Changeset 13976 for stable/HeuristicLab.Problems.DataAnalysis/3.4/Implementation

Timestamp:

07/02/16 09:02:09 (8 years ago)

Author:

gkronber

Message:

#2590 merged r13697:13698, r13700:13702, r13704:13705, r13711, r13715 from trunk to stable

Location:

stable

Files:

• . (modified) (1 prop)
• HeuristicLab.Problems.DataAnalysis (modified) (1 prop)
• HeuristicLab.Problems.DataAnalysis/3.4/Implementation/Regression/RegressionEnsembleModel.cs (modified) (5 diffs)
• HeuristicLab.Problems.DataAnalysis/3.4/Implementation/Regression/RegressionEnsembleSolution.cs (modified) (9 diffs)

stable/HeuristicLab.Problems.DataAnalysis/3.4/Implementation/Regression/RegressionEnsembleModel.cs

@@ -20 +20 @@
 #endregion
 
+using System;
 using System.Collections.Generic;
 using System.Linq;
@@ -32 +33 @@
   [StorableClass]
   [Item("RegressionEnsembleModel", "A regression model that contains an ensemble of multiple regression models")]
-  public class RegressionEnsembleModel : NamedItem, IRegressionEnsembleModel {
+  public sealed class RegressionEnsembleModel : NamedItem, IRegressionEnsembleModel {
 
     private List<IRegressionModel> models;
@@ -45 +46 @@
     }
 
+    private List<double> modelWeights;
+    public IEnumerable<double> ModelWeights {
+      get { return modelWeights; }
+    }
+
+    [Storable(Name = "ModelWeights")]
+    private IEnumerable<double> StorableModelWeights {
+      get { return modelWeights; }
+      set { modelWeights = value.ToList(); }
+    }
+
+    [Storable]
+    private bool averageModelEstimates = true;
+    public bool AverageModelEstimates {
+      get { return averageModelEstimates; }
+      set {
+        if (averageModelEstimates != value) {
+          averageModelEstimates = value;
+          OnChanged();
+        }
+      }
+    }
+
     #region backwards compatiblity 3.3.5
     [Storable(Name = "models", AllowOneWay = true)]
@@ -52 +76 @@
     #endregion
 
+    [StorableHook(HookType.AfterDeserialization)]
+    private void AfterDeserialization() {
+      // BackwardsCompatibility 3.3.14
+      #region Backwards compatible code, remove with 3.4
+      if (modelWeights == null || !modelWeights.Any())
+        modelWeights = new List<double>(models.Select(m => 1.0));
+      #endregion
+    }
+
     [StorableConstructor]
-    protected RegressionEnsembleModel(bool deserializing) : base(deserializing) { }
-    protected RegressionEnsembleModel(RegressionEnsembleModel original, Cloner cloner)
+    private RegressionEnsembleModel(bool deserializing) : base(deserializing) { }
+    private RegressionEnsembleModel(RegressionEnsembleModel original, Cloner cloner)
       : base(original, cloner) {
-      this.models = original.Models.Select(m => cloner.Clone(m)).ToList();
+      this.models = original.Models.Select(cloner.Clone).ToList();
+      this.modelWeights = new List<double>(original.ModelWeights);
+      this.averageModelEstimates = original.averageModelEstimates;
+    }
+    public override IDeepCloneable Clone(Cloner cloner) {
+      return new RegressionEnsembleModel(this, cloner);
     }
 
     public RegressionEnsembleModel() : this(Enumerable.Empty<IRegressionModel>()) { }
-    public RegressionEnsembleModel(IEnumerable<IRegressionModel> models)
+    public RegressionEnsembleModel(IEnumerable<IRegressionModel> models) : this(models, models.Select(m => 1.0)) { }
+    public RegressionEnsembleModel(IEnumerable<IRegressionModel> models, IEnumerable<double> modelWeights)
       : base() {
       this.name = ItemName;
       this.description = ItemDescription;
+
+
       this.models = new List<IRegressionModel>(models);
-    }
-
-    public override IDeepCloneable Clone(Cloner cloner) {
-      return new RegressionEnsembleModel(this, cloner);
-    }
-
-    #region IRegressionEnsembleModel Members
+      this.modelWeights = new List<double>(modelWeights);
+    }
 
     public void Add(IRegressionModel model) {
+      Add(model, 1.0);
+    }
+    public void Add(IRegressionModel model, double weight) {
       models.Add(model);
-    }
+      modelWeights.Add(weight);
+      OnChanged();
+    }
+
+    public void AddRange(IEnumerable<IRegressionModel> models) {
+      AddRange(models, models.Select(m => 1.0));
+    }
+    public void AddRange(IEnumerable<IRegressionModel> models, IEnumerable<double> weights) {
+      this.models.AddRange(models);
+      modelWeights.AddRange(weights);
+      OnChanged();
+    }
+
     public void Remove(IRegressionModel model) {
-      models.Remove(model);
-    }
-
+      var index = models.IndexOf(model);
+      models.RemoveAt(index);
+      modelWeights.RemoveAt(index);
+      OnChanged();
+    }
+    public void RemoveRange(IEnumerable<IRegressionModel> models) {
+      foreach (var model in models) {
+        var index = this.models.IndexOf(model);
+        this.models.RemoveAt(index);
+        modelWeights.RemoveAt(index);
+      }
+      OnChanged();
+    }
+
+    public double GetModelWeight(IRegressionModel model) {
+      var index = models.IndexOf(model);
+      return modelWeights[index];
+    }
+    public void SetModelWeight(IRegressionModel model, double weight) {
+      var index = models.IndexOf(model);
+      modelWeights[index] = weight;
+      OnChanged();
+    }
+
+    #region evaluation
     public IEnumerable<IEnumerable<double>> GetEstimatedValueVectors(IDataset dataset, IEnumerable<int> rows) {
       var estimatedValuesEnumerators = (from model in models
-                                        select model.GetEstimatedValues(dataset, rows).GetEnumerator())
-                                       .ToList();
+                                        let weight = GetModelWeight(model)
+                                        select model.GetEstimatedValues(dataset, rows).Select(e => weight * e)
+                                        .GetEnumerator()).ToList();
 
       while (estimatedValuesEnumerators.All(en => en.MoveNext())) {
@@ -91 +165 @@
     }
 
+    public IEnumerable<double> GetEstimatedValues(IDataset dataset, IEnumerable<int> rows) {
+      double weightsSum = modelWeights.Sum();
+      var summedEstimates = from estimatedValuesVector in GetEstimatedValueVectors(dataset, rows)
+                            select estimatedValuesVector.DefaultIfEmpty(double.NaN).Sum();
+
+      if (AverageModelEstimates)
+        return summedEstimates.Select(v => v / weightsSum);
+      else
+        return summedEstimates;
+
+    }
+
+    public IEnumerable<double> GetEstimatedValues(IDataset dataset, IEnumerable<int> rows, Func<int, IRegressionModel, bool> modelSelectionPredicate) {
+      var estimatedValuesEnumerators = GetEstimatedValueVectors(dataset, rows).GetEnumerator();
+      var rowsEnumerator = rows.GetEnumerator();
+
+      while (rowsEnumerator.MoveNext() & estimatedValuesEnumerators.MoveNext()) {
+        var estimatedValueEnumerator = estimatedValuesEnumerators.Current.GetEnumerator();
+        int currentRow = rowsEnumerator.Current;
+        double weightsSum = 0.0;
+        double filteredEstimatesSum = 0.0;
+
+        for (int m = 0; m < models.Count; m++) {
+          estimatedValueEnumerator.MoveNext();
+          var model = models[m];
+          if (!modelSelectionPredicate(currentRow, model)) continue;
+
+          filteredEstimatesSum += estimatedValueEnumerator.Current;
+          weightsSum += modelWeights[m];
+        }
+
+        if (AverageModelEstimates)
+          yield return filteredEstimatesSum / weightsSum;
+        else
+          yield return filteredEstimatesSum;
+      }
+    }
+
     #endregion
 
-    #region IRegressionModel Members
-
-    public IEnumerable<double> GetEstimatedValues(IDataset dataset, IEnumerable<int> rows) {
-      foreach (var estimatedValuesVector in GetEstimatedValueVectors(dataset, rows)) {
-        yield return estimatedValuesVector.Average();
-      }
-    }
+    public event EventHandler Changed;
+    private void OnChanged() {
+      var handler = Changed;
+      if (handler != null)
+        handler(this, EventArgs.Empty);
+    }
+
 
     public RegressionEnsembleSolution CreateRegressionSolution(IRegressionProblemData problemData) {
-      return new RegressionEnsembleSolution(this.Models, new RegressionEnsembleProblemData(problemData));
+      return new RegressionEnsembleSolution(this, new RegressionEnsembleProblemData(problemData));
     }
     IRegressionSolution IRegressionModel.CreateRegressionSolution(IRegressionProblemData problemData) {
       return CreateRegressionSolution(problemData);
     }
-
-    #endregion
   }
 }

stable/HeuristicLab.Problems.DataAnalysis/3.4/Implementation/Regression/RegressionEnsembleSolution.cs

@@ -79 +79 @@
         }
       }
+
+      RegisterModelEvents();
       RegisterRegressionSolutionsEventHandler();
     }
@@ -93 +95 @@
       }
 
+      evaluationCache = new Dictionary<int, double>(original.ProblemData.Dataset.Rows);
       trainingEvaluationCache = new Dictionary<int, double>(original.ProblemData.TrainingIndices.Count());
       testEvaluationCache = new Dictionary<int, double>(original.ProblemData.TestIndices.Count());
 
       regressionSolutions = cloner.Clone(original.regressionSolutions);
+      RegisterModelEvents();
       RegisterRegressionSolutionsEventHandler();
     }
@@ -106 +110 @@
       regressionSolutions = new ItemCollection<IRegressionSolution>();
 
+      RegisterModelEvents();
       RegisterRegressionSolutionsEventHandler();
     }
 
     public RegressionEnsembleSolution(IRegressionProblemData problemData)
-      : this(Enumerable.Empty<IRegressionModel>(), problemData) {
-    }
-
-    public RegressionEnsembleSolution(IEnumerable<IRegressionModel> models, IRegressionProblemData problemData)
-      : this(models, problemData,
-             models.Select(m => (IntRange)problemData.TrainingPartition.Clone()),
-             models.Select(m => (IntRange)problemData.TestPartition.Clone())
-      ) { }
-
-    public RegressionEnsembleSolution(IEnumerable<IRegressionModel> models, IRegressionProblemData problemData, IEnumerable<IntRange> trainingPartitions, IEnumerable<IntRange> testPartitions)
-      : base(new RegressionEnsembleModel(Enumerable.Empty<IRegressionModel>()), new RegressionEnsembleProblemData(problemData)) {
-      this.trainingPartitions = new Dictionary<IRegressionModel, IntRange>();
-      this.testPartitions = new Dictionary<IRegressionModel, IntRange>();
-      this.regressionSolutions = new ItemCollection<IRegressionSolution>();
-
-      List<IRegressionSolution> solutions = new List<IRegressionSolution>();
-      var modelEnumerator = models.GetEnumerator();
-      var trainingPartitionEnumerator = trainingPartitions.GetEnumerator();
-      var testPartitionEnumerator = testPartitions.GetEnumerator();
-
-      while (modelEnumerator.MoveNext() & trainingPartitionEnumerator.MoveNext() & testPartitionEnumerator.MoveNext()) {
-        var p = (IRegressionProblemData)problemData.Clone();
-        p.TrainingPartition.Start = trainingPartitionEnumerator.Current.Start;
-        p.TrainingPartition.End = trainingPartitionEnumerator.Current.End;
-        p.TestPartition.Start = testPartitionEnumerator.Current.Start;
-        p.TestPartition.End = testPartitionEnumerator.Current.End;
-
-        solutions.Add(modelEnumerator.Current.CreateRegressionSolution(p));
-      }
-      if (modelEnumerator.MoveNext() | trainingPartitionEnumerator.MoveNext() | testPartitionEnumerator.MoveNext()) {
-        throw new ArgumentException();
-      }
-
+      : this(new RegressionEnsembleModel(), problemData) {
+    }
+
+    public RegressionEnsembleSolution(IRegressionEnsembleModel model, IRegressionProblemData problemData)
+      : base(model, new RegressionEnsembleProblemData(problemData)) {
+      trainingPartitions = new Dictionary<IRegressionModel, IntRange>();
+      testPartitions = new Dictionary<IRegressionModel, IntRange>();
+      regressionSolutions = new ItemCollection<IRegressionSolution>();
+
+      evaluationCache = new Dictionary<int, double>(problemData.Dataset.Rows);
       trainingEvaluationCache = new Dictionary<int, double>(problemData.TrainingIndices.Count());
       testEvaluationCache = new Dictionary<int, double>(problemData.TestIndices.Count());
 
+
+      var solutions = model.Models.Select(m => m.CreateRegressionSolution((IRegressionProblemData)problemData.Clone()));
+      foreach (var solution in solutions) {
+        regressionSolutions.Add(solution);
+        trainingPartitions.Add(solution.Model, solution.ProblemData.TrainingPartition);
+        testPartitions.Add(solution.Model, solution.ProblemData.TestPartition);
+      }
+
+      RecalculateResults();
+      RegisterModelEvents();
       RegisterRegressionSolutionsEventHandler();
-      regressionSolutions.AddRange(solutions);
-    }
+    }
+
 
     public override IDeepCloneable Clone(Cloner cloner) {
       return new RegressionEnsembleSolution(this, cloner);
+    }
+
+    private void RegisterModelEvents() {
+      Model.Changed += Model_Changed;
     }
     private void RegisterRegressionSolutionsEventHandler() {
@@ -168 +164 @@
         var rows = ProblemData.TrainingIndices;
         var rowsToEvaluate = rows.Except(trainingEvaluationCache.Keys);
+
         var rowsEnumerator = rowsToEvaluate.GetEnumerator();
-        var valuesEnumerator = GetEstimatedValues(rowsToEvaluate, (r, m) => RowIsTrainingForModel(r, m) && !RowIsTestForModel(r, m)).GetEnumerator();
+        var valuesEnumerator = Model.GetEstimatedValues(ProblemData.Dataset, rowsToEvaluate, (r, m) => RowIsTrainingForModel(r, m) && !RowIsTestForModel(r, m)).GetEnumerator();
 
         while (rowsEnumerator.MoveNext() & valuesEnumerator.MoveNext()) {
@@ -184 +181 @@
         var rowsToEvaluate = rows.Except(testEvaluationCache.Keys);
         var rowsEnumerator = rowsToEvaluate.GetEnumerator();
-        var valuesEnumerator = GetEstimatedValues(rowsToEvaluate, RowIsTestForModel).GetEnumerator();
+        var valuesEnumerator = Model.GetEstimatedValues(ProblemData.Dataset, rowsToEvaluate, RowIsTestForModel).GetEnumerator();
 
         while (rowsEnumerator.MoveNext() & valuesEnumerator.MoveNext()) {
@@ -193 +190 @@
       }
     }
-
-    private IEnumerable<double> GetEstimatedValues(IEnumerable<int> rows, Func<int, IRegressionModel, bool> modelSelectionPredicate) {
-      var estimatedValuesEnumerators = (from model in Model.Models
-                                        select new { Model = model, EstimatedValuesEnumerator = model.GetEstimatedValues(ProblemData.Dataset, rows).GetEnumerator() })
-                                       .ToList();
-      var rowsEnumerator = rows.GetEnumerator();
-      // aggregate to make sure that MoveNext is called for all enumerators 
-      while (rowsEnumerator.MoveNext() & estimatedValuesEnumerators.Select(en => en.EstimatedValuesEnumerator.MoveNext()).Aggregate(true, (acc, b) => acc & b)) {
-        int currentRow = rowsEnumerator.Current;
-
-        var selectedEnumerators = from pair in estimatedValuesEnumerators
-                                  where modelSelectionPredicate(currentRow, pair.Model)
-                                  select pair.EstimatedValuesEnumerator;
-
-        yield return AggregateEstimatedValues(selectedEnumerators.Select(x => x.Current));
-      }
-    }
-
     private bool RowIsTrainingForModel(int currentRow, IRegressionModel model) {
       return trainingPartitions == null || !trainingPartitions.ContainsKey(model) ||
               (trainingPartitions[model].Start <= currentRow && currentRow < trainingPartitions[model].End);
     }
-
     private bool RowIsTestForModel(int currentRow, IRegressionModel model) {
       return testPartitions == null || !testPartitions.ContainsKey(model) ||
@@ -224 +202 @@
       var rowsToEvaluate = rows.Except(evaluationCache.Keys);
       var rowsEnumerator = rowsToEvaluate.GetEnumerator();
-      var valuesEnumerator = (from xs in GetEstimatedValueVectors(ProblemData.Dataset, rowsToEvaluate)
-                              select AggregateEstimatedValues(xs))
-                             .GetEnumerator();
+      var valuesEnumerator = Model.GetEstimatedValues(ProblemData.Dataset, rowsToEvaluate).GetEnumerator();
 
       while (rowsEnumerator.MoveNext() & valuesEnumerator.MoveNext()) {
@@ -235 +211 @@
     }
 
-    public IEnumerable<IEnumerable<double>> GetEstimatedValueVectors(IDataset dataset, IEnumerable<int> rows) {
-      if (!Model.Models.Any()) yield break;
-      var estimatedValuesEnumerators = (from model in Model.Models
-                                        select model.GetEstimatedValues(dataset, rows).GetEnumerator())
-                                       .ToList();
-
-      while (estimatedValuesEnumerators.All(en => en.MoveNext())) {
-        yield return from enumerator in estimatedValuesEnumerators
-                     select enumerator.Current;
-      }
-    }
-
-    private double AggregateEstimatedValues(IEnumerable<double> estimatedValues) {
-      return estimatedValues.DefaultIfEmpty(double.NaN).Average();
+    public IEnumerable<IEnumerable<double>> GetEstimatedValueVectors(IEnumerable<int> rows) {
+      return Model.GetEstimatedValueVectors(ProblemData.Dataset, rows);
     }
     #endregion
@@ -282 +246 @@
     }
 
-    public void AddRegressionSolutions(IEnumerable<IRegressionSolution> solutions) {
-      regressionSolutions.AddRange(solutions);
+    private void Model_Changed(object sender, EventArgs e) {
+      var modelSet = new HashSet<IRegressionModel>(Model.Models);
+      foreach (var model in Model.Models) {
+        if (!trainingPartitions.ContainsKey(model)) trainingPartitions.Add(model, ProblemData.TrainingPartition);
+        if (!testPartitions.ContainsKey(model)) testPartitions.Add(model, ProblemData.TrainingPartition);
+      }
+      foreach (var model in trainingPartitions.Keys) {
+        if (modelSet.Contains(model)) continue;
+        trainingPartitions.Remove(model);
+        testPartitions.Remove(model);
+      }
 
       trainingEvaluationCache.Clear();
       testEvaluationCache.Clear();
       evaluationCache.Clear();
+
+      OnModelChanged();
+    }
+
+    public void AddRegressionSolutions(IEnumerable<IRegressionSolution> solutions) {
+      regressionSolutions.AddRange(solutions);
     }
     public void RemoveRegressionSolutions(IEnumerable<IRegressionSolution> solutions) {
       regressionSolutions.RemoveRange(solutions);
-
-      trainingEvaluationCache.Clear();
-      testEvaluationCache.Clear();
-      evaluationCache.Clear();
     }
 
     private void regressionSolutions_ItemsAdded(object sender, CollectionItemsChangedEventArgs<IRegressionSolution> e) {
-      foreach (var solution in e.Items) AddRegressionSolution(solution);
-      RecalculateResults();
+      foreach (var solution in e.Items) {
+        trainingPartitions.Add(solution.Model, solution.ProblemData.TrainingPartition);
+        testPartitions.Add(solution.Model, solution.ProblemData.TestPartition);
+      }
+      Model.AddRange(e.Items.Select(s => s.Model));
     }
     private void regressionSolutions_ItemsRemoved(object sender, CollectionItemsChangedEventArgs<IRegressionSolution> e) {
-      foreach (var solution in e.Items) RemoveRegressionSolution(solution);
-      RecalculateResults();
+      foreach (var solution in e.Items) {
+        trainingPartitions.Remove(solution.Model);
+        testPartitions.Remove(solution.Model);
+      }
+      Model.RemoveRange(e.Items.Select(s => s.Model));
     }
     private void regressionSolutions_CollectionReset(object sender, CollectionItemsChangedEventArgs<IRegressionSolution> e) {
-      foreach (var solution in e.OldItems) RemoveRegressionSolution(solution);
-      foreach (var solution in e.Items) AddRegressionSolution(solution);
-      RecalculateResults();
-    }
-
-    private void AddRegressionSolution(IRegressionSolution solution) {
-      if (Model.Models.Contains(solution.Model)) throw new ArgumentException();
-      Model.Add(solution.Model);
-      trainingPartitions[solution.Model] = solution.ProblemData.TrainingPartition;
-      testPartitions[solution.Model] = solution.ProblemData.TestPartition;
-
-      trainingEvaluationCache.Clear();
-      testEvaluationCache.Clear();
-      evaluationCache.Clear();
-    }
-
-    private void RemoveRegressionSolution(IRegressionSolution solution) {
-      if (!Model.Models.Contains(solution.Model)) throw new ArgumentException();
-      Model.Remove(solution.Model);
-      trainingPartitions.Remove(solution.Model);
-      testPartitions.Remove(solution.Model);
-
-      trainingEvaluationCache.Clear();
-      testEvaluationCache.Clear();
-      evaluationCache.Clear();
+      foreach (var solution in e.OldItems) {
+        trainingPartitions.Remove(solution.Model);
+        testPartitions.Remove(solution.Model);
+      }
+      Model.RemoveRange(e.OldItems.Select(s => s.Model));
+
+      foreach (var solution in e.Items) {
+        trainingPartitions.Add(solution.Model, solution.ProblemData.TrainingPartition);
+        testPartitions.Add(solution.Model, solution.ProblemData.TestPartition);
+      }
+      Model.AddRange(e.Items.Select(s => s.Model));
     }
   }