Changeset 9270 for trunk/sources/HeuristicLab.Algorithms.DataAnalysis/3.4/Nca/Initialization

Timestamp:

03/01/13 18:32:26 (12 years ago)

Author:

abeham

Message:

#1913: Changed NCA to use LM-BFGS optimization algorithm, added model/solution creators, added operator for gradient calculation

Location:

trunk/sources/HeuristicLab.Algorithms.DataAnalysis/3.4/Nca/Initialization

Files:

• INcaInitializer.cs (modified) (2 diffs)
• LdaInitializer.cs (modified) (2 diffs)
• NcaInitializer.cs (added)
• PcaInitializer.cs (modified) (2 diffs)
• RandomInitializer.cs (modified) (2 diffs)

trunk/sources/HeuristicLab.Algorithms.DataAnalysis/3.4/Nca/Initialization/INcaInitializer.cs

@@ -24 +24 @@
 
 namespace HeuristicLab.Algorithms.DataAnalysis {
-  public interface INCAInitializer : IItem {
+  public interface INcaInitializer : IOperator {
     /// <summary>
     /// Calculates an initial projection for the NCA to start from.
@@ -30 +30 @@
     /// <param name="data">The problem data that contains the AllowedInputVariables and TrainingIndices.</param>
     /// <param name="dimensions">The amount of columns in the matrix</param>
-    /// <returns>A flat representation of a matrix that is read row-wise and contains AllowedInputVariables * TrainingIndices numbers.</returns>
-    double[] Initialize(IClassificationProblemData data, int dimensions);
+    /// <returns>The matrix that projects the input variables into a lower dimensional space.</returns>
+    double[,] Initialize(IClassificationProblemData data, Scaling scaling, int dimensions);
   }
 }

trunk/sources/HeuristicLab.Algorithms.DataAnalysis/3.4/Nca/Initialization/LdaInitializer.cs

@@ -20 +20 @@
 #endregion
 
-using System.Collections.Generic;
 using System.Linq;
 using HeuristicLab.Common;
@@ -30 +29 @@
   [Item("LDA", "Initializes the matrix by performing a linear discriminant analysis.")]
   [StorableClass]
-  public class LDAInitializer : Item, INCAInitializer {
+  public class LdaInitializer : NcaInitializer {
 
     [StorableConstructor]
-    protected LDAInitializer(bool deserializing) : base(deserializing) { }
-    protected LDAInitializer(LDAInitializer original, Cloner cloner) : base(original, cloner) { }
-    public LDAInitializer() : base() { }
+    protected LdaInitializer(bool deserializing) : base(deserializing) { }
+    protected LdaInitializer(LdaInitializer original, Cloner cloner) : base(original, cloner) { }
+    public LdaInitializer() : base() { }
 
     public override IDeepCloneable Clone(Cloner cloner) {
-      return new LDAInitializer(this, cloner);
+      return new LdaInitializer(this, cloner);
     }
 
-    public double[] Initialize(IClassificationProblemData data, int dimensions) {
+    public override double[,] Initialize(IClassificationProblemData data, Scaling scaling, int dimensions) {
       var instances = data.TrainingIndices.Count();
       var attributes = data.AllowedInputVariables.Count();
 
       var ldaDs = new double[instances, attributes + 1];
-      int row, col = 0;
-      foreach (var variable in data.AllowedInputVariables) {
-        row = 0;
-        foreach (var value in data.Dataset.GetDoubleValues(variable, data.TrainingIndices)) {
-          ldaDs[row, col] = value;
-          row++;
+      int j = 0;
+      foreach (var a in data.AllowedInputVariables) {
+        int i = 0;
+        var sv = scaling.GetScaledValues(data.Dataset, a, data.TrainingIndices);
+        foreach (var v in sv) {
+          ldaDs[i++, j] = v;
         }
-        col++;
+        j++;
       }
-      row = 0;
-      var uniqueClasses = new Dictionary<double, int>();
-      foreach (var label in data.Dataset.GetDoubleValues(data.TargetVariable, data.TrainingIndices)) {
-        if (!uniqueClasses.ContainsKey(label))
-          uniqueClasses[label] = uniqueClasses.Count;
-        ldaDs[row++, attributes] = label;
-      }
-      for (row = 0; row < instances; row++)
-        ldaDs[row, attributes] = uniqueClasses[ldaDs[row, attributes]];
+      j = 0;
+      foreach (var tv in data.Dataset.GetDoubleValues(data.TargetVariable, data.TrainingIndices))
+        ldaDs[j++, attributes] = tv;
+
+      var uniqueClasses = data.Dataset.GetDoubleValues(data.TargetVariable, data.TrainingIndices).Distinct().Count();
 
       int info;
       double[,] matrix;
-      alglib.fisherldan(ldaDs, instances, attributes, uniqueClasses.Count, out info, out matrix);
+      alglib.fisherldan(ldaDs, instances, attributes, uniqueClasses, out info, out matrix);
 
-      var result = new double[attributes * dimensions];
-      for (int i = 0; i < attributes; i++)
-        for (int j = 0; j < dimensions; j++)
-          result[i * dimensions + j] = matrix[i, j];
-
-      return result;
+      return matrix;
     }
 

trunk/sources/HeuristicLab.Algorithms.DataAnalysis/3.4/Nca/Initialization/PcaInitializer.cs

@@ -29 +29 @@
   [Item("PCA", "Initializes the matrix by performing a principal components analysis.")]
   [StorableClass]
-  public sealed class PCAInitializer : Item, INCAInitializer {
+  public sealed class PcaInitializer : NcaInitializer {
 
     [StorableConstructor]
-    private PCAInitializer(bool deserializing) : base(deserializing) { }
-    private PCAInitializer(PCAInitializer original, Cloner cloner) : base(original, cloner) { }
-    public PCAInitializer() : base() { }
+    private PcaInitializer(bool deserializing) : base(deserializing) { }
+    private PcaInitializer(PcaInitializer original, Cloner cloner) : base(original, cloner) { }
+    public PcaInitializer() : base() { }
 
     public override IDeepCloneable Clone(Cloner cloner) {
-      return new PCAInitializer(this, cloner);
+      return new PcaInitializer(this, cloner);
     }
 
-    public double[] Initialize(IClassificationProblemData data, int dimensions) {
+    public override double[,] Initialize(IClassificationProblemData data, Scaling scaling, int dimensions) {
       var instances = data.TrainingIndices.Count();
       var attributes = data.AllowedInputVariables.Count();
 
-      var pcaDs = new double[instances, attributes];
-      int col = 0;
-      foreach (var variable in data.AllowedInputVariables) {
-        int row = 0;
-        foreach (var value in data.Dataset.GetDoubleValues(variable, data.TrainingIndices)) {
-          pcaDs[row, col] = value;
-          row++;
-        }
-        col++;
-      }
+      var pcaDs = AlglibUtil.PrepareAndScaleInputMatrix(data.Dataset, data.AllowedInputVariables, data.TrainingIndices, scaling);
 
       int info;
@@ -60 +51 @@
       alglib.pcabuildbasis(pcaDs, instances, attributes, out info, out varianceValues, out matrix);
 
-      var result = new double[attributes * dimensions];
-      for (int i = 0; i < attributes; i++)
-        for (int j = 0; j < dimensions; j++)
-          result[i * dimensions + j] = matrix[i, j];
-
-      return result;
+      return matrix;
     }
 

trunk/sources/HeuristicLab.Algorithms.DataAnalysis/3.4/Nca/Initialization/RandomInitializer.cs

@@ -23 +23 @@
 using HeuristicLab.Common;
 using HeuristicLab.Core;
-using HeuristicLab.Data;
+using HeuristicLab.Optimization;
 using HeuristicLab.Parameters;
 using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
 using HeuristicLab.Problems.DataAnalysis;
-using HeuristicLab.Random;
 
 namespace HeuristicLab.Algorithms.DataAnalysis {
   [Item("Random", "Initializes the matrix randomly.")]
   [StorableClass]
-  public class RandomInitializer : ParameterizedNamedItem, INCAInitializer {
-    private IValueParameter<IntValue> RandomParameter {
-      get { return (IValueParameter<IntValue>)Parameters["Seed"]; }
-    }
-    private IValueParameter<BoolValue> SetSeedRandomlyParameter {
-      get { return (IValueParameter<BoolValue>)Parameters["SetSeedRandomly"]; }
-    }
-
-    public int Seed {
-      get { return RandomParameter.Value.Value; }
-      set { RandomParameter.Value.Value = value; }
-    }
-
-    public bool SetSeedRandomly {
-      get { return SetSeedRandomlyParameter.Value.Value; }
-      set { SetSeedRandomlyParameter.Value.Value = value; }
+  public sealed class RandomInitializer : NcaInitializer, IStochasticOperator {
+    public ILookupParameter<IRandom> RandomParameter {
+      get { return (ILookupParameter<IRandom>)Parameters["Random"]; }
     }
 
     [StorableConstructor]
-    protected RandomInitializer(bool deserializing) : base(deserializing) { }
-    protected RandomInitializer(RandomInitializer original, Cloner cloner) : base(original, cloner) { }
+    private RandomInitializer(bool deserializing) : base(deserializing) { }
+    private RandomInitializer(RandomInitializer original, Cloner cloner) : base(original, cloner) { }
     public RandomInitializer()
       : base() {
-      Parameters.Add(new ValueParameter<IntValue>("Seed", "The seed for the random number generator.", new IntValue(0)));
-      Parameters.Add(new ValueParameter<BoolValue>("SetSeedRandomly", "Whether the seed should be randomized for each call.", new BoolValue(true)));
+      Parameters.Add(new LookupParameter<IRandom>("Random", "The random number generator to use."));
     }
 
@@ -63 +48 @@
     }
 
-    public double[] Initialize(IClassificationProblemData data, int dimensions) {
-      var instances = data.TrainingIndices.Count();
+    public override double[,] Initialize(IClassificationProblemData data, Scaling scaling, int dimensions) {
       var attributes = data.AllowedInputVariables.Count();
 
-      var random = new MersenneTwister();
-      if (SetSeedRandomly) Seed = random.Next();
-      random.Reset(Seed);
-
-      var range = data.AllowedInputVariables.Select(x => data.Dataset.GetDoubleValues(x).Max() - data.Dataset.GetDoubleValues(x).Min()).ToArray();
-      var matrix = new double[attributes * dimensions];
-      for (int i = 0; i < matrix.Length; i++)
-        matrix[i] = random.NextDouble() / range[i / dimensions];
+      var random = RandomParameter.ActualValue;
+      var matrix = new double[attributes, dimensions];
+      for (int i = 0; i < attributes; i++)
+        for (int j = 0; j < dimensions; j++)
+          matrix[i, j] = random.NextDouble();
 
       return matrix;