Changeset 15531

Timestamp:

12/18/17 12:27:14 (6 years ago)

Author:

bwerth

Message:

#2850 added named weights and automatic adaption thereof

Location:

branches/Weighted TSNE/3.4/TSNE

Files:

• Distances/WeightedEuclideanDistance.cs (modified) (6 diffs)
• TSNEAlgorithm.cs (modified) (7 diffs)

branches/Weighted TSNE/3.4/TSNE/Distances/WeightedEuclideanDistance.cs

@@ -29 +29 @@
 using HeuristicLab.Parameters;
 using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
+using HeuristicLab.Problems.DataAnalysis;
 
 namespace HeuristicLab.Algorithms.DataAnalysis {
@@ -34 +35 @@
   [Item("WeightedEuclideanDistance", "A weighted norm function that uses Euclidean distance √(Σ(w[i]²*(p1[i]-p2[i])²))")]
   public class WeightedEuclideanDistance : ParameterizedNamedItem, IDistance<IEnumerable<double>> {
+    [Storable]
+    private double[] weights;
     public const string WeightsParameterName = "Weights";
-    public IValueParameter<DoubleArray> WeigthsParameter {
+    public IValueParameter<DoubleArray> WeightsParameter {
       get { return (IValueParameter<DoubleArray>) Parameters[WeightsParameterName]; }
     }
 
     public DoubleArray Weights {
-      get { return WeigthsParameter.Value; }
-      set { WeigthsParameter.Value = value; }
+      get { return WeightsParameter.Value; }
+      set { WeightsParameter.Value = value; }
     }
 
@@ -52 +55 @@
     protected WeightedEuclideanDistance(WeightedEuclideanDistance original, Cloner cloner) : base(original, cloner) {
       RegisterParameterEvents();
+      weights = original.weights != null ? original.weights.ToArray() : null;
     }
     public override IDeepCloneable Clone(Cloner cloner) {
@@ -57 +61 @@
     }
     public WeightedEuclideanDistance() {
-      Parameters.Add(new ValueParameter<DoubleArray>(WeightsParameterName, "The weights used to modify the euclidean distance."));
+      Parameters.Add(new ValueParameter<DoubleArray>(WeightsParameterName, "The weights used to modify the euclidean distance.", new DoubleArray(new[] {1.0})));
       RegisterParameterEvents();
     }
@@ -76 +80 @@
 
     public double Get(IEnumerable<double> a, IEnumerable<double> b) {
-      return GetDistance(a, b, Weights);
+      return GetDistance(a, b, weights);
     }
     public IComparer<IEnumerable<double>> GetDistanceComparer(IEnumerable<double> item) {
@@ -88 +92 @@
     }
 
+    public void AdaptToProblemData(IDataAnalysisProblemData problemData) {
+      Weights = new DoubleArray(problemData.AllowedInputVariables.Select(v => Weights.ElementNames.Contains(v) ? GetWeight(v) : 1).ToArray())
+        {ElementNames = problemData.AllowedInputVariables};
+    }
+    public void Initialize(IDataAnalysisProblemData problemData) {
+      if (Weights.Length != problemData.AllowedInputVariables.Count()) throw new ArgumentException("Number of Weights does not match the number of input variables");
+      weights = Weights.ElementNames.All(v => v == null || v.Equals(string.Empty)) ? 
+        Weights.ToArray() : 
+        problemData.AllowedInputVariables.Select(GetWeight).ToArray();
+    }
+    private double GetWeight(string v) {
+      var w = Weights;
+      var names = w.ElementNames.ToArray();
+      for (var i = 0; i < w.Length; i++) if (names[i].Equals(v)) return w[i];
+      throw new ArgumentException("weigth for " + v + " was requested but not specified.");
+    }
     private void RegisterParameterEvents() {
-      WeigthsParameter.ValueChanged += OnWeightsArrayChanged;
-      WeigthsParameter.Value.ItemChanged += OnWeightChanged;
+      WeightsParameter.ValueChanged += OnWeightsArrayChanged;
+      WeightsParameter.Value.ItemChanged += OnWeightChanged;
     }
     private void OnWeightChanged(object sender, EventArgs<int> e) {
-      WeigthsParameter.Value.ItemChanged -= OnWeightChanged;
+      WeightsParameter.Value.ItemChanged -= OnWeightChanged;
       Weights[e.Value] = Math.Max(0, Weights[e.Value]);
-      WeigthsParameter.Value.ItemChanged -= OnWeightChanged;
+      WeightsParameter.Value.ItemChanged -= OnWeightChanged;
     }
     private void OnWeightsArrayChanged(object sender, EventArgs e) {
-      for (int i = 0; i < Weights.Length; i++)
+      for (var i = 0; i < Weights.Length; i++)
         Weights[i] = Math.Max(0, Weights[i]);
-      WeigthsParameter.Value.ItemChanged += OnWeightChanged;
+      WeightsParameter.Value.ItemChanged += OnWeightChanged;
     }
   }

branches/Weighted TSNE/3.4/TSNE/TSNEAlgorithm.cs

@@ -195 +195 @@
       set { RandomInitializationParameter.Value.Value = value; }
     }
-
     public int UpdateInterval {
       get { return UpdateIntervalParameter.Value.Value; }
@@ -219 +218 @@
     [StorableHook(HookType.AfterDeserialization)]
     private void AfterDeserialization() {
+      if (Parameters.ContainsKey(RandomInitializationParameterName))
+        Parameters.Add(new FixedValueParameter<BoolValue>(RandomInitializationParameterName, "Wether data points should be randomly initialized or according to the first 2 dimensions", new BoolValue(true)));
       RegisterParameterEvents();
     }
@@ -279 +280 @@
       var problemData = Problem.ProblemData;
       // set up and initialize everything if necessary
+      var wdist = DistanceFunction as WeightedEuclideanDistance;
+      if (wdist != null) wdist.Initialize(problemData);
       if (state == null) {
         if (SetSeedRandomly) Seed = new System.Random().Next();
@@ -299 +302 @@
 
         if (Normalization) data = NormalizeInputData(data);
-        var randomInit = Parameters.ContainsKey(RandomInitializationParameterName) ? RandomInitialization : true;
-        state = TSNEStatic<double[]>.CreateState(data, DistanceFunction, random, NewDimensions, Perplexity, Theta, StopLyingIteration, MomentumSwitchIteration, InitialMomentum, FinalMomentum, Eta, randomInit);
+        state = TSNEStatic<double[]>.CreateState(data, DistanceFunction, random, NewDimensions, Perplexity, Theta, StopLyingIteration, MomentumSwitchIteration, InitialMomentum, FinalMomentum, Eta, RandomInitialization);
         SetUpResults(allindices);
         iter = 0;
@@ -338 +340 @@
       if (Problem == null || Problem.ProblemData == null) return;
       OnPerplexityChanged(this, null);
+      OnColumnsChanged(this, null);
       Problem.ProblemData.Changed += OnPerplexityChanged;
       Problem.ProblemData.Changed += OnColumnsChanged;
@@ -349 +352 @@
     private void OnColumnsChanged(object sender, EventArgs e) {
       if (Problem == null || Problem.ProblemData == null || Problem.ProblemData.Dataset == null || !Parameters.ContainsKey(DistanceFunctionParameterName)) return;
-      DistanceFunctionParameter.ValidValues.OfType<WeightedEuclideanDistance>().Single().Weights = new DoubleArray(Problem.ProblemData.AllowedInputVariables.Select(x => 1.0).ToArray());
+      DistanceFunctionParameter.ValidValues.OfType<WeightedEuclideanDistance>().Single().AdaptWeightVector(Problem.ProblemData);
     }
 
@@ -546 +549 @@
       var colorGradient = ColorGradient.Colors;
       var range = high - low;
-      var h = cell / range * colorGradient.Count;
+      var h = Math.Min(cell / range * colorGradient.Count, colorGradient.Count - 1);
       return colorGradient[(int) h];
     }