[13720] | 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 |
|
---|
| 22 | using HeuristicLab.Analysis;
|
---|
| 23 | using HeuristicLab.Analysis.QualityAnalysis;
|
---|
| 24 | using HeuristicLab.Common;
|
---|
| 25 | using HeuristicLab.Core;
|
---|
| 26 | using HeuristicLab.Data;
|
---|
| 27 | using HeuristicLab.MainForm;
|
---|
| 28 | using HeuristicLab.Optimization;
|
---|
| 29 | using HeuristicLab.OptimizationExpertSystem.Common;
|
---|
[13743] | 30 | using HeuristicLab.Random;
|
---|
[13720] | 31 | using System;
|
---|
| 32 | using System.Collections.Generic;
|
---|
| 33 | using System.Linq;
|
---|
| 34 | using System.Windows.Forms;
|
---|
| 35 | using System.Windows.Forms.DataVisualization.Charting;
|
---|
| 36 |
|
---|
| 37 | namespace HeuristicLab.OptimizationExpertSystem {
|
---|
| 38 | [View("Understanding Solutions")]
|
---|
[13722] | 39 | [Content(typeof(KnowledgeCenter), IsDefaultView = false)]
|
---|
| 40 | public partial class UnderstandingSolutionsView : KnowledgeCenterViewBase {
|
---|
[13720] | 41 | protected virtual bool SuppressEvents { get; set; }
|
---|
| 42 |
|
---|
| 43 | public UnderstandingSolutionsView() {
|
---|
| 44 | InitializeComponent();
|
---|
[13743] | 45 | solutionNetworkProjectionComboBox.SelectedIndex = 0;
|
---|
[13720] | 46 | }
|
---|
| 47 |
|
---|
| 48 | protected override void OnContentChanged() {
|
---|
| 49 | base.OnContentChanged();
|
---|
| 50 | UpdateSimilarityCalculators();
|
---|
| 51 | UpdateNamesComboboxes();
|
---|
| 52 | UpdateSolutionVisualization();
|
---|
| 53 | }
|
---|
| 54 |
|
---|
| 55 | protected override void OnProblemChanged() {
|
---|
| 56 | base.OnProblemInstancesChanged();
|
---|
| 57 | UpdateSimilarityCalculators();
|
---|
| 58 | UpdateNamesComboboxes();
|
---|
| 59 | UpdateSolutionVisualization();
|
---|
| 60 | }
|
---|
| 61 |
|
---|
| 62 | protected override void OnProblemSolutionsChanged() {
|
---|
| 63 | base.OnProblemSolutionsChanged();
|
---|
| 64 | UpdateNamesComboboxes();
|
---|
| 65 | UpdateSolutionVisualization();
|
---|
| 66 | }
|
---|
| 67 |
|
---|
| 68 | protected override void OnSolutionSeedingPoolChanged() {
|
---|
| 69 | base.OnSolutionSeedingPoolChanged();
|
---|
[13743] | 70 | UpdateSolutionNetworkAnalysis(similarityComboBox.SelectedItem as ISolutionSimilarityCalculator, (string)solutionNetworkProjectionComboBox.SelectedItem);
|
---|
[13720] | 71 | }
|
---|
| 72 |
|
---|
| 73 | private void UpdateSimilarityCalculators() {
|
---|
| 74 | var selected = (ISolutionSimilarityCalculator)(similarityComboBox.SelectedIndex >= 0 ? similarityComboBox.SelectedItem : null);
|
---|
| 75 | similarityComboBox.Items.Clear();
|
---|
| 76 |
|
---|
| 77 | if (Content == null || Content.Problem == null) return;
|
---|
| 78 |
|
---|
| 79 | foreach (var calc in Content.Problem.Operators.OfType<ISolutionSimilarityCalculator>()) {
|
---|
| 80 | similarityComboBox.Items.Add(calc);
|
---|
| 81 | if (selected != null && calc.ItemName == selected.ItemName) similarityComboBox.SelectedItem = calc;
|
---|
| 82 | }
|
---|
| 83 | if (selected == null && similarityComboBox.Items.Count > 0)
|
---|
| 84 | similarityComboBox.SelectedIndex = 0;
|
---|
| 85 | }
|
---|
| 86 |
|
---|
| 87 | private void UpdateNamesComboboxes() {
|
---|
| 88 | var selectedSolutionName = solutionNameComboBox.SelectedIndex >= 0 ? (string)solutionNameComboBox.SelectedItem : string.Empty;
|
---|
| 89 |
|
---|
| 90 | solutionNameComboBox.Items.Clear();
|
---|
| 91 | if (Content == null) return;
|
---|
| 92 | var solutionNames = Content.Problem.Solutions.Select(x => x.Solution).OfType<IScope>().SelectMany(x => x.Variables);
|
---|
| 93 |
|
---|
| 94 | foreach (var sn in solutionNames.GroupBy(x => x.Name).OrderBy(x => x.Key)) {
|
---|
| 95 | solutionNameComboBox.Items.Add(sn.Key);
|
---|
| 96 | // either it was previously selected, or the variable value is defined in the HeuristicLab.Encodings sub-namespace
|
---|
| 97 | if (sn.Key == selectedSolutionName || (string.IsNullOrEmpty(selectedSolutionName) && sn.All(x => x.Value != null && x.Value.GetType().FullName.StartsWith("HeuristicLab.Encodings."))))
|
---|
| 98 | solutionNameComboBox.SelectedItem = sn.Key;
|
---|
| 99 | }
|
---|
| 100 | }
|
---|
| 101 |
|
---|
| 102 | public void UpdateSolutionVisualization() {
|
---|
| 103 | if (InvokeRequired) { Invoke((Action)UpdateSolutionVisualization); return; }
|
---|
| 104 | var qualityName = Content.Problem.Problem.Evaluator.QualityParameter.ActualName;
|
---|
| 105 | UpdateSolutionQualityAnalysis(qualityName);
|
---|
| 106 | if (similarityComboBox.SelectedIndex >= 0) {
|
---|
| 107 | var calculator = (ISolutionSimilarityCalculator)similarityComboBox.SelectedItem;
|
---|
| 108 | UpdateSolutionDiversityAnalysis(calculator);
|
---|
| 109 | UpdateSolutionFdcAnalysis(calculator, fdcBetweenBestCheckBox.Checked);
|
---|
| 110 | UpdateSolutionLengthScaleAnalysis(calculator);
|
---|
[13743] | 111 | UpdateSolutionNetworkAnalysis(calculator, (string)solutionNetworkProjectionComboBox.SelectedItem);
|
---|
[13720] | 112 | } else {
|
---|
| 113 | solutionsDiversityViewHost.Content = null;
|
---|
| 114 | solutionsFdcViewHost.Content = null;
|
---|
| 115 | solutionsLengthScaleViewHost.Content = null;
|
---|
| 116 | solutionsNetworkChart.Series.First().Points.Clear();
|
---|
| 117 | }
|
---|
| 118 | }
|
---|
| 119 |
|
---|
| 120 | private void UpdateSolutionQualityAnalysis(string qualityName) {
|
---|
| 121 | var dt = solutionsQualityViewHost.Content as DataTable;
|
---|
| 122 | if (dt == null) {
|
---|
| 123 | dt = QualityDistributionAnalyzer.PrepareTable(qualityName);
|
---|
| 124 | dt.VisualProperties.Title = "Quality Distribution";
|
---|
| 125 | solutionsQualityViewHost.Content = dt;
|
---|
| 126 | }
|
---|
| 127 | QualityDistributionAnalyzer.UpdateTable(dt, GetSolutionScopes().Select(x => GetQuality(x, qualityName) ?? double.NaN).Where(x => !double.IsNaN(x)));
|
---|
| 128 | }
|
---|
| 129 |
|
---|
| 130 | private void UpdateSolutionDiversityAnalysis(ISolutionSimilarityCalculator calculator) {
|
---|
| 131 | try {
|
---|
| 132 | solutionsDiversityViewHost.Content = null;
|
---|
| 133 | var solutionScopes = GetSolutionScopes();
|
---|
| 134 | var similarities = new double[solutionScopes.Count, solutionScopes.Count];
|
---|
| 135 | for (var i = 0; i < solutionScopes.Count; i++) {
|
---|
| 136 | for (var j = 0; j < solutionScopes.Count; j++)
|
---|
| 137 | similarities[i, j] = calculator.CalculateSolutionSimilarity(solutionScopes[i], solutionScopes[j]);
|
---|
| 138 | }
|
---|
| 139 | var hm = new HeatMap(similarities, "Solution Similarities", 0.0, 1.0);
|
---|
| 140 | solutionsDiversityViewHost.Content = hm;
|
---|
| 141 | } catch { }
|
---|
| 142 | }
|
---|
| 143 |
|
---|
| 144 | private void UpdateSolutionFdcAnalysis(ISolutionSimilarityCalculator calculator, bool distanceToBest) {
|
---|
| 145 | try {
|
---|
| 146 | solutionsFdcViewHost.Content = null;
|
---|
[13722] | 147 | fdcSpearmanLabel.Text = "-";
|
---|
| 148 | fdcPearsonLabel.Text = "-";
|
---|
[13720] | 149 | var solutionScopes = GetSolutionScopes();
|
---|
| 150 | var points = new List<Point2D<double>>();
|
---|
| 151 | if (distanceToBest) {
|
---|
| 152 | var maximization = ((IValueParameter<BoolValue>)Content.Problem.MaximizationParameter).Value.Value;
|
---|
| 153 | var bestSolutions = (maximization ? solutionScopes.MaxItems(x => GetQuality(x, calculator.QualityVariableName) ?? double.NegativeInfinity)
|
---|
| 154 | : solutionScopes.MinItems(x => GetQuality(x, calculator.QualityVariableName) ?? double.PositiveInfinity)).ToList();
|
---|
| 155 | foreach (var solScope in solutionScopes.Except(bestSolutions)) {
|
---|
| 156 | var maxSimilarity = bestSolutions.Max(x => calculator.CalculateSolutionSimilarity(solScope, x));
|
---|
| 157 | var qDiff = (GetQuality(solScope, calculator.QualityVariableName) ?? double.NaN)
|
---|
| 158 | - (GetQuality(bestSolutions[0], calculator.QualityVariableName) ?? double.NaN);
|
---|
| 159 | points.Add(new Point2D<double>(Math.Abs(qDiff), 1.0 - maxSimilarity));
|
---|
| 160 | }
|
---|
| 161 | } else {
|
---|
| 162 | for (int i = 0; i < solutionScopes.Count; i++) {
|
---|
| 163 | for (int j = 0; j < solutionScopes.Count; j++) {
|
---|
| 164 | if (i == j) continue;
|
---|
| 165 | var qDiff = (GetQuality(solutionScopes[i], calculator.QualityVariableName) ?? double.NaN)
|
---|
| 166 | - (GetQuality(solutionScopes[j], calculator.QualityVariableName) ?? double.NaN);
|
---|
| 167 | if (double.IsNaN(qDiff)) continue;
|
---|
| 168 | points.Add(new Point2D<double>(Math.Abs(qDiff), 1.0 - calculator.CalculateSolutionSimilarity(solutionScopes[i], solutionScopes[j])));
|
---|
| 169 | }
|
---|
| 170 | }
|
---|
| 171 | }
|
---|
[13722] | 172 | var xs = points.Select(p => p.X).ToArray();
|
---|
| 173 | var ys = points.Select(p => p.Y).ToArray();
|
---|
| 174 | var scorr = alglib.spearmancorr2(xs, ys);
|
---|
| 175 | var pcorr = alglib.pearsoncorr2(xs, ys);
|
---|
| 176 | pcorr = pcorr * pcorr;
|
---|
| 177 | fdcSpearmanLabel.Text = scorr.ToString("F2");
|
---|
| 178 | fdcPearsonLabel.Text = pcorr.ToString("F2");
|
---|
| 179 |
|
---|
[13720] | 180 | var splot = new ScatterPlot("Fitness-Distance", "");
|
---|
| 181 | splot.VisualProperties.XAxisTitle = "Absolute Fitness Difference";
|
---|
[13722] | 182 | splot.VisualProperties.XAxisMinimumAuto = false;
|
---|
[13720] | 183 | splot.VisualProperties.XAxisMinimumFixedValue = 0.0;
|
---|
| 184 | splot.VisualProperties.YAxisTitle = "Solution Distance";
|
---|
[13722] | 185 | splot.VisualProperties.YAxisMinimumAuto = false;
|
---|
[13720] | 186 | splot.VisualProperties.YAxisMinimumFixedValue = 0.0;
|
---|
[13722] | 187 | splot.VisualProperties.YAxisMaximumAuto = false;
|
---|
[13720] | 188 | splot.VisualProperties.YAxisMaximumFixedValue = 1.0;
|
---|
| 189 | var row = new ScatterPlotDataRow("Fdc", "", points);
|
---|
[13722] | 190 | row.VisualProperties.PointSize = 10;
|
---|
| 191 | row.VisualProperties.ShowRegressionLine = true;
|
---|
[13720] | 192 | splot.Rows.Add(row);
|
---|
| 193 | solutionsFdcViewHost.Content = splot;
|
---|
| 194 | } catch { }
|
---|
| 195 | }
|
---|
| 196 |
|
---|
| 197 | private void UpdateSolutionLengthScaleAnalysis(ISolutionSimilarityCalculator calculator) {
|
---|
| 198 | try {
|
---|
| 199 | solutionsLengthScaleViewHost.Content = null;
|
---|
| 200 | var dt = solutionsLengthScaleViewHost.Content as DataTable;
|
---|
| 201 | if (dt == null) {
|
---|
| 202 | dt = QualityDistributionAnalyzer.PrepareTable("Length Scale");
|
---|
| 203 | solutionsLengthScaleViewHost.Content = dt;
|
---|
| 204 | }
|
---|
| 205 | QualityDistributionAnalyzer.UpdateTable(dt, CalculateLengthScale(calculator));
|
---|
| 206 | } catch {
|
---|
| 207 | solutionsLengthScaleViewHost.Content = null;
|
---|
| 208 | }
|
---|
| 209 | }
|
---|
| 210 |
|
---|
| 211 | private IEnumerable<double> CalculateLengthScale(ISolutionSimilarityCalculator calculator) {
|
---|
| 212 | var solutionScopes = GetSolutionScopes();
|
---|
| 213 | for (var i = 0; i < solutionScopes.Count; i++) {
|
---|
| 214 | for (var j = 0; j < solutionScopes.Count; j++) {
|
---|
| 215 | if (i == j) continue;
|
---|
| 216 | var sim = calculator.CalculateSolutionSimilarity(solutionScopes[i], solutionScopes[j]);
|
---|
| 217 | if (sim.IsAlmost(0)) continue;
|
---|
| 218 | var qDiff = (GetQuality(solutionScopes[i], calculator.QualityVariableName) ?? double.NaN)
|
---|
| 219 | - (GetQuality(solutionScopes[j], calculator.QualityVariableName) ?? double.NaN);
|
---|
| 220 | if (!double.IsNaN(qDiff)) yield return Math.Abs(qDiff) / sim;
|
---|
| 221 | }
|
---|
| 222 | }
|
---|
| 223 | }
|
---|
| 224 |
|
---|
[13743] | 225 | private void UpdateSolutionNetworkAnalysis(ISolutionSimilarityCalculator calculator, string projection) {
|
---|
[13720] | 226 | var series = solutionsNetworkChart.Series["SolutionSeries"];
|
---|
| 227 | var seedingSeries = solutionsNetworkChart.Series["SeedingSolutionSeries"];
|
---|
| 228 | try {
|
---|
| 229 | series.Points.Clear();
|
---|
| 230 | seedingSeries.Points.Clear();
|
---|
| 231 | var solutionScopes = GetSolutionScopes();
|
---|
| 232 | var dissimilarities = new DoubleMatrix(solutionScopes.Count, solutionScopes.Count);
|
---|
| 233 | for (var i = 0; i < solutionScopes.Count; i++) {
|
---|
| 234 | for (var j = 0; j < solutionScopes.Count; j++) {
|
---|
| 235 | if (i == j) continue;
|
---|
| 236 | dissimilarities[i, j] = 1.0 - calculator.CalculateSolutionSimilarity(solutionScopes[i], solutionScopes[j]);
|
---|
| 237 | }
|
---|
| 238 | }
|
---|
[13743] | 239 | DoubleMatrix coords = null;
|
---|
| 240 | if (projection == "SOM")
|
---|
| 241 | coords = Som(dissimilarities, new MersenneTwister(42), jittering: true);
|
---|
| 242 | else coords = MultidimensionalScaling.KruskalShepard(dissimilarities);
|
---|
[13720] | 243 | for (var i = 0; i < coords.Rows; i++) {
|
---|
| 244 | var quality = GetQuality(solutionScopes[i], calculator.QualityVariableName) ?? double.NaN;
|
---|
| 245 | var dataPoint = new DataPoint() {
|
---|
| 246 | Name = (i + 1).ToString(),
|
---|
| 247 | XValue = coords[i, 0],
|
---|
| 248 | YValues = new[] {coords[i, 1], quality},
|
---|
| 249 | Label = (i + 1) + ": " + quality,
|
---|
| 250 | Tag = solutionScopes[i]
|
---|
| 251 | };
|
---|
| 252 | if (Content.SolutionSeedingPool.Contains(solutionScopes[i]) && Content.SolutionSeedingPool.ItemChecked(solutionScopes[i]))
|
---|
| 253 | seedingSeries.Points.Add(dataPoint);
|
---|
| 254 | else series.Points.Add(dataPoint);
|
---|
| 255 | }
|
---|
| 256 | } catch {
|
---|
| 257 | // problems in calculating the similarity
|
---|
| 258 | series.Points.Clear();
|
---|
| 259 | seedingSeries.Points.Clear();
|
---|
| 260 | }
|
---|
| 261 | }
|
---|
| 262 |
|
---|
| 263 | private void SimilarityComboBoxOnSelectedIndexChanged(object sender, EventArgs e) {
|
---|
| 264 | if (InvokeRequired) { Invoke((Action<object, EventArgs>)SimilarityComboBoxOnSelectedIndexChanged, sender, e); return; }
|
---|
| 265 | var calculator = (ISolutionSimilarityCalculator)similarityComboBox.SelectedItem;
|
---|
| 266 | if (calculator != null) {
|
---|
| 267 | calculator.SolutionVariableName = (string)solutionNameComboBox.SelectedItem;
|
---|
| 268 | calculator.QualityVariableName = Content.Problem.Problem.Evaluator.QualityParameter.ActualName;
|
---|
| 269 | }
|
---|
| 270 | UpdateSolutionDiversityAnalysis(calculator);
|
---|
| 271 | UpdateSolutionFdcAnalysis(calculator, fdcBetweenBestCheckBox.Checked);
|
---|
| 272 | UpdateSolutionLengthScaleAnalysis(calculator);
|
---|
[13743] | 273 | UpdateSolutionNetworkAnalysis(calculator, (string)solutionNetworkProjectionComboBox.SelectedItem);
|
---|
[13720] | 274 | }
|
---|
| 275 |
|
---|
| 276 | private void SolutionNameComboBoxOnSelectedIndexChanged(object sender, EventArgs e) {
|
---|
| 277 | if (InvokeRequired) { Invoke((Action<object, EventArgs>)SolutionNameComboBoxOnSelectedIndexChanged, sender, e); return; }
|
---|
| 278 | var calculator = (ISolutionSimilarityCalculator)similarityComboBox.SelectedItem;
|
---|
| 279 | if (calculator != null) {
|
---|
| 280 | calculator.SolutionVariableName = (string)solutionNameComboBox.SelectedItem;
|
---|
| 281 | calculator.QualityVariableName = Content.Problem.Problem.Evaluator.QualityParameter.ActualName;
|
---|
| 282 | }
|
---|
| 283 | UpdateSolutionDiversityAnalysis(calculator);
|
---|
| 284 | UpdateSolutionFdcAnalysis(calculator, fdcBetweenBestCheckBox.Checked);
|
---|
| 285 | UpdateSolutionLengthScaleAnalysis(calculator);
|
---|
[13743] | 286 | UpdateSolutionNetworkAnalysis(calculator, (string)solutionNetworkProjectionComboBox.SelectedItem);
|
---|
[13720] | 287 | }
|
---|
| 288 |
|
---|
| 289 | private void FdcBetweenBestCheckBoxOnCheckedChanged(object sender, EventArgs e) {
|
---|
| 290 | if (InvokeRequired) { Invoke((Action<object, EventArgs>)FdcBetweenBestCheckBoxOnCheckedChanged, sender, e); return; }
|
---|
| 291 | UpdateSolutionFdcAnalysis((ISolutionSimilarityCalculator)similarityComboBox.SelectedItem, fdcBetweenBestCheckBox.Checked);
|
---|
| 292 | }
|
---|
| 293 |
|
---|
[13743] | 294 | private void SolutionNetworkProjectionComboBoxOnSelectedIndexChanged(object sender, EventArgs e) {
|
---|
| 295 | if (InvokeRequired) { Invoke((Action<object, EventArgs>)SolutionNetworkProjectionComboBoxOnSelectedIndexChanged, sender, e); return; }
|
---|
| 296 | var calculator = (ISolutionSimilarityCalculator)similarityComboBox.SelectedItem;
|
---|
| 297 | if (calculator != null) {
|
---|
| 298 | calculator.SolutionVariableName = (string)solutionNameComboBox.SelectedItem;
|
---|
| 299 | calculator.QualityVariableName = Content.Problem.Problem.Evaluator.QualityParameter.ActualName;
|
---|
| 300 | }
|
---|
| 301 | UpdateSolutionNetworkAnalysis(calculator, (string)solutionNetworkProjectionComboBox.SelectedItem);
|
---|
| 302 | }
|
---|
| 303 |
|
---|
[13720] | 304 | private void SolutionsNetworkChartOnMouseClick(object sender, MouseEventArgs e) {
|
---|
| 305 | var result = solutionsNetworkChart.HitTest(e.X, e.Y);
|
---|
| 306 | if (result.ChartElementType == ChartElementType.DataPoint) {
|
---|
| 307 | var point = (DataPoint)result.Object;
|
---|
| 308 | var solutionScope = (IScope)point.Tag;
|
---|
| 309 | if (!Content.SolutionSeedingPool.Contains(solutionScope)) return;
|
---|
| 310 | Content.SolutionSeedingPool.SetItemCheckedState(solutionScope, !Content.SolutionSeedingPool.ItemChecked(solutionScope));
|
---|
| 311 | }
|
---|
| 312 | }
|
---|
| 313 |
|
---|
| 314 | private void SolutionsNetworkChartOnMouseDoubleClick(object sender, MouseEventArgs e) {
|
---|
| 315 | var result = solutionsNetworkChart.HitTest(e.X, e.Y);
|
---|
| 316 | if (result.ChartElementType == ChartElementType.DataPoint) {
|
---|
| 317 | var point = (DataPoint)result.Object;
|
---|
| 318 | var solutionScope = (IScope)point.Tag;
|
---|
| 319 | MainForm.ShowContent(solutionScope, true);
|
---|
| 320 | }
|
---|
| 321 | }
|
---|
| 322 |
|
---|
| 323 | #region Helpers
|
---|
| 324 | private List<IScope> GetSolutionScopes() {
|
---|
| 325 | return Content.Problem.Solutions.Select(x => x.Solution).OfType<IScope>().ToList();
|
---|
| 326 | }
|
---|
| 327 |
|
---|
| 328 | private double? GetQuality(IScope scope, string qualityName) {
|
---|
| 329 | IVariable v;
|
---|
| 330 | if (!scope.Variables.TryGetValue(qualityName, out v)) return null;
|
---|
| 331 | var dval = v.Value as DoubleValue;
|
---|
| 332 | if (dval == null) return null;
|
---|
| 333 | return dval.Value;
|
---|
| 334 | }
|
---|
[13743] | 335 |
|
---|
| 336 | #region SOM projection
|
---|
| 337 | /// <summary>
|
---|
| 338 | /// This is the online algorithm described in
|
---|
| 339 | /// Olteanu, M. and Villa-Vialaneix, N. 2015.
|
---|
| 340 | /// On-line relational and multiple relational SOM.
|
---|
| 341 | /// Neurocomputing 147, pp. 15-30.
|
---|
| 342 | /// </summary>
|
---|
| 343 | /// <param name="dissimilarities">The full NxN matrix containing all dissimilarities between N points.</param>
|
---|
| 344 | /// <param name="random">The random number generator to use.</param>
|
---|
| 345 | /// <param name="somSize">The length of a side of the SOM grid (there are somSize * somSize neurons).</param>
|
---|
| 346 | /// <param name="iterations">The amount of iterations to perform in learning.</param>
|
---|
| 347 | /// <param name="learningRate">The initial learning rate.</param>
|
---|
| 348 | /// <param name="learningRadius">The initial learning radius.</param>
|
---|
| 349 | /// <param name="jittering">If the final coordinates should be jittered slightly within the grid.</param>
|
---|
| 350 | /// <returns>A matrix of coordinates having N rows and 2 columns.</returns>
|
---|
| 351 | private DoubleMatrix Som(DoubleMatrix dissimilarities, IRandom random, int somSize = 5, int iterations = 100, double learningRate = double.NaN, double learningRadius = 5.0, bool jittering = true) {
|
---|
| 352 | var K = somSize * somSize;
|
---|
| 353 | var N = dissimilarities.Rows;
|
---|
| 354 | if (double.IsNaN(learningRate)) learningRate = 1.0 / Math.Sqrt(2.0 * N);
|
---|
| 355 | var fixedLearningRate = learningRate / 10.0;
|
---|
| 356 | var varLearningRate = 9.0 * fixedLearningRate;
|
---|
| 357 | Func<int, int, double> learningRateT = (maxIter, iter) => {
|
---|
| 358 | return varLearningRate * ((maxIter - iter) / (double)maxIter) + fixedLearningRate;
|
---|
| 359 | };
|
---|
| 360 | Func<int, int> getX = (neuron) => neuron % somSize;
|
---|
| 361 | Func<int, int> getY = (neuron) => neuron / somSize;
|
---|
| 362 | Func<int, int, int, int, double> neighborhood = (maxIter, iter, k, bmu) => {
|
---|
| 363 | var sigma = 1.0 * ((maxIter - iter) / (double)maxIter) + 0.0001;
|
---|
| 364 | var xK = getX(k);
|
---|
| 365 | var yK = getY(k);
|
---|
| 366 | var xW = getX(bmu);
|
---|
| 367 | var yW = getY(bmu);
|
---|
| 368 | var d = (xK - xW) * (xK - xW) + (yK - yW) * (yK - yW);
|
---|
| 369 | return Math.Exp(-d / (2.0 * sigma * sigma));
|
---|
| 370 | };
|
---|
| 371 | var alphas = Enumerable.Range(0, K).Select(k => Enumerable.Range(0, N).Select(_ => random.NextDouble()).ToArray()).ToArray();
|
---|
| 372 | // normalize s.t. sum(alphas[k]) = 1
|
---|
| 373 | for (var k = 0; k < K; k++) {
|
---|
| 374 | var sum = alphas[k].Sum();
|
---|
| 375 | for (var i = 0; i < alphas[k].Length; i++) alphas[k][i] /= sum;
|
---|
| 376 | }
|
---|
| 377 | var oldAlphas = alphas.Select(x => (double[])x.Clone()).ToArray();
|
---|
| 378 |
|
---|
| 379 | for (var iter = 0; iter < iterations; iter++) {
|
---|
| 380 | var pointShuffle = Enumerable.Range(0, N).Shuffle(random).ToArray();
|
---|
| 381 | for (var p = 0; p < N; p++) {
|
---|
| 382 | var i = pointShuffle[p];
|
---|
| 383 | var bmu = GetBestMatchingUnit(dissimilarities, alphas, i);
|
---|
| 384 |
|
---|
| 385 | for (var k = 0; k < K; k++) {
|
---|
| 386 | for (var j = 0; j < N; j++) {
|
---|
| 387 | alphas[k][j] = oldAlphas[k][j] + learningRateT(iterations, iter) * neighborhood(iterations, iter, k, bmu) * ((i == j ? 1.0 : 0.0) - oldAlphas[k][j]);
|
---|
| 388 | }
|
---|
| 389 | }
|
---|
| 390 | }
|
---|
| 391 | for (var k = 0; k < K; k++) {
|
---|
| 392 | for (var j = 0; j < N; j++) {
|
---|
| 393 | oldAlphas[k][j] = alphas[k][j];
|
---|
| 394 | }
|
---|
| 395 | }
|
---|
| 396 | }
|
---|
| 397 |
|
---|
| 398 | var result = new DoubleMatrix(N, 2);
|
---|
| 399 | for (var i = 0; i < N; i++) {
|
---|
| 400 | var bmu = GetBestMatchingUnit(dissimilarities, alphas, i);
|
---|
| 401 | if (!jittering) {
|
---|
| 402 | result[i, 0] = getX(bmu);
|
---|
| 403 | result[i, 1] = getY(bmu);
|
---|
| 404 | } else {
|
---|
| 405 | result[i, 0] = getX(bmu) + random.NextDouble() * 0.8;
|
---|
| 406 | result[i, 1] = getY(bmu) + random.NextDouble() * 0.8;
|
---|
| 407 | }
|
---|
| 408 | }
|
---|
| 409 | return result;
|
---|
| 410 | }
|
---|
| 411 |
|
---|
| 412 | private int GetBestMatchingUnit(DoubleMatrix D, double[][] alphas, int i) {
|
---|
| 413 | var bmu = -1;
|
---|
| 414 | var minV = double.MaxValue;
|
---|
| 415 | for (var k = 0; k < alphas.Length; k++) {
|
---|
| 416 | var Daki = 0.0;
|
---|
| 417 | var akDak = 0.0;
|
---|
| 418 | for (var r = 0; r < D.Rows; r++) {
|
---|
| 419 | var Dakr = 0.0;
|
---|
| 420 | for (var s = 0; s < D.Rows; s++) {
|
---|
| 421 | Dakr += D[r, s] * alphas[k][s];
|
---|
| 422 | }
|
---|
| 423 | if (r == i) Daki = Dakr;
|
---|
| 424 | akDak += alphas[k][r] * Dakr;
|
---|
| 425 | }
|
---|
| 426 | var v = Daki - 0.5 * akDak;
|
---|
| 427 | if (v < minV) {
|
---|
| 428 | bmu = k;
|
---|
| 429 | minV = v;
|
---|
| 430 | }
|
---|
| 431 | }
|
---|
| 432 | return bmu;
|
---|
| 433 | }
|
---|
[13720] | 434 | #endregion
|
---|
[13743] | 435 |
|
---|
| 436 | private DoubleMatrix Mds(DoubleMatrix dissimilarities) {
|
---|
| 437 | return MultidimensionalScaling.KruskalShepard(dissimilarities);
|
---|
| 438 | }
|
---|
| 439 | #endregion
|
---|
[13720] | 440 | }
|
---|
| 441 | }
|
---|