source: trunk/sources/HeuristicLab.Algorithms.DataAnalysis/3.4/GaussianProcess/CovarianceFunctions/CovarianceNeuralNetwork.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: 5.8 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 HeuristicLab.Common;
25using HeuristicLab.Core;
26using HeuristicLab.Data;
27using HeuristicLab.Parameters;
28using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
29
30namespace HeuristicLab.Algorithms.DataAnalysis {
31  [StorableClass]
32  [Item(Name = "CovarianceNeuralNetwork",
33    Description = "Neural network covariance function for Gaussian processes.")]
34  public sealed class CovarianceNeuralNetwork : ParameterizedNamedItem, ICovarianceFunction {
35    public IValueParameter<DoubleValue> ScaleParameter {
36      get { return (IValueParameter<DoubleValue>)Parameters["Scale"]; }
37    }
38
39    public IValueParameter<DoubleValue> LengthParameter {
40      get { return (IValueParameter<DoubleValue>)Parameters["Length"]; }
41    }
42    private bool HasFixedScaleParameter {
43      get { return ScaleParameter.Value != null; }
44    }
45    private bool HasFixedLengthParameter {
46      get { return LengthParameter.Value != null; }
47    }
48
49    [StorableConstructor]
50    private CovarianceNeuralNetwork(bool deserializing)
51      : base(deserializing) {
52    }
53
54    private CovarianceNeuralNetwork(CovarianceNeuralNetwork original, Cloner cloner)
55      : base(original, cloner) {
56    }
57
58    public CovarianceNeuralNetwork()
59      : base() {
60      Name = ItemName;
61      Description = ItemDescription;
62
63      Parameters.Add(new OptionalValueParameter<DoubleValue>("Scale", "The scale parameter."));
64      Parameters.Add(new OptionalValueParameter<DoubleValue>("Length", "The length parameter."));
65    }
66
67    public override IDeepCloneable Clone(Cloner cloner) {
68      return new CovarianceNeuralNetwork(this, cloner);
69    }
70
71    public int GetNumberOfParameters(int numberOfVariables) {
72      return
73        (HasFixedScaleParameter ? 0 : 1) +
74        (HasFixedLengthParameter ? 0 : 1);
75    }
76
77    public void SetParameter(double[] p) {
78      double scale, length;
79      GetParameterValues(p, out scale, out length);
80      ScaleParameter.Value = new DoubleValue(scale);
81      LengthParameter.Value = new DoubleValue(length);
82    }
83
84
85    private void GetParameterValues(double[] p, out double scale, out double length) {
86      // gather parameter values
87      int c = 0;
88      if (HasFixedLengthParameter) {
89        length = LengthParameter.Value.Value;
90      } else {
91        length = Math.Exp(2 * p[c]);
92        c++;
93      }
94
95      if (HasFixedScaleParameter) {
96        scale = ScaleParameter.Value.Value;
97      } else {
98        scale = Math.Exp(2 * p[c]);
99        c++;
100      }
101      if (p.Length != c) throw new ArgumentException("The length of the parameter vector does not match the number of free parameters for CovarianceNeuralNetwork", "p");
102    }
103
104    public ParameterizedCovarianceFunction GetParameterizedCovarianceFunction(double[] p, int[] columnIndices) {
105      double length, scale;
106      GetParameterValues(p, out scale, out length);
107      var fixedLength = HasFixedLengthParameter;
108      var fixedScale = HasFixedScaleParameter;
109
110      var cov = new ParameterizedCovarianceFunction();
111      cov.Covariance = (x, i, j) => {
112        double sx = 1.0;
113        double s1 = 1.0;
114        double s2 = 1.0;
115        for (int c = 0; c < columnIndices.Length; c++) {
116          var col = columnIndices[c];
117          sx += x[i, col] * x[j, col];
118          s1 += x[i, col] * x[i, col];
119          s2 += x[j, col] * x[j, col];
120        }
121
122        return (scale * Math.Asin(sx / (Math.Sqrt((length + s1) * (length + s2)))));
123      };
124      cov.CrossCovariance = (x, xt, i, j) => {
125        double sx = 1.0;
126        double s1 = 1.0;
127        double s2 = 1.0;
128        for (int c = 0; c < columnIndices.Length; c++) {
129          var col = columnIndices[c];
130          sx += x[i, col] * xt[j, col];
131          s1 += x[i, col] * x[i, col];
132          s2 += xt[j, col] * xt[j, col];
133        }
134
135        return (scale * Math.Asin(sx / (Math.Sqrt((length + s1) * (length + s2)))));
136      };
137      cov.CovarianceGradient = (x, i, j) => GetGradient(x, i, j, length, scale, columnIndices, fixedLength, fixedScale);
138      return cov;
139    }
140
141    // order of returned gradients must match the order in GetParameterValues!
142    private static IList<double> GetGradient(double[,] x, int i, int j, double length, double scale, int[] columnIndices,
143      bool fixedLength, bool fixedScale) {
144      double sx = 1.0;
145      double s1 = 1.0;
146      double s2 = 1.0;
147      for (int c = 0; c < columnIndices.Length; c++) {
148        var col = columnIndices[c];
149        sx += x[i, col] * x[j, col];
150        s1 += x[i, col] * x[i, col];
151        s2 += x[j, col] * x[j, col];
152      }
153      var h = (length + s1) * (length + s2);
154      var f = sx / Math.Sqrt(h);
155
156      var g = new List<double>(2);
157      if (!fixedLength) g.Add(-scale / Math.Sqrt(1.0 - f * f) * ((length * sx * (2.0 * length + s1 + s2)) / Math.Pow(h, 3.0 / 2.0)));
158      if (!fixedScale) g.Add(2.0 * scale * Math.Asin(f));
159      return g;
160    }
161  }
162}
Note: See TracBrowser for help on using the repository browser.