source: trunk/sources/HeuristicLab.Algorithms.DataAnalysis/3.4/NeuralNetwork/NeuralNetworkRegression.cs @ 14185

Last change on this file since 14185 was 14185, checked in by swagner, 3 years ago

#2526: Updated year of copyrights in license headers

File size: 11.2 KB
Line 
1#region License Information
2/* HeuristicLab
3 * Copyright (C) 2002-2016 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.Optimization;
29using HeuristicLab.Parameters;
30using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
31using HeuristicLab.Problems.DataAnalysis;
32
33namespace HeuristicLab.Algorithms.DataAnalysis {
34  /// <summary>
35  /// Neural network regression data analysis algorithm.
36  /// </summary>
37  [Item("Neural Network Regression (NN)", "Neural network regression data analysis algorithm (wrapper for ALGLIB). Further documentation: http://www.alglib.net/dataanalysis/neuralnetworks.php")]
38  [Creatable(CreatableAttribute.Categories.DataAnalysisRegression, Priority = 130)]
39  [StorableClass]
40  public sealed class NeuralNetworkRegression : FixedDataAnalysisAlgorithm<IRegressionProblem> {
41    private const string DecayParameterName = "Decay";
42    private const string HiddenLayersParameterName = "HiddenLayers";
43    private const string NodesInFirstHiddenLayerParameterName = "NodesInFirstHiddenLayer";
44    private const string NodesInSecondHiddenLayerParameterName = "NodesInSecondHiddenLayer";
45    private const string RestartsParameterName = "Restarts";
46    private const string NeuralNetworkRegressionModelResultName = "Neural network regression solution";
47
48    #region parameter properties
49    public IFixedValueParameter<DoubleValue> DecayParameter {
50      get { return (IFixedValueParameter<DoubleValue>)Parameters[DecayParameterName]; }
51    }
52    public IConstrainedValueParameter<IntValue> HiddenLayersParameter {
53      get { return (IConstrainedValueParameter<IntValue>)Parameters[HiddenLayersParameterName]; }
54    }
55    public IFixedValueParameter<IntValue> NodesInFirstHiddenLayerParameter {
56      get { return (IFixedValueParameter<IntValue>)Parameters[NodesInFirstHiddenLayerParameterName]; }
57    }
58    public IFixedValueParameter<IntValue> NodesInSecondHiddenLayerParameter {
59      get { return (IFixedValueParameter<IntValue>)Parameters[NodesInSecondHiddenLayerParameterName]; }
60    }
61    public IFixedValueParameter<IntValue> RestartsParameter {
62      get { return (IFixedValueParameter<IntValue>)Parameters[RestartsParameterName]; }
63    }
64    #endregion
65
66    #region properties
67    public double Decay {
68      get { return DecayParameter.Value.Value; }
69      set {
70        if (value < 0.001 || value > 100) throw new ArgumentException("The decay parameter should be set to a value between 0.001 and 100.", "Decay");
71        DecayParameter.Value.Value = value;
72      }
73    }
74    public int HiddenLayers {
75      get { return HiddenLayersParameter.Value.Value; }
76      set {
77        if (value < 0 || value > 2) throw new ArgumentException("The number of hidden layers should be set to 0, 1, or 2.", "HiddenLayers");
78        HiddenLayersParameter.Value = (from v in HiddenLayersParameter.ValidValues
79                                       where v.Value == value
80                                       select v)
81                                      .Single();
82      }
83    }
84    public int NodesInFirstHiddenLayer {
85      get { return NodesInFirstHiddenLayerParameter.Value.Value; }
86      set {
87        if (value < 1) throw new ArgumentException("The number of nodes in the first hidden layer must be at least one.", "NodesInFirstHiddenLayer");
88        NodesInFirstHiddenLayerParameter.Value.Value = value;
89      }
90    }
91    public int NodesInSecondHiddenLayer {
92      get { return NodesInSecondHiddenLayerParameter.Value.Value; }
93      set {
94        if (value < 1) throw new ArgumentException("The number of nodes in the first second layer must be at least one.", "NodesInSecondHiddenLayer");
95        NodesInSecondHiddenLayerParameter.Value.Value = value;
96      }
97    }
98    public int Restarts {
99      get { return RestartsParameter.Value.Value; }
100      set {
101        if (value < 0) throw new ArgumentException("The number of restarts must be positive.", "Restarts");
102        RestartsParameter.Value.Value = value;
103      }
104    }
105    #endregion
106
107
108    [StorableConstructor]
109    private NeuralNetworkRegression(bool deserializing) : base(deserializing) { }
110    private NeuralNetworkRegression(NeuralNetworkRegression original, Cloner cloner)
111      : base(original, cloner) {
112      RegisterEventHandlers();
113    }
114    public NeuralNetworkRegression()
115      : base() {
116      var validHiddenLayerValues = new ItemSet<IntValue>(new IntValue[] { 
117        (IntValue)new IntValue(0).AsReadOnly(), 
118        (IntValue)new IntValue(1).AsReadOnly(), 
119        (IntValue)new IntValue(2).AsReadOnly() });
120      var selectedHiddenLayerValue = (from v in validHiddenLayerValues
121                                      where v.Value == 1
122                                      select v)
123                                     .Single();
124      Parameters.Add(new FixedValueParameter<DoubleValue>(DecayParameterName, "The decay parameter for the training phase of the neural network. This parameter determines the strengh of regularization and should be set to a value between 0.001 (weak regularization) to 100 (very strong regularization). The correct value should be determined via cross-validation.", new DoubleValue(1)));
125      Parameters.Add(new ConstrainedValueParameter<IntValue>(HiddenLayersParameterName, "The number of hidden layers for the neural network (0, 1, or 2)", validHiddenLayerValues, selectedHiddenLayerValue));
126      Parameters.Add(new FixedValueParameter<IntValue>(NodesInFirstHiddenLayerParameterName, "The number of nodes in the first hidden layer. This value is not used if the number of hidden layers is zero.", new IntValue(10)));
127      Parameters.Add(new FixedValueParameter<IntValue>(NodesInSecondHiddenLayerParameterName, "The number of nodes in the second hidden layer. This value is not used if the number of hidden layers is zero or one.", new IntValue(10)));
128      Parameters.Add(new FixedValueParameter<IntValue>(RestartsParameterName, "The number of restarts for learning.", new IntValue(2)));
129
130      RestartsParameter.Hidden = true;
131      NodesInSecondHiddenLayerParameter.Hidden = true;
132
133      RegisterEventHandlers();
134
135      Problem = new RegressionProblem();
136    }
137
138    private void RegisterEventHandlers() {
139      HiddenLayersParameter.Value.ValueChanged += HiddenLayersParameterValueValueChanged;
140      HiddenLayersParameter.ValueChanged += HiddenLayersParameterValueChanged;
141    }
142
143    [StorableHook(HookType.AfterDeserialization)]
144    private void AfterDeserialization() {
145      RegisterEventHandlers();
146    }
147
148    public override IDeepCloneable Clone(Cloner cloner) {
149      return new NeuralNetworkRegression(this, cloner);
150    }
151
152    private void HiddenLayersParameterValueChanged(object source, EventArgs e) {
153      HiddenLayersParameter.Value.ValueChanged += HiddenLayersParameterValueValueChanged;
154      HiddenLayersParameterValueValueChanged(this, EventArgs.Empty);
155    }
156
157    private void HiddenLayersParameterValueValueChanged(object source, EventArgs e) {
158      if (HiddenLayers == 0) {
159        NodesInFirstHiddenLayerParameter.Hidden = true;
160        NodesInSecondHiddenLayerParameter.Hidden = true;
161      } else if (HiddenLayers == 1) {
162        NodesInFirstHiddenLayerParameter.Hidden = false;
163        NodesInSecondHiddenLayerParameter.Hidden = true;
164      } else {
165        NodesInFirstHiddenLayerParameter.Hidden = false;
166        NodesInSecondHiddenLayerParameter.Hidden = false;
167      }
168    }
169
170
171    #region neural network
172    protected override void Run() {
173      double rmsError, avgRelError;
174      var solution = CreateNeuralNetworkRegressionSolution(Problem.ProblemData, HiddenLayers, NodesInFirstHiddenLayer, NodesInSecondHiddenLayer, Decay, Restarts, out rmsError, out avgRelError);
175      Results.Add(new Result(NeuralNetworkRegressionModelResultName, "The neural network regression solution.", solution));
176      Results.Add(new Result("Root mean square error", "The root of the mean of squared errors of the neural network regression solution on the training set.", new DoubleValue(rmsError)));
177      Results.Add(new Result("Average relative error", "The average of relative errors of the neural network regression solution on the training set.", new PercentValue(avgRelError)));
178    }
179
180    public static IRegressionSolution CreateNeuralNetworkRegressionSolution(IRegressionProblemData problemData, int nLayers, int nHiddenNodes1, int nHiddenNodes2, double decay, int restarts,
181      out double rmsError, out double avgRelError) {
182      var dataset = problemData.Dataset;
183      string targetVariable = problemData.TargetVariable;
184      IEnumerable<string> allowedInputVariables = problemData.AllowedInputVariables;
185      IEnumerable<int> rows = problemData.TrainingIndices;
186      double[,] inputMatrix = AlglibUtil.PrepareInputMatrix(dataset, allowedInputVariables.Concat(new string[] { targetVariable }), rows);
187      if (inputMatrix.Cast<double>().Any(x => double.IsNaN(x) || double.IsInfinity(x)))
188        throw new NotSupportedException("Neural network regression does not support NaN or infinity values in the input dataset.");
189
190      alglib.multilayerperceptron multiLayerPerceptron = null;
191      if (nLayers == 0) {
192        alglib.mlpcreate0(allowedInputVariables.Count(), 1, out multiLayerPerceptron);
193      } else if (nLayers == 1) {
194        alglib.mlpcreate1(allowedInputVariables.Count(), nHiddenNodes1, 1, out multiLayerPerceptron);
195      } else if (nLayers == 2) {
196        alglib.mlpcreate2(allowedInputVariables.Count(), nHiddenNodes1, nHiddenNodes2, 1, out multiLayerPerceptron);
197      } else throw new ArgumentException("Number of layers must be zero, one, or two.", "nLayers");
198      alglib.mlpreport rep;
199      int nRows = inputMatrix.GetLength(0);
200
201      int info;
202      // using mlptrainlm instead of mlptraines or mlptrainbfgs because only one parameter is necessary
203      alglib.mlptrainlm(multiLayerPerceptron, inputMatrix, nRows, decay, restarts, out info, out rep);
204      if (info != 2) throw new ArgumentException("Error in calculation of neural network regression solution");
205
206      rmsError = alglib.mlprmserror(multiLayerPerceptron, inputMatrix, nRows);
207      avgRelError = alglib.mlpavgrelerror(multiLayerPerceptron, inputMatrix, nRows);
208
209      return new NeuralNetworkRegressionSolution(new NeuralNetworkModel(multiLayerPerceptron, targetVariable, allowedInputVariables), (IRegressionProblemData)problemData.Clone());
210    }
211    #endregion
212  }
213}
Note: See TracBrowser for help on using the repository browser.