkingSuited(double startingFitness) {
-      if (adaptiveWalkPerformanceModel == null) return true;
-      if (startingFitness < AdaptivewalkingStat.Min(x => x.Item1) || startingFitness > AdaptivewalkingStat.Max(x => x.Item1))
-        return true;
-      return Random.NextDouble() < ProbabilityAcceptRelativePerformanceModel(startingFitness, new List<double> { startingFitness }, adaptiveWalkPerformanceModel);
-    }
-    #endregion
-
-    public IConfidenceRegressionSolution GetSolution(IConfidenceRegressionModel model, IEnumerable<Tuple<double, double>> data) {
-      return new ConfidenceRegressionSolution(model, PrepareRegression(ToListRow(data.ToList())));
-    }
-    public IConfidenceRegressionSolution GetSolution(IConfidenceRegressionModel model, IEnumerable<Tuple<double, double, double>> data) {
-      return new ConfidenceRegressionSolution(model, PrepareRegression(ToListRow(data.ToList())));
-    }
-    public IConfidenceRegressionSolution GetSolution(IConfidenceRegressionModel model, IEnumerable<Tuple<double, double, double, double>> data) {
-      return new ConfidenceRegressionSolution(model, PrepareRegression(ToListRow(data.ToList())));
-    }
-
-    protected RegressionProblemData PrepareRegression(List<List<double>> data) {
-      var columns = data.First().Select(y => new List<double>()).ToList();
-      foreach (var next in data.Shuffle(Random)) {
-        for (var i = 0; i < next.Count; i++) {
-          columns[i].Add(next[i]);
-        }
-      }
-      var ds = new Dataset(columns.Select((v, i) => i < columns.Count - 1 ? "in" + i : "out").ToList(), columns);
-      var regPrb = new RegressionProblemData(ds, Enumerable.Range(0, columns.Count - 1).Select(x => "in" + x), "out") {
-        TrainingPartition = { Start = 0, End = Math.Min(50, data.Count) },
-        TestPartition = { Start = Math.Min(50, data.Count), End = data.Count }
-      };
-      return regPrb;
-    }
-
-    protected static IConfidenceRegressionSolution RunRegression(RegressionProblemData trainingData, IConfidenceRegressionModel baseLineModel = null) {
-      var targetValues = trainingData.Dataset.GetDoubleValues(trainingData.TargetVariable, trainingData.TrainingIndices).ToList();
-      var baseline = baseLineModel != null ? new ConfidenceRegressionSolution(baseLineModel, trainingData) : null;
-      var constantSolution = new ConfidenceRegressionSolution(new ConfidenceConstantModel(targetValues.Average(), targetValues.Variance(), trainingData.TargetVariable), trainingData);
-      var gpr = new GaussianProcessRegression { Problem = { ProblemData = trainingData } };
-      if (trainingData.InputVariables.CheckedItems.Any(x => alglib.pearsoncorr2(trainingData.Dataset.GetDoubleValues(x.Value.Value).ToArray(), trainingData.TargetVariableValues.ToArray()) > 0.8)) {
-        gpr.MeanFunction = new MeanZero();
-        var cov1 = new CovarianceSum();
-        cov1.Terms.Add(new CovarianceLinearArd());
-        cov1.Terms.Add(new CovarianceConst());
-        gpr.CovarianceFunction = cov1;
-      }
-      IConfidenceRegressionSolution solution = null;
-      var cnt = 0;
-      do {
-        ExecuteAlgorithm(gpr);
-        solution = (IConfidenceRegressionSolution)gpr.Results["Solution"].Value;
-        cnt++;
-      } while (cnt < 10 && (solution == null || solution.TrainingRSquared.IsAlmost(0)));
-
-      return GetBestRegressionSolution(constantSolution, baseline, solution);
-    }
-
-    private static IConfidenceRegressionSolution GetBestRegressionSolution(IConfidenceRegressionSolution constant, IConfidenceRegressionSolution baseline, IConfidenceRegressionSolution solution) {
-      if (baseline == null)
-        return constant.TrainingMeanAbsoluteError < solution.TrainingMeanAbsoluteError ? constant : solution;
-
-      double a, b, c;
-      if (constant.ProblemData.Dataset.Rows < 60) {
-        c = constant.TrainingMeanAbsoluteError;
-        b = baseline.TrainingMeanAbsoluteError;
-        a = solution.TrainingMeanAbsoluteError;
-      } else {
-        c = constant.TestMeanAbsoluteError;
-        b = baseline.TestMeanAbsoluteError;
-        a = solution.TestMeanAbsoluteError;
-      }
-      if (c < b && (c < a || b < a)) return constant;
-      if (b < c && (b < a || c < a)) return baseline;
-      return solution;
-    }
-
-    protected static void ExecuteAlgorithm(IAlgorithm algorithm) {
-      using (var evt = new AutoResetEvent(false)) {
-        EventHandler exeStateChanged = (o, args) => {
-          if (algorithm.ExecutionState != ExecutionState.Started)
-            evt.Set();
-        };
-        algorithm.ExecutionStateChanged += exeStateChanged;
-        if (algorithm.ExecutionState != ExecutionState.Prepared) {
-          algorithm.Prepare(true);
-          evt.WaitOne();
-        }
-        algorithm.Start();
-        evt.WaitOne();
-        algorithm.ExecutionStateChanged -= exeStateChanged;
-      }
-    }
-
-    private double ProbabilityAcceptAbsolutePerformanceModel(List<double> inputs, IConfidenceRegressionModel model) {
-      var inputVariables = inputs.Select((v, i) => "in" + i);
-      var ds = new Dataset(inputVariables.Concat( new [] { "out" }), inputs.Select(x => new List<double> { x }).Concat(new [] { new List<double> { double.NaN } }));
-      var mean = model.GetEstimatedValues(ds, new[] { 0 }).Single();
-      var sdev = Math.Sqrt(model.GetEstimatedVariances(ds, new[] { 0 }).Single());
-
-      // calculate the fitness goal
-      var goal = Maximization ? Population.Min(x => x.Fitness) : Population.Max(x => x.Fitness);
-      var z = (goal - mean) / sdev;
-      // return the probability of achieving or surpassing that goal
-      var y = alglib.invnormaldistribution(z);
-      return Maximization ? 1.0 - y /* P(X >= z) */ : y; // P(X <= z)
-    }
-
-    private double ProbabilityAcceptRelativePerformanceModel(double basePerformance, List<double> inputs, IConfidenceRegressionModel model) {
-      var inputVariables = inputs.Select((v, i) => "in" + i);
-      var ds = new Dataset(inputVariables.Concat(new[] { "out" }), inputs.Select(x => new List<double> { x }).Concat(new[] { new List<double> { double.NaN } }));
-      var mean = model.GetEstimatedValues(ds, new[] { 0 }).Single();
-      var sdev = Math.Sqrt(model.GetEstimatedVariances(ds, new[] { 0 }).Single());
-
-      // calculate the improvement goal
-      var goal = Maximization ? Population.Min(x => x.Fitness) - basePerformance : basePerformance - Population.Max(x => x.Fitness);
-      var z = (goal - mean) / sdev;
-      // return the probability of achieving or surpassing that goal
-      return 1.0 - alglib.invnormaldistribution(z); /* P(X >= z) */
-    }
-
-    private static List<List<double>> ToListRow(List<Tuple<double, double>> rows) {
-      return rows.Select(x => new List<double> { x.Item1, x.Item2 }).ToList();
-    }
-    private static List<List<double>> ToListRow(List<Tuple<double, double, double>> rows) {
-      return rows.Select(x => new List<double> { x.Item1, x.Item2, x.Item3 }).ToList();
-    }
-    private static List<List<double>> ToListRow(List<Tuple<double, double, double, double>> rows) {
-      return rows.Select(x => new List<double> { x.Item1, x.Item2, x.Item3, x.Item4 }).ToList();
-    }
+      return true;
+    }
+    #endregion
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]