source: trunk/sources/HeuristicLab.Problems.DataAnalysis.Symbolic.Classification.Views/3.4/InteractiveSymbolicDiscriminantFunctionClassificationSolutionSimplifierView.cs @ 7027

Last change on this file since 7027 was 7027, checked in by gkronber, 11 years ago

#1685: fixed a bug in the InteractiveSymbolicDiscriminantFunctionClassificationSolutionSimplifierView. The drawback is that the threshold calculation is much more complex now and can lead to an unresponsive UI as we do not calculate impacts in the background.

File size: 7.6 KB
Line 
1#region License Information
2/* HeuristicLab
3 * Copyright (C) 2002-2011 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.Encodings.SymbolicExpressionTreeEncoding;
27using HeuristicLab.Problems.DataAnalysis.Symbolic.Views;
28
29namespace HeuristicLab.Problems.DataAnalysis.Symbolic.Classification.Views {
30  public partial class InteractiveSymbolicDiscriminantFunctionClassificationSolutionSimplifierView : InteractiveSymbolicDataAnalysisSolutionSimplifierView {
31    private readonly ConstantTreeNode constantNode;
32    private readonly SymbolicExpressionTree tempTree;
33
34    public new SymbolicDiscriminantFunctionClassificationSolution Content {
35      get { return (SymbolicDiscriminantFunctionClassificationSolution)base.Content; }
36      set { base.Content = value; }
37    }
38
39    public InteractiveSymbolicDiscriminantFunctionClassificationSolutionSimplifierView()
40      : base() {
41      InitializeComponent();
42      this.Caption = "Interactive Classification Solution Simplifier";
43
44      constantNode = ((ConstantTreeNode)new Constant().CreateTreeNode());
45      ISymbolicExpressionTreeNode root = new ProgramRootSymbol().CreateTreeNode();
46      ISymbolicExpressionTreeNode start = new StartSymbol().CreateTreeNode();
47      root.AddSubtree(start);
48      tempTree = new SymbolicExpressionTree(root);
49    }
50
51    protected override void UpdateModel(ISymbolicExpressionTree tree) {
52      Content.Model = new SymbolicDiscriminantFunctionClassificationModel(tree, Content.Model.Interpreter);
53      // the default policy for setting thresholds in classification models is the accuarcy maximizing policy.
54      // This is rather slow to calculate and can lead to a very laggy UI in the interactive solution simplifier.
55      // However, since we automatically prune sub-trees based on the threshold reaching the maximum accuracy we must
56      // also use maximum accuracy threshold calculation here in order to prevent incoherent behavior of the simplifier.
57      Content.SetAccuracyMaximizingThresholds();
58    }
59
60    protected override Dictionary<ISymbolicExpressionTreeNode, double> CalculateReplacementValues(ISymbolicExpressionTree tree) {
61      Dictionary<ISymbolicExpressionTreeNode, double> replacementValues = new Dictionary<ISymbolicExpressionTreeNode, double>();
62      foreach (ISymbolicExpressionTreeNode node in tree.Root.GetSubtree(0).GetSubtree(0).IterateNodesPrefix()) {
63        replacementValues[node] = CalculateReplacementValue(node, tree);
64      }
65      return replacementValues;
66    }
67
68    protected override Dictionary<ISymbolicExpressionTreeNode, double> CalculateImpactValues(ISymbolicExpressionTree tree) {
69      var interpreter = Content.Model.Interpreter;
70      var dataset = Content.ProblemData.Dataset;
71      var rows = Content.ProblemData.TrainingIndizes;
72      string targetVariable = Content.ProblemData.TargetVariable;
73      Dictionary<ISymbolicExpressionTreeNode, double> impactValues = new Dictionary<ISymbolicExpressionTreeNode, double>();
74      List<ISymbolicExpressionTreeNode> nodes = tree.Root.GetSubtree(0).GetSubtree(0).IterateNodesPostfix().ToList();
75
76      var targetClassValues = dataset.GetDoubleValues(targetVariable, rows);
77      var originalOutput = interpreter.GetSymbolicExpressionTreeValues(tree, dataset, rows)
78        .LimitToRange(Content.Model.LowerEstimationLimit, Content.Model.UpperEstimationLimit)
79        .ToArray();
80      double[] classValues;
81      double[] thresholds;
82      // normal distribution cut points are used as thresholds here because they are a lot faster to calculate than the accuracy maximizing thresholds
83      NormalDistributionCutPointsThresholdCalculator.CalculateThresholds(Content.ProblemData, originalOutput, targetClassValues, out classValues, out thresholds);
84      var classifier = new SymbolicDiscriminantFunctionClassificationModel(tree, interpreter);
85      classifier.SetThresholdsAndClassValues(thresholds, classValues);
86      OnlineCalculatorError errorState;
87      double originalAccuracy = OnlineAccuracyCalculator.Calculate(targetClassValues, classifier.GetEstimatedClassValues(dataset, rows), out errorState);
88      if (errorState != OnlineCalculatorError.None) originalAccuracy = 0.0;
89
90      foreach (ISymbolicExpressionTreeNode node in nodes) {
91        var parent = node.Parent;
92        constantNode.Value = CalculateReplacementValue(node, tree);
93        ISymbolicExpressionTreeNode replacementNode = constantNode;
94        SwitchNode(parent, node, replacementNode);
95        var newOutput = interpreter.GetSymbolicExpressionTreeValues(tree, dataset, rows)
96          .LimitToRange(Content.Model.LowerEstimationLimit, Content.Model.UpperEstimationLimit)
97          .ToArray();
98        NormalDistributionCutPointsThresholdCalculator.CalculateThresholds(Content.ProblemData, newOutput, targetClassValues, out classValues, out thresholds);
99        classifier = new SymbolicDiscriminantFunctionClassificationModel(tree, interpreter);
100        classifier.SetThresholdsAndClassValues(thresholds, classValues);
101        double newAccuracy = OnlineAccuracyCalculator.Calculate(targetClassValues, classifier.GetEstimatedClassValues(dataset, rows), out errorState);
102        if (errorState != OnlineCalculatorError.None) newAccuracy = 0.0;
103
104        // impact = 0 if no change
105        // impact < 0 if new solution is better
106        // impact > 0 if new solution is worse
107        impactValues[node] = originalAccuracy - newAccuracy;
108        SwitchNode(parent, replacementNode, node);
109      }
110      return impactValues;
111    }
112
113    private double CalculateReplacementValue(ISymbolicExpressionTreeNode node, ISymbolicExpressionTree sourceTree) {
114      // remove old ADFs
115      while (tempTree.Root.SubtreeCount > 1) tempTree.Root.RemoveSubtree(1);
116      // clone ADFs of source tree
117      for (int i = 1; i < sourceTree.Root.SubtreeCount; i++) {
118        tempTree.Root.AddSubtree((ISymbolicExpressionTreeNode)sourceTree.Root.GetSubtree(i).Clone());
119      }
120      var start = tempTree.Root.GetSubtree(0);
121      while (start.SubtreeCount > 0) start.RemoveSubtree(0);
122      start.AddSubtree((ISymbolicExpressionTreeNode)node.Clone());
123      var interpreter = Content.Model.Interpreter;
124      var rows = Content.ProblemData.TrainingIndizes;
125      return interpreter.GetSymbolicExpressionTreeValues(tempTree, Content.ProblemData.Dataset, rows).Median();
126    }
127
128
129    private void SwitchNode(ISymbolicExpressionTreeNode root, ISymbolicExpressionTreeNode oldBranch, ISymbolicExpressionTreeNode newBranch) {
130      for (int i = 0; i < root.SubtreeCount; i++) {
131        if (root.GetSubtree(i) == oldBranch) {
132          root.RemoveSubtree(i);
133          root.InsertSubtree(i, newBranch);
134          return;
135        }
136      }
137    }
138
139    protected override void btnOptimizeConstants_Click(object sender, EventArgs e) {
140
141    }
142  }
143}
Note: See TracBrowser for help on using the repository browser.