Changeset 14887 for branches/RBFRegression
- Timestamp:
- 04/24/17 18:31:44 (8 years ago)
- Location:
- branches/RBFRegression/HeuristicLab.Algorithms.DataAnalysis/3.4
- Files:
-
- 1 added
- 2 deleted
- 10 edited
Legend:
- Unmodified
- Added
- Removed
-
branches/RBFRegression/HeuristicLab.Algorithms.DataAnalysis/3.4/HeuristicLab.Algorithms.DataAnalysis-3.4.csproj
r14885 r14887 309 309 <Compile Include="KernelRidgeRegression\KernelFunctions\CicularKernel.cs" /> 310 310 <Compile Include="KernelRidgeRegression\KernelFunctions\GaussianKernel.cs" /> 311 <Compile Include="KernelRidgeRegression\KernelFunctions\IKernel.cs" /> 311 312 <Compile Include="KernelRidgeRegression\KernelFunctions\InverseMultiquadraticKernel.cs" /> 312 313 <Compile Include="KernelRidgeRegression\KernelFunctions\KernelBase.cs" /> 313 <Compile Include="KernelRidgeRegression\KernelFunctions\LaplacianKernel.cs" />314 314 <Compile Include="KernelRidgeRegression\KernelFunctions\MultiquadraticKernel.cs" /> 315 315 <Compile Include="KernelRidgeRegression\KernelFunctions\PolysplineKernel.cs" /> -
branches/RBFRegression/HeuristicLab.Algorithms.DataAnalysis/3.4/KernelRidgeRegression/KernelFunctions/CicularKernel.cs
r14883 r14887 23 23 using HeuristicLab.Common; 24 24 using HeuristicLab.Core; 25 using HeuristicLab.Data;26 using HeuristicLab.Parameters;27 25 using HeuristicLab.Persistence.Default.CompositeSerializers.Storable; 28 26 29 namespace HeuristicLab.Algorithms.DataAnalysis {27 namespace HeuristicLab.Algorithms.DataAnalysis.KernelRidgeRegression { 30 28 [StorableClass] 31 [Item("CircularKernel", "A circular kernel function ")]29 [Item("CircularKernel", "A circular kernel function 2*pi*(acos(-d)-d*(1-n²)^(0.5)) where n = ||x-c|| and d = n/beta")] 32 30 public class CircularKernel : KernelBase { 33 31 … … 39 37 protected CircularKernel(CircularKernel original, Cloner cloner) : base(original, cloner) { } 40 38 public CircularKernel() { 41 Parameters.Add(new FixedValueParameter<DoubleValue>(BetaParameterName, "The beta in the kernel function 2*pi*(acos(-d)-d*(1-n²)^(0.5)) where n = ||x-c|| and d = n/beta", new DoubleValue(2)));42 39 } 43 40 public override IDeepCloneable Clone(Cloner cloner) { … … 47 44 48 45 protected override double Get(double norm) { 49 if (Math.Abs(Beta) < double.Epsilon) return double.NaN; 50 if (norm >= Beta) return 0; 51 var d = norm / Beta; 46 var beta = Beta.Value; 47 if (Math.Abs(beta) < double.Epsilon) return double.NaN; 48 if (norm >= beta) return 0; 49 var d = norm / beta; 52 50 return Math.Acos(-d) - d * Math.Sqrt(1 - d * d) - Math.PI / 2; 53 51 } 54 52 55 53 protected override double GetGradient(double norm) { 56 if (Math.Abs(Beta) < double.Epsilon) return double.NaN; 57 if (Beta < norm) return 0; 58 return -2*Math.Pow(norm,3)/(Math.Pow(Beta,4)*Math.Sqrt(1-norm*norm/(Beta*Beta))); 54 var beta = Beta.Value; 55 if (Math.Abs(beta) < double.Epsilon) return double.NaN; 56 if (beta < norm) return 0; 57 return -2 * Math.Pow(norm, 3) / (Math.Pow(beta, 4) * Math.Sqrt(1 - norm * norm / (beta * beta))); 59 58 } 60 59 } -
branches/RBFRegression/HeuristicLab.Algorithms.DataAnalysis/3.4/KernelRidgeRegression/KernelFunctions/GaussianKernel.cs
r14883 r14887 21 21 22 22 using System; 23 23 24 using HeuristicLab.Common; 24 25 using HeuristicLab.Core; 25 using HeuristicLab.Data; 26 using HeuristicLab.Parameters; 26 27 27 using HeuristicLab.Persistence.Default.CompositeSerializers.Storable; 28 28 29 namespace HeuristicLab.Algorithms.DataAnalysis {29 namespace HeuristicLab.Algorithms.DataAnalysis.KernelRidgeRegression { 30 30 [StorableClass] 31 [Item("GaussianKernel", "A kernel function that uses Gaussian function ")]31 [Item("GaussianKernel", "A kernel function that uses Gaussian function exp(-||x-c||/beta²). Positive definite beta > 0")] 32 32 public class GaussianKernel : KernelBase { 33 33 … … 39 39 protected GaussianKernel(GaussianKernel original, Cloner cloner) : base(original, cloner) { } 40 40 public GaussianKernel() { 41 Parameters.Add(new FixedValueParameter<DoubleValue>(BetaParameterName, "The beta in the kernelfunction exp(-||x-c||/beta²)", new DoubleValue(2)));42 41 } 43 42 public override IDeepCloneable Clone(Cloner cloner) { … … 47 46 48 47 protected override double Get(double norm) { 49 if (Math.Abs(Beta) < double.Epsilon) return double.NaN; 50 return Math.Exp(-norm * norm / (Beta * Beta)); 48 var beta = Beta.Value; 49 if (Math.Abs(beta) < double.Epsilon) return double.NaN; 50 return Math.Exp(-norm * norm / (beta * beta)); 51 51 } 52 52 53 53 protected override double GetGradient(double norm) { 54 if (Math.Abs(Beta) < double.Epsilon) return double.NaN; 55 return 2 * norm * norm / Math.Pow(Beta, 3) * Math.Exp(-norm * norm / (Beta * Beta)); 54 var beta = Beta.Value; 55 if (Math.Abs(beta) < double.Epsilon) return double.NaN; 56 return 2 * norm * norm / Math.Pow(beta, 3) * Math.Exp(-norm * norm / (beta * beta)); 56 57 } 57 58 } -
branches/RBFRegression/HeuristicLab.Algorithms.DataAnalysis/3.4/KernelRidgeRegression/KernelFunctions/InverseMultiquadraticKernel.cs
r14883 r14887 22 22 using System; 23 23 using HeuristicLab.Common; 24 using HeuristicLab.Core; 25 using HeuristicLab.Data; 26 using HeuristicLab.Parameters; 24 using HeuristicLab.Core; 27 25 using HeuristicLab.Persistence.Default.CompositeSerializers.Storable; 28 26 29 namespace HeuristicLab.Algorithms.DataAnalysis {27 namespace HeuristicLab.Algorithms.DataAnalysis.KernelRidgeRegression { 30 28 [StorableClass] 31 [Item("InverseMultiquadraticKernel", "A kernel function that uses the inverse multi quadratic function")]29 [Item("InverseMultiquadraticKernel", "A kernel function that uses the inverse multi-quadratic function 1 / sqrt(1+||x-c||^2/beta). Positive definite: beta > 0")] 32 30 public class InverseMultiquadraticKernel : KernelBase { 33 31 #region HLConstructors & Boilerplate … … 38 36 protected InverseMultiquadraticKernel(InverseMultiquadraticKernel original, Cloner cloner) : base(original, cloner) { } 39 37 public InverseMultiquadraticKernel() { 40 Parameters.Add(new FixedValueParameter<DoubleValue>(BetaParameterName, "The beta in the kernel function 1 / sqrt(1+||x-c||^2/beta)", new DoubleValue(2)));41 38 } 42 39 public override IDeepCloneable Clone(Cloner cloner) { … … 46 43 47 44 protected override double Get(double norm) { 48 if (Math.Abs(Beta) < double.Epsilon) return double.NaN; 49 return 1 / Math.Sqrt(1 + norm * norm / Beta); 45 var beta = Beta.Value; 46 if (Math.Abs(beta) < double.Epsilon) return double.NaN; 47 return 1 / Math.Sqrt(1 + norm * norm / beta); 50 48 } 51 49 52 50 protected override double GetGradient(double norm) { 53 if (Math.Abs(Beta) < double.Epsilon) return double.NaN; 54 return norm * norm / (2 * Beta * Beta * Math.Pow((norm * norm + Beta) / Beta, 1.5)); 51 var beta = Beta.Value; 52 if (Math.Abs(beta) < double.Epsilon) return double.NaN; 53 return norm * norm / (2 * beta * beta * Math.Pow((norm * norm + beta) / beta, 1.5)); 55 54 } 56 55 } -
branches/RBFRegression/HeuristicLab.Algorithms.DataAnalysis/3.4/KernelRidgeRegression/KernelFunctions/KernelBase.cs
r14872 r14887 21 21 22 22 using System; 23 using System.Collections;24 23 using System.Collections.Generic; 24 using System.Linq; 25 25 using HeuristicLab.Common; 26 26 using HeuristicLab.Core; 27 using HeuristicLab.Data;28 27 using HeuristicLab.Parameters; 29 28 using HeuristicLab.Persistence.Default.CompositeSerializers.Storable; 30 29 31 namespace HeuristicLab.Algorithms.DataAnalysis {30 namespace HeuristicLab.Algorithms.DataAnalysis.KernelRidgeRegression { 32 31 [StorableClass] 33 public abstract class KernelBase : ParameterizedNamedItem, I CovarianceFunction{32 public abstract class KernelBase : ParameterizedNamedItem, IKernel { 34 33 35 34 #region Parameternames 36 35 private const string DistanceParameterName = "Distance"; 37 protected const string BetaParameterName = "Beta";38 36 #endregion 39 37 #region Parameterproperties … … 42 40 } 43 41 44 public IFixedValueParameter<DoubleValue> BetaParameter { 45 get { return Parameters[BetaParameterName] as FixedValueParameter<DoubleValue>; } 46 } 47 42 [Storable] 43 public double? Beta { get; set; } 48 44 #endregion 49 45 #region Properties 50 46 public IDistance Distance { 51 47 get { return DistanceParameter.Value; } 52 } 53 54 public double Beta { 55 get { return BetaParameter.Value.Value; } 48 set { DistanceParameter.Value = value; } 56 49 } 57 50 … … 64 57 65 58 protected KernelBase(KernelBase original, Cloner cloner) 66 : base(original, cloner) { } 59 : base(original, cloner) { 60 Beta = original.Beta; 61 } 67 62 68 63 protected KernelBase() { … … 78 73 79 74 public int GetNumberOfParameters(int numberOfVariables) { 80 return 1;75 return Beta.HasValue ? 0 : 1; 81 76 } 82 77 83 78 public void SetParameter(double[] p) { 84 if (p != null && p.Length == 1) Beta Parameter.Value.Value = p[0];79 if (p != null && p.Length == 1) Beta = new double?(p[0]); 85 80 } 86 81 87 82 public ParameterizedCovarianceFunction GetParameterizedCovarianceFunction(double[] p, int[] columnIndices) { 88 if (p == null || p.Length != 1) throw new ArgumentException("Illegal parametrization");83 if (p.Length != GetNumberOfParameters(columnIndices.Length)) throw new ArgumentException("Illegal parametrization"); 89 84 var myClone = (KernelBase)Clone(new Cloner()); 90 myClone. BetaParameter.Value.Value = p[0];85 myClone.SetParameter(p); 91 86 var cov = new ParameterizedCovarianceFunction { 92 87 Covariance = (x, i, j) => myClone.Get(GetNorm(x, x, i, j, columnIndices)), … … 102 97 var dist = Distance as IDistance<IEnumerable<double>>; 103 98 if (dist == null) throw new ArgumentException("The distance needs to apply to double vectors"); 104 var r1 = new IndexedEnumerable(x, i, columnIndices);105 var r2 = new IndexedEnumerable(xt, j, columnIndices);99 var r1 = columnIndices.Select(c => x[i, c]); 100 var r2 = columnIndices.Select(c => xt[j, c]); 106 101 return dist.Get(r1, r2); 107 }108 internal class IndexedEnumerable : IEnumerable<double> {109 private readonly double[,] data;110 private readonly int row;111 private readonly int[] columnIndices;112 113 public IndexedEnumerable(double[,] data, int row, int[] columnIndices) {114 this.data = data;115 this.row = row;116 this.columnIndices = columnIndices;117 }118 119 public IEnumerator<double> GetEnumerator() {120 return new IndexedEnumerator(data, row, columnIndices);121 }122 123 IEnumerator IEnumerable.GetEnumerator() {124 return new IndexedEnumerator(data, row, columnIndices);125 }126 }127 internal class IndexedEnumerator : IEnumerator<double> {128 private readonly IEnumerator<int> column;129 private readonly double[,] data;130 private readonly int row;131 132 public IndexedEnumerator(double[,] data, int row, int[] columnIndices) {133 this.data = data;134 this.row = row;135 column = ((IEnumerable<int>)columnIndices).GetEnumerator();136 }137 138 public double Current {139 get { return data[row, column.Current]; }140 }141 142 object IEnumerator.Current {143 get {144 return data[row, column.Current];145 }146 }147 148 public void Dispose() { }149 150 public bool MoveNext() {151 return column.MoveNext();152 }153 154 public void Reset() {155 column.Reset();156 }157 102 } 158 103 } -
branches/RBFRegression/HeuristicLab.Algorithms.DataAnalysis/3.4/KernelRidgeRegression/KernelFunctions/MultiquadraticKernel.cs
r14883 r14887 22 22 using System; 23 23 using HeuristicLab.Common; 24 using HeuristicLab.Core; 25 using HeuristicLab.Data; 26 using HeuristicLab.Parameters; 24 using HeuristicLab.Core; 27 25 using HeuristicLab.Persistence.Default.CompositeSerializers.Storable; 28 26 29 namespace HeuristicLab.Algorithms.DataAnalysis {27 namespace HeuristicLab.Algorithms.DataAnalysis.KernelRidgeRegression { 30 28 [StorableClass] 31 [Item("MultiquadraticKernel", "A kernel function that uses the multiquadratic function")] 29 // conditionally positive definite. (need to add polynomials) see http://num.math.uni-goettingen.de/schaback/teaching/sc.pdf 30 [Item("MultiquadraticKernel", "A kernel function that uses the multi-quadratic function (sqrt(1+||x-c||²/β).")] 32 31 public class MultiquadraticKernel : KernelBase { 33 32 … … 41 40 42 41 public MultiquadraticKernel() { 43 Parameters.Add(new FixedValueParameter<DoubleValue>(BetaParameterName, "The beta in the kernel function sqrt(1+||x-c||²/beta)", new DoubleValue(2)));44 42 } 45 43 public override IDeepCloneable Clone(Cloner cloner) { … … 48 46 #endregion 49 47 protected override double Get(double norm) { 50 if (Math.Abs(Beta) < double.Epsilon) return double.NaN; 51 return Math.Sqrt(1 + norm * norm / Beta); 48 var beta = Beta.Value; 49 if (Math.Abs(beta) < double.Epsilon) return double.NaN; 50 return Math.Sqrt(1 + norm * norm / beta); 52 51 } 53 52 54 53 protected override double GetGradient(double norm) { 55 if (Math.Abs(Beta) < double.Epsilon) return double.NaN; 56 var dividend = 2 * Beta * Beta * Math.Sqrt((Beta + norm * norm) / Beta); 54 var beta = Beta.Value; 55 if (Math.Abs(beta) < double.Epsilon) return double.NaN; 56 var dividend = 2 * beta * beta * Math.Sqrt((beta + norm * norm) / beta); 57 57 return -norm * norm / dividend; 58 58 } -
branches/RBFRegression/HeuristicLab.Algorithms.DataAnalysis/3.4/KernelRidgeRegression/KernelFunctions/PolysplineKernel.cs
r14883 r14887 22 22 using System; 23 23 using HeuristicLab.Common; 24 using HeuristicLab.Core; 25 using HeuristicLab.Data; 26 using HeuristicLab.Parameters; 24 using HeuristicLab.Core; 27 25 using HeuristicLab.Persistence.Default.CompositeSerializers.Storable; 28 26 29 namespace HeuristicLab.Algorithms.DataAnalysis {27 namespace HeuristicLab.Algorithms.DataAnalysis.KernelRidgeRegression { 30 28 [StorableClass] 31 [Item("PolysplineKernel", "A kernel function that uses the multiquadratic function")] 29 // conditionally positive definite. (need to add polynomials) see http://num.math.uni-goettingen.de/schaback/teaching/sc.pdf 30 [Item("PolysplineKernel", "A kernel function that uses the poly-spline function ||x-c||^Beta.")] 32 31 public class PolysplineKernel : KernelBase { 33 32 … … 40 39 : base(original, cloner) { } 41 40 public PolysplineKernel() { 42 Parameters.Add(new FixedValueParameter<DoubleValue>(BetaParameterName, "The Beta in the kernelfunction ||x-c||^Beta", new DoubleValue(1.5)));43 41 } 44 42 public override IDeepCloneable Clone(Cloner cloner) { … … 48 46 49 47 protected override double Get(double norm) { 50 return Math.Pow(norm, Beta );48 return Math.Pow(norm, Beta.Value); 51 49 } 52 50 53 51 protected override double GetGradient(double norm) { 54 return Math.Pow(norm, Beta ) * Math.Log(norm);52 return Math.Pow(norm, Beta.Value) * Math.Log(norm); 55 53 } 56 54 } -
branches/RBFRegression/HeuristicLab.Algorithms.DataAnalysis/3.4/KernelRidgeRegression/KernelFunctions/ThinPlatePolysplineKernel.cs
r14883 r14887 23 23 using HeuristicLab.Common; 24 24 using HeuristicLab.Core; 25 using HeuristicLab.Data;26 using HeuristicLab.Parameters;27 25 using HeuristicLab.Persistence.Default.CompositeSerializers.Storable; 28 26 29 namespace HeuristicLab.Algorithms.DataAnalysis {27 namespace HeuristicLab.Algorithms.DataAnalysis.KernelRidgeRegression { 30 28 [StorableClass] 31 [Item("ThinPlatePolysplineKernel", "A kernel function that uses the ThinPlatePolyspline function")] 29 // conditionally positive definite. (need to add polynomials) see http://num.math.uni-goettingen.de/schaback/teaching/sc.pdf 30 [Item("ThinPlatePolysplineKernel", "A kernel function that uses the ThinPlatePolyspline function ||x-c||^(2*Beta)*log(||x-c||^Beta)")] 32 31 public class ThinPlatePolysplineKernel : KernelBase { 33 32 #region HLConstructors & Boilerplate … … 38 37 protected ThinPlatePolysplineKernel(ThinPlatePolysplineKernel original, Cloner cloner) : base(original, cloner) { } 39 38 public ThinPlatePolysplineKernel() { 40 Parameters.Add(new FixedValueParameter<DoubleValue>(BetaParameterName, "The Beta in the kernelfunction ||x-c||^(2*Beta)*log(||x-c||^Beta)", new DoubleValue(1)));41 39 } 42 40 public override IDeepCloneable Clone(Cloner cloner) { … … 46 44 47 45 protected override double Get(double norm) { 48 if (Math.Pow(norm, Beta) <= 0) return double.NaN; 49 return Math.Pow(norm, 2 * Beta) * Math.Log(1 + Math.Pow(norm, Beta)); 46 var beta = Beta.Value; 47 if (Math.Pow(norm, beta) < 0) return double.NaN; 48 return Math.Pow(norm, 2 * beta) * Math.Log(1 + Math.Pow(norm, beta)); 50 49 } 51 50 52 51 protected override double GetGradient(double norm) { 53 if (Math.Pow(norm, Beta) <= 0) return double.NaN; 54 return 2 * Math.Log(norm) * Math.Pow(norm, 2 * Beta) * Math.Log(1 + Math.Pow(norm, Beta)) + Math.Pow(norm, 3 * Beta) * Math.Log(norm) / (Math.Pow(norm, Beta) + 1); 52 var beta = Beta.Value; 53 if (Math.Pow(norm, beta) <= 0) return double.NaN; 54 return 2 * Math.Log(norm) * Math.Pow(norm, 2 * beta) * Math.Log(1 + Math.Pow(norm, beta)) + Math.Pow(norm, 3 * beta) * Math.Log(norm) / (Math.Pow(norm, beta) + 1); 55 55 } 56 56 } -
branches/RBFRegression/HeuristicLab.Algorithms.DataAnalysis/3.4/KernelRidgeRegression/KernelRidgeRegression.cs
r14885 r14887 31 31 using HeuristicLab.Problems.DataAnalysis; 32 32 33 namespace HeuristicLab.Algorithms.DataAnalysis { 34 /// <summary> 35 /// Linear regression data analysis algorithm. 36 /// </summary> 37 [Item("Radial Basis Function Regression (RBF-R)", "Radial basis function regression data analysis algorithm.")] 33 namespace HeuristicLab.Algorithms.DataAnalysis.KernelRidgeRegression { 34 [Item("Kernel Ridge Regression", "Kernelized ridge regression e.g. for radial basis function (RBF) regression.")] 38 35 [Creatable(CreatableAttribute.Categories.DataAnalysisRegression, Priority = 100)] 39 36 [StorableClass] 40 public sealed class RadialBasisRegression : BasicAlgorithm {41 private const string RBFRegressionSolutionResultName = "RBFregression solution";37 public sealed class KernelRidgeRegression : BasicAlgorithm { 38 private const string SolutionResultName = "Kernel ridge regression solution"; 42 39 43 40 public override bool SupportsPause { … … 55 52 private const string KernelParameterName = "Kernel"; 56 53 private const string ScaleInputVariablesParameterName = "ScaleInputVariables"; 54 private const string LambdaParameterName = "LogLambda"; 55 private const string BetaParameterName = "Beta"; 57 56 #endregion 58 57 59 58 #region parameter properties 60 public ValueParameter<I CovarianceFunction> KernelParameter {61 get { return (ValueParameter<I CovarianceFunction>)Parameters[KernelParameterName]; }59 public ValueParameter<IKernel> KernelParameter { 60 get { return (ValueParameter<IKernel>)Parameters[KernelParameterName]; } 62 61 } 63 62 … … 65 64 get { return (IFixedValueParameter<BoolValue>)Parameters[ScaleInputVariablesParameterName]; } 66 65 } 66 67 public IFixedValueParameter<DoubleValue> LogLambdaParameter { 68 get { return (IFixedValueParameter<DoubleValue>)Parameters[LambdaParameterName]; } 69 } 70 71 public IFixedValueParameter<DoubleValue> BetaParameter { 72 get { return (IFixedValueParameter<DoubleValue>)Parameters[BetaParameterName]; } 73 } 67 74 #endregion 68 75 69 76 #region properties 70 public I CovarianceFunctionKernel {77 public IKernel Kernel { 71 78 get { return KernelParameter.Value; } 72 79 } … … 77 84 } 78 85 86 public double LogLambda { 87 get { return LogLambdaParameter.Value.Value; } 88 set { LogLambdaParameter.Value.Value = value; } 89 } 90 91 public double Beta { 92 get { return BetaParameter.Value.Value; } 93 set { BetaParameter.Value.Value = value; } 94 } 79 95 #endregion 80 96 81 97 [StorableConstructor] 82 private RadialBasisRegression(bool deserializing) : base(deserializing) { }83 private RadialBasisRegression(RadialBasisRegression original, Cloner cloner)98 private KernelRidgeRegression(bool deserializing) : base(deserializing) { } 99 private KernelRidgeRegression(KernelRidgeRegression original, Cloner cloner) 84 100 : base(original, cloner) { 85 101 } 86 public RadialBasisRegression() {102 public KernelRidgeRegression() { 87 103 Problem = new RegressionProblem(); 88 Parameters.Add(new ValueParameter<I CovarianceFunction>(KernelParameterName, "The radial basis function"));104 Parameters.Add(new ValueParameter<IKernel>(KernelParameterName, "The kernel", new GaussianKernel())); 89 105 Parameters.Add(new FixedValueParameter<BoolValue>(ScaleInputVariablesParameterName, "Set to true if the input variables should be scaled to the interval [0..1]", new BoolValue(true))); 90 var kernel = new GaussianKernel();91 KernelParameter.Value = kernel;106 Parameters.Add(new FixedValueParameter<DoubleValue>(LambdaParameterName, "The log10-transformed weight for the regularization term lambda [-inf..+inf]. Small values produce more complex models, large values produce models with larger errors. Set to very small value (e.g. -1.0e15) for almost exact approximation", new DoubleValue(-2))); 107 Parameters.Add(new FixedValueParameter<DoubleValue>(BetaParameterName, "The beta parameter for the kernel", new DoubleValue(2))); 92 108 } 93 109 [StorableHook(HookType.AfterDeserialization)] … … 95 111 96 112 public override IDeepCloneable Clone(Cloner cloner) { 97 return new RadialBasisRegression(this, cloner);113 return new KernelRidgeRegression(this, cloner); 98 114 } 99 115 100 116 protected override void Run(CancellationToken cancellationToken) { 101 double loocvrmse, rmsError; 102 var solution = CreateRadialBasisRegressionSolution(Problem.ProblemData, Kernel, ScaleInputVariables, out loocvrmse, out rmsError); 103 Results.Add(new Result(RBFRegressionSolutionResultName, "The RBF regression solution.", solution)); 104 Results.Add(new Result("LOOCVRMSE", "The root mean squared error of a leave-one-out-cross-validation on the training set", new DoubleValue(loocvrmse))); 117 double rmsError; 118 var kernel = Kernel; 119 kernel.Beta = Beta; 120 var solution = CreateRadialBasisRegressionSolution(Problem.ProblemData, kernel, Math.Pow(10, LogLambda), ScaleInputVariables, out rmsError); 121 Results.Add(new Result(SolutionResultName, "The kernel ridge regression solution.", solution)); 105 122 Results.Add(new Result("RMSE (test)", "The root mean squared error of the solution on the test set.", new DoubleValue(rmsError))); 106 123 } 107 124 108 public static IRegressionSolution CreateRadialBasisRegressionSolution(IRegressionProblemData problemData, ICovarianceFunction kernel, bool scaleInputs, out double loocvRmsError, out double rmsError) { 109 var model = new RadialBasisFunctionModel(problemData.Dataset, problemData.TargetVariable, problemData.AllowedInputVariables, problemData.TrainingIndices, scaleInputs, kernel); 110 loocvRmsError = model.LeaveOneOutCrossValidationRootMeanSquaredError(); 111 rmsError = Math.Sqrt(model.GetEstimatedValues(problemData.Dataset, problemData.TestIndices) 112 .Zip(problemData.TargetVariableTestValues, (a, b) => (a - b) * (a - b)) 113 .Average()); 125 public static IRegressionSolution CreateRadialBasisRegressionSolution(IRegressionProblemData problemData, ICovarianceFunction kernel, double lambda, bool scaleInputs, out double rmsError) { 126 var model = new KernelRidgeRegressionModel(problemData.Dataset, problemData.TargetVariable, problemData.AllowedInputVariables, problemData.TrainingIndices, scaleInputs, kernel, lambda); 127 rmsError = double.NaN; 128 if (problemData.TestIndices.Any()) { 129 rmsError = Math.Sqrt(model.GetEstimatedValues(problemData.Dataset, problemData.TestIndices) 130 .Zip(problemData.TargetVariableTestValues, (a, b) => (a - b) * (a - b)) 131 .Average()); 132 } 114 133 var solution = model.CreateRegressionSolution((IRegressionProblemData)problemData.Clone()); 115 solution.Model.Name = " RBF Regression Model";116 solution.Name = "RBF Regression Solution";134 solution.Model.Name = "Kernel ridge regression model"; 135 solution.Name = SolutionResultName; 117 136 return solution; 118 137 } -
branches/RBFRegression/HeuristicLab.Algorithms.DataAnalysis/3.4/KernelRidgeRegression/KernelRidgeRegressionModel.cs
r14885 r14887 21 21 22 22 using System; 23 using System.Collections.Generic; 24 using System.Diagnostics; 23 using System.Collections.Generic; 25 24 using System.Linq; 26 25 using HeuristicLab.Common; … … 29 28 using HeuristicLab.Problems.DataAnalysis; 30 29 31 namespace HeuristicLab.Algorithms.DataAnalysis { 32 /// <summary> 33 /// Represents an RBF regression model. 34 /// </summary> 30 namespace HeuristicLab.Algorithms.DataAnalysis.KernelRidgeRegression { 35 31 [StorableClass] 36 [Item(" RBFModel", "An RBFregression model")]37 public sealed class RadialBasisFunctionModel : RegressionModel, IConfidenceRegressionModel {32 [Item("KernelRidgeRegressionModel", "A kernel ridge regression model")] 33 public sealed class KernelRidgeRegressionModel : RegressionModel { 38 34 public override IEnumerable<string> VariablesUsedForPrediction { 39 35 get { return allowedInputVariables; } … … 58 54 private readonly ICovarianceFunction kernel; 59 55 60 private double[,] gramInvert; // not storable as it can be large (recreate after deserialization as required) 56 [Storable] 57 private readonly double lambda; 61 58 62 59 [Storable] 63 private readonly double meanOffset; // implementation works for zero-mean target variables 60 private readonly double yOffset; // implementation works for zero-mean target variables 61 62 [Storable] 63 private readonly double yScale; 64 64 65 65 [StorableConstructor] 66 private RadialBasisFunctionModel(bool deserializing) : base(deserializing) { }67 private RadialBasisFunctionModel(RadialBasisFunctionModel original, Cloner cloner)66 private KernelRidgeRegressionModel(bool deserializing) : base(deserializing) { } 67 private KernelRidgeRegressionModel(KernelRidgeRegressionModel original, Cloner cloner) 68 68 : base(original, cloner) { 69 69 // shallow copies of arrays because they cannot be modified … … 71 71 alpha = original.alpha; 72 72 trainX = original.trainX; 73 gramInvert = original.gramInvert;74 73 scaling = original.scaling; 74 lambda = original.lambda; 75 75 76 meanOffset = original.meanOffset; 76 yOffset = original.yOffset; 77 yScale = original.yScale; 77 78 if (original.kernel != null) 78 79 kernel = cloner.Clone(original.kernel); 79 80 } 80 public RadialBasisFunctionModel(IDataset dataset, string targetVariable, IEnumerable<string> allowedInputVariables, IEnumerable<int> rows, 81 bool scaleInputs, ICovarianceFunction kernel) 82 : base(targetVariable) { 81 public override IDeepCloneable Clone(Cloner cloner) { 82 return new KernelRidgeRegressionModel(this, cloner); 83 } 84 85 public KernelRidgeRegressionModel(IDataset dataset, string targetVariable, IEnumerable<string> allowedInputVariables, IEnumerable<int> rows, 86 bool scaleInputs, ICovarianceFunction kernel, double lambda = 0.1) : base(targetVariable) { 83 87 if (kernel.GetNumberOfParameters(allowedInputVariables.Count()) > 0) throw new ArgumentException("All parameters in the kernel function must be specified."); 84 88 name = ItemName; … … 87 91 var trainingRows = rows.ToArray(); 88 92 this.kernel = (ICovarianceFunction)kernel.Clone(); 93 this.lambda = lambda; 89 94 try { 90 95 if (scaleInputs) … … 92 97 trainX = ExtractData(dataset, trainingRows, scaling); 93 98 var y = dataset.GetDoubleValues(targetVariable, trainingRows).ToArray(); 94 meanOffset = y.Average(); 95 for (int i = 0; i < y.Length; i++) y[i] -= meanOffset; 99 yOffset = y.Average(); 100 yScale = 1.0 / y.StandardDeviation(); 101 for (int i = 0; i < y.Length; i++) { 102 y[i] -= yOffset; 103 y[i] *= yScale; 104 } 96 105 int info; 97 // TODO: improve efficiency by decomposing matrix once instead of solving the system and then inverting the matrix 98 alglib.densesolverlsreport denseSolveRep; 99 gramInvert = BuildGramMatrix(trainX); 106 alglib.densesolverreport denseSolveRep; 107 var gram = BuildGramMatrix(trainX, lambda); 100 108 int n = trainX.GetLength(0); 101 alglib.rmatrixsolvels(gramInvert, n, n, y, 0.0, out info, out denseSolveRep, out alpha); 109 110 // cholesky decomposition 111 var res = alglib.trfac.spdmatrixcholesky(ref gram, n, false); 112 if(res == false) throw new ArgumentException("Could not decompose matrix. Is it quadratic symmetric positive definite?"); 113 114 alglib.spdmatrixcholeskysolve(gram, n, false, y, out info, out denseSolveRep, out alpha); 102 115 if (info != 1) throw new ArgumentException("Could not create model."); 103 104 alglib.matinvreport report;105 alglib.rmatrixinverse(ref gramInvert, out info, out report);106 if (info != 1) throw new ArgumentException("Could not invert matrix. Is it quadratic symmetric positive definite?");107 108 116 } catch (alglib.alglibexception ae) { 109 117 // wrap exception so that calling code doesn't have to know about alglib implementation 110 throw new ArgumentException("There was a problem in the calculation of the RBFmodel", ae);118 throw new ArgumentException("There was a problem in the calculation of the kernel ridge regression model", ae); 111 119 } 120 } 121 122 123 #region IRegressionModel Members 124 public override IEnumerable<double> GetEstimatedValues(IDataset dataset, IEnumerable<int> rows) { 125 var newX = ExtractData(dataset, rows, scaling); 126 var dim = newX.GetLength(1); 127 var cov = kernel.GetParameterizedCovarianceFunction(new double[0], Enumerable.Range(0, dim).ToArray()); 128 129 var pred = new double[newX.GetLength(0)]; 130 for (int i = 0; i < pred.Length; i++) { 131 double sum = 0.0; 132 for (int j = 0; j < alpha.Length; j++) { 133 sum += alpha[j] * cov.CrossCovariance(trainX, newX, j, i); 134 } 135 pred[i] = sum / yScale + yOffset; 136 } 137 return pred; 138 } 139 public override IRegressionSolution CreateRegressionSolution(IRegressionProblemData problemData) { 140 return new RegressionSolution(this, new RegressionProblemData(problemData)); 141 } 142 #endregion 143 144 #region helpers 145 private double[,] BuildGramMatrix(double[,] data, double lambda) { 146 var n = data.GetLength(0); 147 var dim = data.GetLength(1); 148 var cov = kernel.GetParameterizedCovarianceFunction(new double[0], Enumerable.Range(0, dim).ToArray()); 149 var gram = new double[n, n]; 150 // G = (K + λ I) 151 for (var i = 0; i < n; i++) { 152 for (var j = i; j < n; j++) { 153 gram[j, i] = cov.Covariance(data, i, j); // symmetric matrix --> fill only lower triangle 154 } 155 gram[i, i] += lambda; 156 } 157 return gram; 112 158 } 113 159 … … 145 191 return res; 146 192 } 147 148 public override IDeepCloneable Clone(Cloner cloner) {149 return new RadialBasisFunctionModel(this, cloner);150 }151 152 #region IRegressionModel Members153 public override IEnumerable<double> GetEstimatedValues(IDataset dataset, IEnumerable<int> rows) {154 var newX = ExtractData(dataset, rows, scaling);155 var dim = newX.GetLength(1);156 var cov = kernel.GetParameterizedCovarianceFunction(new double[0], Enumerable.Range(0, dim).ToArray());157 158 var pred = new double[newX.GetLength(0)];159 for (int i = 0; i < pred.Length; i++) {160 double sum = meanOffset;161 for (int j = 0; j < alpha.Length; j++) {162 sum += alpha[j] * cov.CrossCovariance(trainX, newX, j, i);163 }164 pred[i] = sum;165 }166 return pred;167 }168 public override IRegressionSolution CreateRegressionSolution(IRegressionProblemData problemData) {169 return new ConfidenceRegressionSolution(this, new RegressionProblemData(problemData));170 }171 #endregion172 173 public IEnumerable<double> GetEstimatedVariances(IDataset dataset, IEnumerable<int> rows) {174 if (gramInvert == null) CreateAndInvertGramMatrix();175 int n = gramInvert.GetLength(0);176 var newData = ExtractData(dataset, rows, scaling);177 var dim = newData.GetLength(1);178 var cov = kernel.GetParameterizedCovarianceFunction(new double[0], Enumerable.Range(0, dim).ToArray());179 180 // TODO perf (matrix matrix multiplication)181 for (int i = 0; i < newData.GetLength(0); i++) {182 double[] p = new double[n];183 184 for (int j = 0; j < trainX.GetLength(0); j++) {185 p[j] = cov.CrossCovariance(trainX, newData, j, i);186 }187 188 var Ap = new double[n];189 alglib.ablas.rmatrixmv(n, n, gramInvert, 0, 0, 0, p, 0, ref Ap, 0);190 var res = 0.0;191 // dot product192 for (int j = 0; j < p.Length; j++) res += p[j] * Ap[j];193 yield return res > 0 ? res : 0;194 }195 }196 public double LeaveOneOutCrossValidationRootMeanSquaredError() {197 if (gramInvert == null) CreateAndInvertGramMatrix();198 var n = gramInvert.GetLength(0);199 var s = 1.0 / n;200 201 var sum = 0.0;202 for (int i = 0; i < alpha.Length; i++) {203 var x = alpha[i] / gramInvert[i, i];204 sum += x * x;205 }206 sum *= s;207 return Math.Sqrt(sum);208 }209 210 private void CreateAndInvertGramMatrix() {211 try {212 gramInvert = BuildGramMatrix(trainX);213 int info = 0;214 alglib.matinvreport report;215 alglib.rmatrixinverse(ref gramInvert, out info, out report);216 if (info != 1)217 throw new ArgumentException("Could not invert matrix. Is it quadratic symmetric positive definite?");218 } catch (alglib.alglibexception) {219 // wrap exception so that calling code doesn't have to know about alglib implementation220 throw new ArgumentException("Could not invert matrix. Is it quadratic symmetric positive definite?");221 }222 }223 #region helpers224 private double[,] BuildGramMatrix(double[,] data) {225 var n = data.GetLength(0);226 var dim = data.GetLength(1);227 var cov = kernel.GetParameterizedCovarianceFunction(new double[0], Enumerable.Range(0, dim).ToArray());228 var gram = new double[n, n];229 for (var i = 0; i < n; i++)230 for (var j = i; j < n; j++) {231 gram[j, i] = gram[i, j] = cov.Covariance(data, i, j); // symmetric matrix --> half of the work232 }233 return gram;234 }235 236 193 #endregion 237 194 }
Note: See TracChangeset
for help on using the changeset viewer.