1 | #region License Information
|
---|
2 | /* HeuristicLab
|
---|
3 | * Copyright (C) 2002-2010 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 |
|
---|
22 | using System.Collections.Generic;
|
---|
23 | using System.Linq;
|
---|
24 | using System.Windows.Forms;
|
---|
25 | using alglib;
|
---|
26 | using HeuristicLab.Common;
|
---|
27 | using HeuristicLab.Data;
|
---|
28 | using HeuristicLab.MainForm;
|
---|
29 | using HeuristicLab.MainForm.WindowsForms;
|
---|
30 | using HeuristicLab.Optimization;
|
---|
31 | using System;
|
---|
32 | using HeuristicLab.Problems.DataAnalysis.Regression.Symbolic;
|
---|
33 | using HeuristicLab.Problems.DataAnalysis.Symbolic.Symbols;
|
---|
34 | using HeuristicLab.Problems.DataAnalysis.Evaluators;
|
---|
35 |
|
---|
36 | namespace HeuristicLab.Problems.DataAnalysis.Views {
|
---|
37 | [Content(typeof(RunCollection), false)]
|
---|
38 | [View("RunCollection Monte-Carlo Variable Impact View")]
|
---|
39 | public partial class RunCollectionMonteCarloVariableImpactView : AsynchronousContentView {
|
---|
40 | private const string validationBestModelResultName = "Best solution (on validation set)";
|
---|
41 | public RunCollectionMonteCarloVariableImpactView() {
|
---|
42 | InitializeComponent();
|
---|
43 | }
|
---|
44 |
|
---|
45 | public new RunCollection Content {
|
---|
46 | get { return (RunCollection)base.Content; }
|
---|
47 | set { base.Content = value; }
|
---|
48 | }
|
---|
49 |
|
---|
50 | protected override void RegisterContentEvents() {
|
---|
51 | base.RegisterContentEvents();
|
---|
52 | this.Content.ItemsAdded += new HeuristicLab.Collections.CollectionItemsChangedEventHandler<IRun>(Content_ItemsAdded);
|
---|
53 | this.Content.ItemsRemoved += new HeuristicLab.Collections.CollectionItemsChangedEventHandler<IRun>(Content_ItemsRemoved);
|
---|
54 | this.Content.CollectionReset += new HeuristicLab.Collections.CollectionItemsChangedEventHandler<IRun>(Content_CollectionReset);
|
---|
55 | }
|
---|
56 | protected override void DeregisterContentEvents() {
|
---|
57 | base.RegisterContentEvents();
|
---|
58 | this.Content.ItemsAdded -= new HeuristicLab.Collections.CollectionItemsChangedEventHandler<IRun>(Content_ItemsAdded);
|
---|
59 | this.Content.ItemsRemoved -= new HeuristicLab.Collections.CollectionItemsChangedEventHandler<IRun>(Content_ItemsRemoved);
|
---|
60 | this.Content.CollectionReset -= new HeuristicLab.Collections.CollectionItemsChangedEventHandler<IRun>(Content_CollectionReset);
|
---|
61 | }
|
---|
62 |
|
---|
63 | protected override void OnContentChanged() {
|
---|
64 | base.OnContentChanged();
|
---|
65 | this.UpdateData();
|
---|
66 | }
|
---|
67 | private void Content_ItemsAdded(object sender, HeuristicLab.Collections.CollectionItemsChangedEventArgs<IRun> e) {
|
---|
68 | this.UpdateData();
|
---|
69 | }
|
---|
70 | private void Content_ItemsRemoved(object sender, HeuristicLab.Collections.CollectionItemsChangedEventArgs<IRun> e) {
|
---|
71 | this.UpdateData();
|
---|
72 | }
|
---|
73 | private void Content_CollectionReset(object sender, HeuristicLab.Collections.CollectionItemsChangedEventArgs<IRun> e) {
|
---|
74 | this.UpdateData();
|
---|
75 | }
|
---|
76 |
|
---|
77 | private void UpdateData() {
|
---|
78 | matrixView.Content = CalculateVariableImpactMatrix();
|
---|
79 | }
|
---|
80 |
|
---|
81 | private DoubleMatrix CalculateVariableImpactMatrix() {
|
---|
82 | DoubleMatrix matrix = null;
|
---|
83 | if (Content != null) {
|
---|
84 | List<IRun> runsWithSolutions = (from run in Content
|
---|
85 | where run.Results.ContainsKey(validationBestModelResultName)
|
---|
86 | select run)
|
---|
87 | .ToList();
|
---|
88 | IEnumerable<SymbolicRegressionSolution> allSolutions = (from run in Content
|
---|
89 | where run.Results.ContainsKey(validationBestModelResultName)
|
---|
90 | select run.Results[validationBestModelResultName]).Cast<SymbolicRegressionSolution>();
|
---|
91 |
|
---|
92 | Dictionary<SymbolicRegressionSolution, IEnumerable<string>> variableReferences = new Dictionary<SymbolicRegressionSolution, IEnumerable<string>>();
|
---|
93 | foreach (var solution in allSolutions) {
|
---|
94 | variableReferences[solution] = GetVariableReferences(solution).Distinct();
|
---|
95 | }
|
---|
96 |
|
---|
97 | List<string> variableNames = (from modelVarRefs in variableReferences.Values
|
---|
98 | from variableName in modelVarRefs
|
---|
99 | select variableName)
|
---|
100 | .Distinct()
|
---|
101 | .ToList();
|
---|
102 |
|
---|
103 | List<string> statictics = new List<string> { "Median Impact", "Mean Impact", "StdDev", "pValue" };
|
---|
104 | List<string> columnNames = (from run in runsWithSolutions
|
---|
105 | select run.Name).ToList();
|
---|
106 | columnNames.AddRange(statictics);
|
---|
107 |
|
---|
108 | matrix = new DoubleMatrix(variableNames.Count, columnNames.Count);
|
---|
109 | matrix.SortableView = true;
|
---|
110 | matrix.RowNames = variableNames;
|
---|
111 | matrix.ColumnNames = columnNames;
|
---|
112 | Random random = new Random();
|
---|
113 | int columnIndex = 0;
|
---|
114 | foreach (SymbolicRegressionSolution solution in variableReferences.Keys) {
|
---|
115 | foreach (string variableName in variableReferences[solution]) {
|
---|
116 | int rowIndex = variableNames.IndexOf(variableName);
|
---|
117 | if (rowIndex > -1) {
|
---|
118 | matrix[rowIndex, columnIndex] = ApproximatePermutationImpact(random, variableName, solution);
|
---|
119 | }
|
---|
120 | }
|
---|
121 | columnIndex++;
|
---|
122 | }
|
---|
123 | List<List<double>> variableImpactValues = (from row in Enumerable.Range(0, variableNames.Count())
|
---|
124 | select GetRowValues(matrix, row).ToList())
|
---|
125 | .ToList();
|
---|
126 | List<double> referenceValues = (from variableImpacts in variableImpactValues
|
---|
127 | orderby variableImpacts.Average()
|
---|
128 | select variableImpacts)
|
---|
129 | .First();
|
---|
130 | for (int row = 0; row < variableNames.Count; row++) {
|
---|
131 | List<double> rowValues = variableImpactValues[row];
|
---|
132 | matrix[row, columnIndex] = rowValues.Median();
|
---|
133 | matrix[row, columnIndex + 1] = rowValues.Average();
|
---|
134 | matrix[row, columnIndex + 2] = rowValues.StandardDeviation();
|
---|
135 |
|
---|
136 | double bothTails, leftTail, rightTail;
|
---|
137 | bothTails = leftTail = rightTail = 0.0;
|
---|
138 | double[] z = new double[rowValues.Count()];
|
---|
139 | for (int i = 0; i < z.Length; i++) {
|
---|
140 | z[i] = rowValues[i] - referenceValues[i];
|
---|
141 | }
|
---|
142 | alglib.wsr.wilcoxonsignedranktest(z, z.Length, 0.0, ref bothTails, ref leftTail, ref rightTail);
|
---|
143 | matrix[row, columnIndex + 3] = bothTails;
|
---|
144 | }
|
---|
145 | }
|
---|
146 | return matrix;
|
---|
147 | }
|
---|
148 |
|
---|
149 | private IEnumerable<double> GetRowValues(DoubleMatrix matrix, int row) {
|
---|
150 | return from col in Enumerable.Range(0, matrix.Columns)
|
---|
151 | select matrix[row, col];
|
---|
152 | }
|
---|
153 |
|
---|
154 | private IEnumerable<string> GetVariableReferences(SymbolicRegressionSolution solution) {
|
---|
155 | return from node in solution.Model.SymbolicExpressionTree.IterateNodesPostfix().OfType<VariableTreeNode>()
|
---|
156 | select node.VariableName;
|
---|
157 | }
|
---|
158 |
|
---|
159 | private double ApproximatePermutationImpact(Random random, string variableName, SymbolicRegressionSolution solution) {
|
---|
160 | int permutations = 10;
|
---|
161 | int variableIndex = solution.ProblemData.Dataset.GetVariableIndex(variableName);
|
---|
162 | List<double> originalOutput = new List<double>(solution.EstimatedValues);
|
---|
163 | Dataset originalDataset = solution.ProblemData.Dataset;
|
---|
164 |
|
---|
165 | int rows = solution.ProblemData.Dataset.Rows;
|
---|
166 | int columns = solution.ProblemData.Dataset.Columns;
|
---|
167 | List<int> rowIndexPermutation = Enumerable.Range(0, rows).ToList();
|
---|
168 | double mseSum = 0.0;
|
---|
169 | for (int rep = 0; rep < permutations; rep++) {
|
---|
170 | double[,] manipulatedData = new double[rows, columns];
|
---|
171 | Shuffle(random, rowIndexPermutation);
|
---|
172 | for (int row = 0; row < rows; row++) {
|
---|
173 | for (int column = 0; column < columns; column++) {
|
---|
174 | if (column == variableIndex) {
|
---|
175 | manipulatedData[row, column] = solution.ProblemData.Dataset[row, column];
|
---|
176 | } else {
|
---|
177 | manipulatedData[row, column] = solution.ProblemData.Dataset[rowIndexPermutation[row], column];
|
---|
178 | }
|
---|
179 | }
|
---|
180 | }
|
---|
181 |
|
---|
182 | Dataset manipulatedDataset = new Dataset(solution.ProblemData.Dataset.VariableNames, manipulatedData);
|
---|
183 | solution.ProblemData.Dataset = manipulatedDataset;
|
---|
184 | double mse = SimpleMSEEvaluator.Calculate(originalOutput, solution.EstimatedValues);
|
---|
185 | mseSum += mse;
|
---|
186 | }
|
---|
187 |
|
---|
188 | solution.ProblemData.Dataset = originalDataset;
|
---|
189 | return mseSum / permutations;
|
---|
190 | }
|
---|
191 |
|
---|
192 | private void Shuffle(Random random, List<int> xs) {
|
---|
193 | for (int i = xs.Count; i > 1; i--) {
|
---|
194 | int j = random.Next(i);
|
---|
195 | int tmp = xs[j];
|
---|
196 | xs[j] = xs[i - 1];
|
---|
197 | xs[i - 1] = tmp;
|
---|
198 | }
|
---|
199 | }
|
---|
200 | }
|
---|
201 | }
|
---|