source: trunk/sources/HeuristicLab.Algorithms.DataAnalysis/3.4/GaussianProcess/CovarianceFunctions/CovarianceRationalQuadraticArd.cs @ 13784

Last change on this file since 13784 was 13784, checked in by pfleck, 5 years ago

#2591 Made the creation of a GaussianProcessModel faster by avoiding additional iterators during calculation of the hyperparameter gradients.
The gradients of the hyperparameters are now calculated in one sweep and returned as IList, instead of returning an iterator (with yield return).
This avoids a large amount of Move-calls of the iterator, especially for covariance functions with a lot of hyperparameters.
Besides, the signature of the CovarianceGradientFunctionDelegate is changed, to return an IList instead of an IEnumerable to avoid unnececary ToList or ToArray calls.

File size: 6.8 KB
Line 
1#region License Information
2/* HeuristicLab
3 * Copyright (C) 2002-2015 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
4 *
5 * This file is part of HeuristicLab.
6 *
7 * HeuristicLab is free software: you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation, either version 3 of the License, or
10 * (at your option) any later version.
11 *
12 * HeuristicLab is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with HeuristicLab. If not, see <http://www.gnu.org/licenses/>.
19 */
20#endregion
21
22using System;
23using System.Collections.Generic;
24using System.Linq;
25using HeuristicLab.Common;
26using HeuristicLab.Core;
27using HeuristicLab.Data;
28using HeuristicLab.Parameters;
29using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
30
31namespace HeuristicLab.Algorithms.DataAnalysis {
32  [StorableClass]
33  [Item(Name = "CovarianceRationalQuadraticArd",
34    Description = "Rational quadratic covariance function with automatic relevance determination for Gaussian processes.")]
35  public sealed class CovarianceRationalQuadraticArd : ParameterizedNamedItem, ICovarianceFunction {
36    public IValueParameter<DoubleValue> ScaleParameter {
37      get { return (IValueParameter<DoubleValue>)Parameters["Scale"]; }
38    }
39
40    public IValueParameter<DoubleArray> InverseLengthParameter {
41      get { return (IValueParameter<DoubleArray>)Parameters["InverseLength"]; }
42    }
43
44    public IValueParameter<DoubleValue> ShapeParameter {
45      get { return (IValueParameter<DoubleValue>)Parameters["Shape"]; }
46    }
47    private bool HasFixedScaleParameter {
48      get { return ScaleParameter.Value != null; }
49    }
50    private bool HasFixedInverseLengthParameter {
51      get { return InverseLengthParameter.Value != null; }
52    }
53    private bool HasFixedShapeParameter {
54      get { return ShapeParameter.Value != null; }
55    }
56
57    [StorableConstructor]
58    private CovarianceRationalQuadraticArd(bool deserializing)
59      : base(deserializing) {
60    }
61
62    private CovarianceRationalQuadraticArd(CovarianceRationalQuadraticArd original, Cloner cloner)
63      : base(original, cloner) {
64    }
65
66    public CovarianceRationalQuadraticArd()
67      : base() {
68      Name = ItemName;
69      Description = ItemDescription;
70
71      Parameters.Add(new OptionalValueParameter<DoubleValue>("Scale", "The scale parameter of the rational quadratic covariance function with ARD."));
72      Parameters.Add(new OptionalValueParameter<DoubleArray>("InverseLength", "The inverse length parameter for automatic relevance determination."));
73      Parameters.Add(new OptionalValueParameter<DoubleValue>("Shape", "The shape parameter (alpha) of the rational quadratic covariance function with ARD."));
74    }
75
76    public override IDeepCloneable Clone(Cloner cloner) {
77      return new CovarianceRationalQuadraticArd(this, cloner);
78    }
79
80    public int GetNumberOfParameters(int numberOfVariables) {
81      return
82        (HasFixedScaleParameter ? 0 : 1) +
83        (HasFixedShapeParameter ? 0 : 1) +
84        (HasFixedInverseLengthParameter ? 0 : numberOfVariables);
85    }
86
87    public void SetParameter(double[] p) {
88      double scale, shape;
89      double[] inverseLength;
90      GetParameterValues(p, out scale, out shape, out inverseLength);
91      ScaleParameter.Value = new DoubleValue(scale);
92      ShapeParameter.Value = new DoubleValue(shape);
93      InverseLengthParameter.Value = new DoubleArray(inverseLength);
94    }
95
96    private void GetParameterValues(double[] p, out double scale, out double shape, out double[] inverseLength) {
97      int c = 0;
98      // gather parameter values
99      if (HasFixedInverseLengthParameter) {
100        inverseLength = InverseLengthParameter.Value.ToArray();
101      } else {
102        int length = p.Length;
103        if (!HasFixedScaleParameter) length--;
104        if (!HasFixedShapeParameter) length--;
105        inverseLength = p.Select(e => 1.0 / Math.Exp(e)).Take(length).ToArray();
106        c += inverseLength.Length;
107      }
108      if (HasFixedScaleParameter) {
109        scale = ScaleParameter.Value.Value;
110      } else {
111        scale = Math.Exp(2 * p[c]);
112        c++;
113      }
114      if (HasFixedShapeParameter) {
115        shape = ShapeParameter.Value.Value;
116      } else {
117        shape = Math.Exp(p[c]);
118        c++;
119      }
120      if (p.Length != c) throw new ArgumentException("The length of the parameter vector does not match the number of free parameters for CovarianceRationalQuadraticArd", "p");
121    }
122
123    public ParameterizedCovarianceFunction GetParameterizedCovarianceFunction(double[] p, int[] columnIndices) {
124      double scale, shape;
125      double[] inverseLength;
126      GetParameterValues(p, out scale, out shape, out inverseLength);
127      var fixedInverseLength = HasFixedInverseLengthParameter;
128      var fixedScale = HasFixedScaleParameter;
129      var fixedShape = HasFixedShapeParameter;
130      // create functions
131      var cov = new ParameterizedCovarianceFunction();
132      cov.Covariance = (x, i, j) => {
133        double d = i == j
134                    ? 0.0
135                    : Util.SqrDist(x, i, j, inverseLength, columnIndices);
136        return scale * Math.Pow(1 + 0.5 * d / shape, -shape);
137      };
138      cov.CrossCovariance = (x, xt, i, j) => {
139        double d = Util.SqrDist(x, i, xt, j, inverseLength, columnIndices);
140        return scale * Math.Pow(1 + 0.5 * d / shape, -shape);
141      };
142      cov.CovarianceGradient = (x, i, j) => GetGradient(x, i, j, columnIndices, scale, shape, inverseLength, fixedInverseLength, fixedScale, fixedShape);
143      return cov;
144    }
145
146    private static IList<double> GetGradient(double[,] x, int i, int j, int[] columnIndices, double scale, double shape, double[] inverseLength,
147      bool fixedInverseLength, bool fixedScale, bool fixedShape) {
148      double d = i == j
149                   ? 0.0
150                   : Util.SqrDist(x, i, j, inverseLength, columnIndices);
151      double b = 1 + 0.5 * d / shape;
152      int k = 0;
153      var g = new List<double>(columnIndices.Length + 2);
154      if (!fixedInverseLength) {
155        foreach (var columnIndex in columnIndices) {
156          g.Add(
157            scale * Math.Pow(b, -shape - 1) *
158            Util.SqrDist(x[i, columnIndex] * inverseLength[k], x[j, columnIndex] * inverseLength[k]));
159          k++;
160        }
161      }
162      if (!fixedScale) g.Add(2 * scale * Math.Pow(b, -shape));
163      if (!fixedShape) g.Add(scale * Math.Pow(b, -shape) * (0.5 * d / b - shape * Math.Log(b)));
164      return g;
165    }
166  }
167}
Note: See TracBrowser for help on using the repository browser.