| 1 | #region License Information
|
|---|
| 2 | /* HeuristicLab
|
|---|
| 3 | * Copyright (C) 2002-2019 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;
|
|---|
| 23 | using System.Collections.Generic;
|
|---|
| 24 | using System.Drawing;
|
|---|
| 25 | using System.Linq;
|
|---|
| 26 | using System.Threading;
|
|---|
| 27 | using HeuristicLab.Analysis;
|
|---|
| 28 | using HeuristicLab.Common;
|
|---|
| 29 | using HeuristicLab.Core;
|
|---|
| 30 | using HeuristicLab.Data;
|
|---|
| 31 | using HeuristicLab.Optimization;
|
|---|
| 32 | using HeuristicLab.Parameters;
|
|---|
| 33 | using HEAL.Attic;
|
|---|
| 34 | using HeuristicLab.PluginInfrastructure;
|
|---|
| 35 | using HeuristicLab.Problems.DataAnalysis;
|
|---|
| 36 | using HeuristicLab.Random;
|
|---|
| 37 |
|
|---|
| 38 | namespace HeuristicLab.Algorithms.DataAnalysis {
|
|---|
| 39 | /// <summary>
|
|---|
| 40 | /// t-Distributed Stochastic Neighbor Embedding (tSNE) projects the data in a low dimensional
|
|---|
| 41 | /// space to allow visual cluster identification.
|
|---|
| 42 | /// </summary>
|
|---|
| 43 | [Item("t-Distributed Stochastic Neighbor Embedding (tSNE)", "t-Distributed Stochastic Neighbor Embedding projects the data in a low " +
|
|---|
| 44 | "dimensional space to allow visual cluster identification. Implemented similar to: https://lvdmaaten.github.io/tsne/#implementations (Barnes-Hut t-SNE). Described in : https://lvdmaaten.github.io/publications/papers/JMLR_2014.pdf")]
|
|---|
| 45 | [Creatable(CreatableAttribute.Categories.DataAnalysis, Priority = 100)]
|
|---|
| 46 | [StorableType("1CE58B5E-C319-4DEB-B66B-994171370B06")]
|
|---|
| 47 | public sealed class TSNEAlgorithm : BasicAlgorithm {
|
|---|
| 48 | public override bool SupportsPause {
|
|---|
| 49 | get { return true; }
|
|---|
| 50 | }
|
|---|
| 51 | public override Type ProblemType {
|
|---|
| 52 | get { return typeof(IDataAnalysisProblem); }
|
|---|
| 53 | }
|
|---|
| 54 | public new IDataAnalysisProblem Problem {
|
|---|
| 55 | get { return (IDataAnalysisProblem)base.Problem; }
|
|---|
| 56 | set { base.Problem = value; }
|
|---|
| 57 | }
|
|---|
| 58 |
|
|---|
| 59 | #region Parameter names
|
|---|
| 60 | private const string DistanceFunctionParameterName = "DistanceFunction";
|
|---|
| 61 | private const string PerplexityParameterName = "Perplexity";
|
|---|
| 62 | private const string ThetaParameterName = "Theta";
|
|---|
| 63 | private const string NewDimensionsParameterName = "Dimensions";
|
|---|
| 64 | private const string MaxIterationsParameterName = "MaxIterations";
|
|---|
| 65 | private const string StopLyingIterationParameterName = "StopLyingIteration";
|
|---|
| 66 | private const string MomentumSwitchIterationParameterName = "MomentumSwitchIteration";
|
|---|
| 67 | private const string InitialMomentumParameterName = "InitialMomentum";
|
|---|
| 68 | private const string FinalMomentumParameterName = "FinalMomentum";
|
|---|
| 69 | private const string EtaParameterName = "Eta";
|
|---|
| 70 | private const string SetSeedRandomlyParameterName = "SetSeedRandomly";
|
|---|
| 71 | private const string SeedParameterName = "Seed";
|
|---|
| 72 | private const string ClassesNameParameterName = "ClassesName";
|
|---|
| 73 | private const string NormalizationParameterName = "Normalization";
|
|---|
| 74 | private const string RandomInitializationParameterName = "RandomInitialization";
|
|---|
| 75 | private const string UpdateIntervalParameterName = "UpdateInterval";
|
|---|
| 76 | #endregion
|
|---|
| 77 |
|
|---|
| 78 | #region Result names
|
|---|
| 79 | private const string IterationResultName = "Iteration";
|
|---|
| 80 | private const string ErrorResultName = "Error";
|
|---|
| 81 | private const string ErrorPlotResultName = "Error plot";
|
|---|
| 82 | private const string ScatterPlotResultName = "Scatterplot";
|
|---|
| 83 | private const string DataResultName = "Projected data";
|
|---|
| 84 | #endregion
|
|---|
| 85 |
|
|---|
| 86 | #region Parameter properties
|
|---|
| 87 | public IFixedValueParameter<DoubleValue> PerplexityParameter {
|
|---|
| 88 | get { return (IFixedValueParameter<DoubleValue>)Parameters[PerplexityParameterName]; }
|
|---|
| 89 | }
|
|---|
| 90 | public IFixedValueParameter<PercentValue> ThetaParameter {
|
|---|
| 91 | get { return (IFixedValueParameter<PercentValue>)Parameters[ThetaParameterName]; }
|
|---|
| 92 | }
|
|---|
| 93 | public IFixedValueParameter<IntValue> NewDimensionsParameter {
|
|---|
| 94 | get { return (IFixedValueParameter<IntValue>)Parameters[NewDimensionsParameterName]; }
|
|---|
| 95 | }
|
|---|
| 96 | public IConstrainedValueParameter<IDistance<double[]>> DistanceFunctionParameter {
|
|---|
| 97 | get { return (IConstrainedValueParameter<IDistance<double[]>>)Parameters[DistanceFunctionParameterName]; }
|
|---|
| 98 | }
|
|---|
| 99 | public IFixedValueParameter<IntValue> MaxIterationsParameter {
|
|---|
| 100 | get { return (IFixedValueParameter<IntValue>)Parameters[MaxIterationsParameterName]; }
|
|---|
| 101 | }
|
|---|
| 102 | public IFixedValueParameter<IntValue> StopLyingIterationParameter {
|
|---|
| 103 | get { return (IFixedValueParameter<IntValue>)Parameters[StopLyingIterationParameterName]; }
|
|---|
| 104 | }
|
|---|
| 105 | public IFixedValueParameter<IntValue> MomentumSwitchIterationParameter {
|
|---|
| 106 | get { return (IFixedValueParameter<IntValue>)Parameters[MomentumSwitchIterationParameterName]; }
|
|---|
| 107 | }
|
|---|
| 108 | public IFixedValueParameter<DoubleValue> InitialMomentumParameter {
|
|---|
| 109 | get { return (IFixedValueParameter<DoubleValue>)Parameters[InitialMomentumParameterName]; }
|
|---|
| 110 | }
|
|---|
| 111 | public IFixedValueParameter<DoubleValue> FinalMomentumParameter {
|
|---|
| 112 | get { return (IFixedValueParameter<DoubleValue>)Parameters[FinalMomentumParameterName]; }
|
|---|
| 113 | }
|
|---|
| 114 | public IFixedValueParameter<DoubleValue> EtaParameter {
|
|---|
| 115 | get { return (IFixedValueParameter<DoubleValue>)Parameters[EtaParameterName]; }
|
|---|
| 116 | }
|
|---|
| 117 | public IFixedValueParameter<BoolValue> SetSeedRandomlyParameter {
|
|---|
| 118 | get { return (IFixedValueParameter<BoolValue>)Parameters[SetSeedRandomlyParameterName]; }
|
|---|
| 119 | }
|
|---|
| 120 | public IFixedValueParameter<IntValue> SeedParameter {
|
|---|
| 121 | get { return (IFixedValueParameter<IntValue>)Parameters[SeedParameterName]; }
|
|---|
| 122 | }
|
|---|
| 123 | public IConstrainedValueParameter<StringValue> ClassesNameParameter {
|
|---|
| 124 | get { return (IConstrainedValueParameter<StringValue>)Parameters[ClassesNameParameterName]; }
|
|---|
| 125 | }
|
|---|
| 126 | public IFixedValueParameter<BoolValue> NormalizationParameter {
|
|---|
| 127 | get { return (IFixedValueParameter<BoolValue>)Parameters[NormalizationParameterName]; }
|
|---|
| 128 | }
|
|---|
| 129 | public IFixedValueParameter<BoolValue> RandomInitializationParameter {
|
|---|
| 130 | get { return (IFixedValueParameter<BoolValue>)Parameters[RandomInitializationParameterName]; }
|
|---|
| 131 | }
|
|---|
| 132 | public IFixedValueParameter<IntValue> UpdateIntervalParameter {
|
|---|
| 133 | get { return (IFixedValueParameter<IntValue>)Parameters[UpdateIntervalParameterName]; }
|
|---|
| 134 | }
|
|---|
| 135 | #endregion
|
|---|
| 136 |
|
|---|
| 137 | #region Properties
|
|---|
| 138 | public IDistance<double[]> DistanceFunction {
|
|---|
| 139 | get { return DistanceFunctionParameter.Value; }
|
|---|
| 140 | }
|
|---|
| 141 | public double Perplexity {
|
|---|
| 142 | get { return PerplexityParameter.Value.Value; }
|
|---|
| 143 | set { PerplexityParameter.Value.Value = value; }
|
|---|
| 144 | }
|
|---|
| 145 | public double Theta {
|
|---|
| 146 | get { return ThetaParameter.Value.Value; }
|
|---|
| 147 | set { ThetaParameter.Value.Value = value; }
|
|---|
| 148 | }
|
|---|
| 149 | public int NewDimensions {
|
|---|
| 150 | get { return NewDimensionsParameter.Value.Value; }
|
|---|
| 151 | set { NewDimensionsParameter.Value.Value = value; }
|
|---|
| 152 | }
|
|---|
| 153 | public int MaxIterations {
|
|---|
| 154 | get { return MaxIterationsParameter.Value.Value; }
|
|---|
| 155 | set { MaxIterationsParameter.Value.Value = value; }
|
|---|
| 156 | }
|
|---|
| 157 | public int StopLyingIteration {
|
|---|
| 158 | get { return StopLyingIterationParameter.Value.Value; }
|
|---|
| 159 | set { StopLyingIterationParameter.Value.Value = value; }
|
|---|
| 160 | }
|
|---|
| 161 | public int MomentumSwitchIteration {
|
|---|
| 162 | get { return MomentumSwitchIterationParameter.Value.Value; }
|
|---|
| 163 | set { MomentumSwitchIterationParameter.Value.Value = value; }
|
|---|
| 164 | }
|
|---|
| 165 | public double InitialMomentum {
|
|---|
| 166 | get { return InitialMomentumParameter.Value.Value; }
|
|---|
| 167 | set { InitialMomentumParameter.Value.Value = value; }
|
|---|
| 168 | }
|
|---|
| 169 | public double FinalMomentum {
|
|---|
| 170 | get { return FinalMomentumParameter.Value.Value; }
|
|---|
| 171 | set { FinalMomentumParameter.Value.Value = value; }
|
|---|
| 172 | }
|
|---|
| 173 | public double Eta {
|
|---|
| 174 | get { return EtaParameter.Value.Value; }
|
|---|
| 175 | set { EtaParameter.Value.Value = value; }
|
|---|
| 176 | }
|
|---|
| 177 | public bool SetSeedRandomly {
|
|---|
| 178 | get { return SetSeedRandomlyParameter.Value.Value; }
|
|---|
| 179 | set { SetSeedRandomlyParameter.Value.Value = value; }
|
|---|
| 180 | }
|
|---|
| 181 | public int Seed {
|
|---|
| 182 | get { return SeedParameter.Value.Value; }
|
|---|
| 183 | set { SeedParameter.Value.Value = value; }
|
|---|
| 184 | }
|
|---|
| 185 | public string ClassesName {
|
|---|
| 186 | get { return ClassesNameParameter.Value != null ? ClassesNameParameter.Value.Value : null; }
|
|---|
| 187 | set { ClassesNameParameter.Value.Value = value; }
|
|---|
| 188 | }
|
|---|
| 189 | public bool Normalization {
|
|---|
| 190 | get { return NormalizationParameter.Value.Value; }
|
|---|
| 191 | set { NormalizationParameter.Value.Value = value; }
|
|---|
| 192 | }
|
|---|
| 193 | public bool RandomInitialization {
|
|---|
| 194 | get { return RandomInitializationParameter.Value.Value; }
|
|---|
| 195 | set { RandomInitializationParameter.Value.Value = value; }
|
|---|
| 196 | }
|
|---|
| 197 | public int UpdateInterval {
|
|---|
| 198 | get { return UpdateIntervalParameter.Value.Value; }
|
|---|
| 199 | set { UpdateIntervalParameter.Value.Value = value; }
|
|---|
| 200 | }
|
|---|
| 201 | #endregion
|
|---|
| 202 |
|
|---|
| 203 | #region Storable poperties
|
|---|
| 204 | [Storable]
|
|---|
| 205 | private Dictionary<string, IList<int>> dataRowIndices;
|
|---|
| 206 | [Storable]
|
|---|
| 207 | private TSNEStatic<double[]>.TSNEState state;
|
|---|
| 208 | #endregion
|
|---|
| 209 |
|
|---|
| 210 | #region Constructors & Cloning
|
|---|
| 211 | [StorableConstructor]
|
|---|
| 212 | private TSNEAlgorithm(StorableConstructorFlag _) : base(_) { }
|
|---|
| 213 |
|
|---|
| 214 | [StorableHook(HookType.AfterDeserialization)]
|
|---|
| 215 | private void AfterDeserialization() {
|
|---|
| 216 | if (!Parameters.ContainsKey(RandomInitializationParameterName))
|
|---|
| 217 | Parameters.Add(new FixedValueParameter<BoolValue>(RandomInitializationParameterName, "Wether data points should be randomly initialized or according to the first 2 dimensions", new BoolValue(true)));
|
|---|
| 218 | RegisterParameterEvents();
|
|---|
| 219 | }
|
|---|
| 220 | private TSNEAlgorithm(TSNEAlgorithm original, Cloner cloner) : base(original, cloner) {
|
|---|
| 221 | if (original.dataRowIndices != null)
|
|---|
| 222 | dataRowIndices = new Dictionary<string, IList<int>>(original.dataRowIndices);
|
|---|
| 223 | if (original.state != null)
|
|---|
| 224 | state = cloner.Clone(original.state);
|
|---|
| 225 | RegisterParameterEvents();
|
|---|
| 226 | }
|
|---|
| 227 | public override IDeepCloneable Clone(Cloner cloner) {
|
|---|
| 228 | return new TSNEAlgorithm(this, cloner);
|
|---|
| 229 | }
|
|---|
| 230 | public TSNEAlgorithm() {
|
|---|
| 231 | var distances = new ItemSet<IDistance<double[]>>(ApplicationManager.Manager.GetInstances<IDistance<double[]>>());
|
|---|
| 232 | Parameters.Add(new ConstrainedValueParameter<IDistance<double[]>>(DistanceFunctionParameterName, "The distance function used to differentiate similar from non-similar points", distances, distances.OfType<EuclideanDistance>().FirstOrDefault()));
|
|---|
| 233 | Parameters.Add(new FixedValueParameter<DoubleValue>(PerplexityParameterName, "Perplexity-parameter of tSNE. Comparable to k in a k-nearest neighbour algorithm. Recommended value is floor(number of points /3) or lower", new DoubleValue(25)));
|
|---|
| 234 | Parameters.Add(new FixedValueParameter<PercentValue>(ThetaParameterName, "Value describing how much appoximated " +
|
|---|
| 235 | "gradients my differ from exact gradients. Set to 0 for exact calculation and in [0,1] otherwise. " +
|
|---|
| 236 | "Appropriate values for theta are between 0.1 and 0.7 (default = 0.5). CAUTION: exact calculation of " +
|
|---|
| 237 | "forces requires building a non-sparse N*N matrix where N is the number of data points. This may " +
|
|---|
| 238 | "exceed memory limitations. The function is designed to run on large (N > 5000) data sets. It may give" +
|
|---|
| 239 | " poor performance on very small data sets(it is better to use a standard t - SNE implementation on such data).", new PercentValue(0)));
|
|---|
| 240 | Parameters.Add(new FixedValueParameter<IntValue>(NewDimensionsParameterName, "Dimensionality of projected space (usually 2 for easy visual analysis)", new IntValue(2)));
|
|---|
| 241 | Parameters.Add(new FixedValueParameter<IntValue>(MaxIterationsParameterName, "Maximum number of iterations for gradient descent.", new IntValue(1000)));
|
|---|
| 242 | Parameters.Add(new FixedValueParameter<IntValue>(StopLyingIterationParameterName, "Number of iterations after which p is no longer approximated.", new IntValue(0)));
|
|---|
| 243 | Parameters.Add(new FixedValueParameter<IntValue>(MomentumSwitchIterationParameterName, "Number of iterations after which the momentum in the gradient descent is switched.", new IntValue(0)));
|
|---|
| 244 | Parameters.Add(new FixedValueParameter<DoubleValue>(InitialMomentumParameterName, "The initial momentum in the gradient descent.", new DoubleValue(0.5)));
|
|---|
| 245 | Parameters.Add(new FixedValueParameter<DoubleValue>(FinalMomentumParameterName, "The final momentum.", new DoubleValue(0.8)));
|
|---|
| 246 | Parameters.Add(new FixedValueParameter<DoubleValue>(EtaParameterName, "Gradient descent learning rate.", new DoubleValue(10)));
|
|---|
| 247 | Parameters.Add(new FixedValueParameter<BoolValue>(SetSeedRandomlyParameterName, "If the seed should be random.", new BoolValue(true)));
|
|---|
| 248 | Parameters.Add(new FixedValueParameter<IntValue>(SeedParameterName, "The seed used if it should not be random.", new IntValue(0)));
|
|---|
| 249 | Parameters.Add(new OptionalConstrainedValueParameter<StringValue>(ClassesNameParameterName, "Name of the column specifying the class lables of each data point. If this is not set training/test is used as labels."));
|
|---|
| 250 | Parameters.Add(new FixedValueParameter<BoolValue>(NormalizationParameterName, "Whether the data should be zero centered and have variance of 1 for each variable, so different scalings are ignored.", new BoolValue(true)));
|
|---|
| 251 | Parameters.Add(new FixedValueParameter<IntValue>(UpdateIntervalParameterName, "The interval after which the results will be updated.", new IntValue(50)));
|
|---|
| 252 | Parameters.Add(new FixedValueParameter<BoolValue>(RandomInitializationParameterName, "Wether data points should be randomly initialized or according to the first 2 dimensions", new BoolValue(true)));
|
|---|
| 253 |
|
|---|
| 254 | UpdateIntervalParameter.Hidden = true;
|
|---|
| 255 | MomentumSwitchIterationParameter.Hidden = true;
|
|---|
| 256 | InitialMomentumParameter.Hidden = true;
|
|---|
| 257 | FinalMomentumParameter.Hidden = true;
|
|---|
| 258 | StopLyingIterationParameter.Hidden = true;
|
|---|
| 259 | EtaParameter.Hidden = false;
|
|---|
| 260 | Problem = new RegressionProblem();
|
|---|
| 261 | RegisterParameterEvents();
|
|---|
| 262 | }
|
|---|
| 263 | #endregion
|
|---|
| 264 |
|
|---|
| 265 | public override void Prepare() {
|
|---|
| 266 | base.Prepare();
|
|---|
| 267 | dataRowIndices = null;
|
|---|
| 268 | state = null;
|
|---|
| 269 | }
|
|---|
| 270 |
|
|---|
| 271 | protected override void Run(CancellationToken cancellationToken) {
|
|---|
| 272 | var problemData = Problem.ProblemData;
|
|---|
| 273 | // set up and initialize everything if necessary
|
|---|
| 274 | var wdist = DistanceFunction as WeightedEuclideanDistance;
|
|---|
| 275 | if (wdist != null) wdist.Initialize(problemData);
|
|---|
| 276 | if (state == null) {
|
|---|
| 277 | if (SetSeedRandomly) Seed = RandomSeedGenerator.GetSeed();
|
|---|
| 278 | var random = new MersenneTwister((uint)Seed);
|
|---|
| 279 | var dataset = problemData.Dataset;
|
|---|
| 280 | var allowedInputVariables = problemData.AllowedInputVariables.ToArray();
|
|---|
| 281 | var allindices = Problem.ProblemData.AllIndices.ToArray();
|
|---|
| 282 |
|
|---|
| 283 | // jagged array is required to meet the static method declarations of TSNEStatic<T>
|
|---|
| 284 | var data = Enumerable.Range(0, dataset.Rows).Select(x => new double[allowedInputVariables.Length]).ToArray();
|
|---|
| 285 | var col = 0;
|
|---|
| 286 | foreach (var s in allowedInputVariables) {
|
|---|
| 287 | var row = 0;
|
|---|
| 288 | foreach (var d in dataset.GetDoubleValues(s)) {
|
|---|
| 289 | data[row][col] = d;
|
|---|
| 290 | row++;
|
|---|
| 291 | }
|
|---|
| 292 | col++;
|
|---|
| 293 | }
|
|---|
| 294 | if (Normalization) data = NormalizeInputData(data);
|
|---|
| 295 | state = TSNEStatic<double[]>.CreateState(data, DistanceFunction, random, NewDimensions, Perplexity, Theta, StopLyingIteration, MomentumSwitchIteration, InitialMomentum, FinalMomentum, Eta, RandomInitialization);
|
|---|
| 296 | SetUpResults(allindices);
|
|---|
| 297 | }
|
|---|
| 298 | while (state.iter < MaxIterations && !cancellationToken.IsCancellationRequested) {
|
|---|
| 299 | if (state.iter % UpdateInterval == 0) Analyze(state);
|
|---|
| 300 | TSNEStatic<double[]>.Iterate(state);
|
|---|
| 301 | }
|
|---|
| 302 | Analyze(state);
|
|---|
| 303 | }
|
|---|
| 304 |
|
|---|
| 305 | #region Events
|
|---|
| 306 | protected override void OnProblemChanged() {
|
|---|
| 307 | base.OnProblemChanged();
|
|---|
| 308 | if (Problem == null) return;
|
|---|
| 309 | OnProblemDataChanged(this, null);
|
|---|
| 310 | }
|
|---|
| 311 |
|
|---|
| 312 | protected override void RegisterProblemEvents() {
|
|---|
| 313 | base.RegisterProblemEvents();
|
|---|
| 314 | if (Problem == null) return;
|
|---|
| 315 | Problem.ProblemDataChanged += OnProblemDataChanged;
|
|---|
| 316 | if (Problem.ProblemData == null) return;
|
|---|
| 317 | Problem.ProblemData.Changed += OnPerplexityChanged;
|
|---|
| 318 | Problem.ProblemData.Changed += OnColumnsChanged;
|
|---|
| 319 | if (Problem.ProblemData.Dataset == null) return;
|
|---|
| 320 | Problem.ProblemData.Dataset.RowsChanged += OnPerplexityChanged;
|
|---|
| 321 | Problem.ProblemData.Dataset.ColumnsChanged += OnColumnsChanged;
|
|---|
| 322 | }
|
|---|
| 323 |
|
|---|
| 324 | protected override void DeregisterProblemEvents() {
|
|---|
| 325 | base.DeregisterProblemEvents();
|
|---|
| 326 | if (Problem == null) return;
|
|---|
| 327 | Problem.ProblemDataChanged -= OnProblemDataChanged;
|
|---|
| 328 | if (Problem.ProblemData == null) return;
|
|---|
| 329 | Problem.ProblemData.Changed -= OnPerplexityChanged;
|
|---|
| 330 | Problem.ProblemData.Changed -= OnColumnsChanged;
|
|---|
| 331 | if (Problem.ProblemData.Dataset == null) return;
|
|---|
| 332 | Problem.ProblemData.Dataset.RowsChanged -= OnPerplexityChanged;
|
|---|
| 333 | Problem.ProblemData.Dataset.ColumnsChanged -= OnColumnsChanged;
|
|---|
| 334 | }
|
|---|
| 335 |
|
|---|
| 336 | protected override void OnStopped() {
|
|---|
| 337 | base.OnStopped();
|
|---|
| 338 | //bwerth: state objects can be very large; avoid state serialization
|
|---|
| 339 | state = null;
|
|---|
| 340 | dataRowIndices = null;
|
|---|
| 341 | }
|
|---|
| 342 |
|
|---|
| 343 | private void OnProblemDataChanged(object sender, EventArgs args) {
|
|---|
| 344 | if (Problem == null || Problem.ProblemData == null) return;
|
|---|
| 345 | OnPerplexityChanged(this, null);
|
|---|
| 346 | OnColumnsChanged(this, null);
|
|---|
| 347 | Problem.ProblemData.Changed += OnPerplexityChanged;
|
|---|
| 348 | Problem.ProblemData.Changed += OnColumnsChanged;
|
|---|
| 349 | if (Problem.ProblemData.Dataset == null) return;
|
|---|
| 350 | Problem.ProblemData.Dataset.RowsChanged += OnPerplexityChanged;
|
|---|
| 351 | Problem.ProblemData.Dataset.ColumnsChanged += OnColumnsChanged;
|
|---|
| 352 | if (!Parameters.ContainsKey(ClassesNameParameterName)) return;
|
|---|
| 353 | ClassesNameParameter.ValidValues.Clear();
|
|---|
| 354 | foreach (var input in Problem.ProblemData.InputVariables) ClassesNameParameter.ValidValues.Add(input);
|
|---|
| 355 | }
|
|---|
| 356 |
|
|---|
| 357 | private void OnColumnsChanged(object sender, EventArgs e) {
|
|---|
| 358 | if (Problem == null || Problem.ProblemData == null || Problem.ProblemData.Dataset == null || !Parameters.ContainsKey(DistanceFunctionParameterName)) return;
|
|---|
| 359 | DistanceFunctionParameter.ValidValues.OfType<WeightedEuclideanDistance>().Single().AdaptToProblemData(Problem.ProblemData);
|
|---|
| 360 | }
|
|---|
| 361 |
|
|---|
| 362 | private void RegisterParameterEvents() {
|
|---|
| 363 | PerplexityParameter.Value.ValueChanged += OnPerplexityChanged;
|
|---|
| 364 | }
|
|---|
| 365 |
|
|---|
| 366 | private void OnPerplexityChanged(object sender, EventArgs e) {
|
|---|
| 367 | if (Problem == null || Problem.ProblemData == null || Problem.ProblemData.Dataset == null || !Parameters.ContainsKey(PerplexityParameterName)) return;
|
|---|
| 368 | PerplexityParameter.Value.Value = Math.Max(1, Math.Min((Problem.ProblemData.Dataset.Rows - 1) / 3.0, Perplexity));
|
|---|
| 369 | }
|
|---|
| 370 | #endregion
|
|---|
| 371 |
|
|---|
| 372 | #region Helpers
|
|---|
| 373 | private void SetUpResults(IReadOnlyList<int> allIndices) {
|
|---|
| 374 | if (Results == null) return;
|
|---|
| 375 | var results = Results;
|
|---|
| 376 | dataRowIndices = new Dictionary<string, IList<int>>();
|
|---|
| 377 | var problemData = Problem.ProblemData;
|
|---|
| 378 |
|
|---|
| 379 | if (!results.ContainsKey(IterationResultName)) results.Add(new Result(IterationResultName, new IntValue(0)));
|
|---|
| 380 | if (!results.ContainsKey(ErrorResultName)) results.Add(new Result(ErrorResultName, new DoubleValue(0)));
|
|---|
| 381 | if (!results.ContainsKey(ScatterPlotResultName)) results.Add(new Result(ScatterPlotResultName, "Plot of the projected data", new ScatterPlot(DataResultName, "")));
|
|---|
| 382 | if (!results.ContainsKey(DataResultName)) results.Add(new Result(DataResultName, "Projected Data", new DoubleMatrix()));
|
|---|
| 383 | if (!results.ContainsKey(ErrorPlotResultName)) {
|
|---|
| 384 | var errortable = new DataTable(ErrorPlotResultName, "Development of errors during gradient descent") {
|
|---|
| 385 | VisualProperties = {
|
|---|
| 386 | XAxisTitle = "UpdateIntervall",
|
|---|
| 387 | YAxisTitle = "Error",
|
|---|
| 388 | YAxisLogScale = true
|
|---|
| 389 | }
|
|---|
| 390 | };
|
|---|
| 391 | errortable.Rows.Add(new DataRow("Errors"));
|
|---|
| 392 | errortable.Rows["Errors"].VisualProperties.StartIndexZero = true;
|
|---|
| 393 | results.Add(new Result(ErrorPlotResultName, errortable));
|
|---|
| 394 | }
|
|---|
| 395 |
|
|---|
| 396 | //color datapoints acording to classes variable (be it double, datetime or string)
|
|---|
| 397 | if (!problemData.Dataset.VariableNames.Contains(ClassesName)) {
|
|---|
| 398 | dataRowIndices.Add("Training", problemData.TrainingIndices.ToList());
|
|---|
| 399 | dataRowIndices.Add("Test", problemData.TestIndices.ToList());
|
|---|
| 400 | return;
|
|---|
| 401 | }
|
|---|
| 402 |
|
|---|
| 403 | var classificationData = problemData as ClassificationProblemData;
|
|---|
| 404 | if (classificationData != null && classificationData.TargetVariable.Equals(ClassesName)) {
|
|---|
| 405 | var classNames = classificationData.ClassValues.Zip(classificationData.ClassNames, (v, n) => new {v, n}).ToDictionary(x => x.v, x => x.n);
|
|---|
| 406 | var classes = classificationData.Dataset.GetDoubleValues(classificationData.TargetVariable, allIndices).Select(v => classNames[v]).ToArray();
|
|---|
| 407 | for (var i = 0; i < classes.Length; i++) {
|
|---|
| 408 | if (!dataRowIndices.ContainsKey(classes[i])) dataRowIndices.Add(classes[i], new List<int>());
|
|---|
| 409 | dataRowIndices[classes[i]].Add(i);
|
|---|
| 410 | }
|
|---|
| 411 | } else if (((Dataset)problemData.Dataset).VariableHasType<string>(ClassesName)) {
|
|---|
| 412 | var classes = problemData.Dataset.GetStringValues(ClassesName, allIndices).ToArray();
|
|---|
| 413 | for (var i = 0; i < classes.Length; i++) {
|
|---|
| 414 | if (!dataRowIndices.ContainsKey(classes[i])) dataRowIndices.Add(classes[i], new List<int>());
|
|---|
| 415 | dataRowIndices[classes[i]].Add(i);
|
|---|
| 416 | }
|
|---|
| 417 | } else if (((Dataset)problemData.Dataset).VariableHasType<double>(ClassesName)) {
|
|---|
| 418 | var clusterdata = new Dataset(problemData.Dataset.DoubleVariables, problemData.Dataset.DoubleVariables.Select(v => problemData.Dataset.GetDoubleValues(v, allIndices).ToList()));
|
|---|
| 419 | const int contours = 8;
|
|---|
| 420 | Dictionary<int, string> contourMap;
|
|---|
| 421 | IClusteringModel clusterModel;
|
|---|
| 422 | double[][] borders;
|
|---|
| 423 | CreateClusters(clusterdata, ClassesName, contours, out clusterModel, out contourMap, out borders);
|
|---|
| 424 | var contourorder = borders.Select((x, i) => new {x, i}).OrderBy(x => x.x[0]).Select(x => x.i).ToArray();
|
|---|
| 425 | for (var i = 0; i < contours; i++) {
|
|---|
| 426 | var c = contourorder[i];
|
|---|
| 427 | var contourname = contourMap[c];
|
|---|
| 428 | dataRowIndices.Add(contourname, new List<int>());
|
|---|
| 429 | var row = new ScatterPlotDataRow(contourname, "", new List<Point2D<double>>()) {VisualProperties = {Color = GetHeatMapColor(i, contours), PointSize = 8}};
|
|---|
| 430 | ((ScatterPlot)results[ScatterPlotResultName].Value).Rows.Add(row);
|
|---|
| 431 | }
|
|---|
| 432 | var allClusters = clusterModel.GetClusterValues(clusterdata, Enumerable.Range(0, clusterdata.Rows)).ToArray();
|
|---|
| 433 | for (var i = 0; i < clusterdata.Rows; i++) dataRowIndices[contourMap[allClusters[i] - 1]].Add(i);
|
|---|
| 434 | } else if (((Dataset)problemData.Dataset).VariableHasType<DateTime>(ClassesName)) {
|
|---|
| 435 | var clusterdata = new Dataset(problemData.Dataset.DateTimeVariables, problemData.Dataset.DateTimeVariables.Select(v => problemData.Dataset.GetDoubleValues(v, allIndices).ToList()));
|
|---|
| 436 | const int contours = 8;
|
|---|
| 437 | Dictionary<int, string> contourMap;
|
|---|
| 438 | IClusteringModel clusterModel;
|
|---|
| 439 | double[][] borders;
|
|---|
| 440 | CreateClusters(clusterdata, ClassesName, contours, out clusterModel, out contourMap, out borders);
|
|---|
| 441 | var contourorder = borders.Select((x, i) => new {x, i}).OrderBy(x => x.x[0]).Select(x => x.i).ToArray();
|
|---|
| 442 | for (var i = 0; i < contours; i++) {
|
|---|
| 443 | var c = contourorder[i];
|
|---|
| 444 | var contourname = contourMap[c];
|
|---|
| 445 | dataRowIndices.Add(contourname, new List<int>());
|
|---|
| 446 | var row = new ScatterPlotDataRow(contourname, "", new List<Point2D<double>>()) {VisualProperties = {Color = GetHeatMapColor(i, contours), PointSize = 8}};
|
|---|
| 447 | row.VisualProperties.PointSize = 8;
|
|---|
| 448 | ((ScatterPlot)results[ScatterPlotResultName].Value).Rows.Add(row);
|
|---|
| 449 | }
|
|---|
| 450 | var allClusters = clusterModel.GetClusterValues(clusterdata, Enumerable.Range(0, clusterdata.Rows)).ToArray();
|
|---|
| 451 | for (var i = 0; i < clusterdata.Rows; i++) dataRowIndices[contourMap[allClusters[i] - 1]].Add(i);
|
|---|
| 452 | } else {
|
|---|
| 453 | dataRowIndices.Add("Training", problemData.TrainingIndices.ToList());
|
|---|
| 454 | dataRowIndices.Add("Test", problemData.TestIndices.ToList());
|
|---|
| 455 | }
|
|---|
| 456 | }
|
|---|
| 457 |
|
|---|
| 458 | private void Analyze(TSNEStatic<double[]>.TSNEState tsneState) {
|
|---|
| 459 | if (Results == null) return;
|
|---|
| 460 | var results = Results;
|
|---|
| 461 | var plot = results[ErrorPlotResultName].Value as DataTable;
|
|---|
| 462 | if (plot == null) throw new ArgumentException("Could not create/access error data table in results collection.");
|
|---|
| 463 | var errors = plot.Rows["Errors"].Values;
|
|---|
| 464 | var c = tsneState.EvaluateError();
|
|---|
| 465 | errors.Add(c);
|
|---|
| 466 | ((IntValue)results[IterationResultName].Value).Value = tsneState.iter;
|
|---|
| 467 | ((DoubleValue)results[ErrorResultName].Value).Value = errors.Last();
|
|---|
| 468 |
|
|---|
| 469 | var ndata = NormalizeProjectedData(tsneState.newData);
|
|---|
| 470 | results[DataResultName].Value = new DoubleMatrix(ndata);
|
|---|
| 471 | var splot = results[ScatterPlotResultName].Value as ScatterPlot;
|
|---|
| 472 | FillScatterPlot(ndata, splot, dataRowIndices);
|
|---|
| 473 | }
|
|---|
| 474 |
|
|---|
| 475 | private static void FillScatterPlot(double[,] lowDimData, ScatterPlot plot, Dictionary<string, IList<int>> dataRowIndices) {
|
|---|
| 476 | foreach (var rowName in dataRowIndices.Keys) {
|
|---|
| 477 | if (!plot.Rows.ContainsKey(rowName)) {
|
|---|
| 478 | plot.Rows.Add(new ScatterPlotDataRow(rowName, "", new List<Point2D<double>>()));
|
|---|
| 479 | plot.Rows[rowName].VisualProperties.PointSize = 8;
|
|---|
| 480 | }
|
|---|
| 481 | plot.Rows[rowName].Points.Replace(dataRowIndices[rowName].Select(i => new Point2D<double>(lowDimData[i, 0], lowDimData[i, 1])));
|
|---|
| 482 | }
|
|---|
| 483 | }
|
|---|
| 484 |
|
|---|
| 485 | private static double[,] NormalizeProjectedData(double[,] data) {
|
|---|
| 486 | var max = new double[data.GetLength(1)];
|
|---|
| 487 | var min = new double[data.GetLength(1)];
|
|---|
| 488 | var res = new double[data.GetLength(0), data.GetLength(1)];
|
|---|
| 489 | for (var i = 0; i < max.Length; i++) max[i] = min[i] = data[0, i];
|
|---|
| 490 | for (var i = 0; i < data.GetLength(0); i++)
|
|---|
| 491 | for (var j = 0; j < data.GetLength(1); j++) {
|
|---|
| 492 | var v = data[i, j];
|
|---|
| 493 | max[j] = Math.Max(max[j], v);
|
|---|
| 494 | min[j] = Math.Min(min[j], v);
|
|---|
| 495 | }
|
|---|
| 496 | for (var i = 0; i < data.GetLength(0); i++) {
|
|---|
| 497 | for (var j = 0; j < data.GetLength(1); j++) {
|
|---|
| 498 | var d = max[j] - min[j];
|
|---|
| 499 | var s = data[i, j] - (max[j] + min[j]) / 2; //shift data
|
|---|
| 500 | if (d.IsAlmost(0)) res[i, j] = data[i, j]; //no scaling possible
|
|---|
| 501 | else res[i, j] = s / d; //scale data
|
|---|
| 502 | }
|
|---|
| 503 | }
|
|---|
| 504 | return res;
|
|---|
| 505 | }
|
|---|
| 506 |
|
|---|
| 507 | private static double[][] NormalizeInputData(IReadOnlyList<IReadOnlyList<double>> data) {
|
|---|
| 508 | // as in tSNE implementation by van der Maaten
|
|---|
| 509 | var n = data[0].Count;
|
|---|
| 510 | var mean = new double[n];
|
|---|
| 511 | var max = new double[n];
|
|---|
| 512 | var nData = new double[data.Count][];
|
|---|
| 513 | for (var i = 0; i < n; i++) {
|
|---|
| 514 | mean[i] = Enumerable.Range(0, data.Count).Select(x => data[x][i]).Average();
|
|---|
| 515 | max[i] = Enumerable.Range(0, data.Count).Max(x => Math.Abs(data[x][i]));
|
|---|
| 516 | }
|
|---|
| 517 | for (var i = 0; i < data.Count; i++) {
|
|---|
| 518 | nData[i] = new double[n];
|
|---|
| 519 | for (var j = 0; j < n; j++)
|
|---|
| 520 | nData[i][j] = max[j].IsAlmost(0) ? data[i][j] - mean[j] : (data[i][j] - mean[j]) / max[j];
|
|---|
| 521 | }
|
|---|
| 522 | return nData;
|
|---|
| 523 | }
|
|---|
| 524 |
|
|---|
| 525 | private static Color GetHeatMapColor(int contourNr, int noContours) {
|
|---|
| 526 | return ConvertTotalToRgb(0, noContours, contourNr);
|
|---|
| 527 | }
|
|---|
| 528 |
|
|---|
| 529 | private static void CreateClusters(IDataset data, string target, int contours, out IClusteringModel contourCluster, out Dictionary<int, string> contourNames, out double[][] borders) {
|
|---|
| 530 | var cpd = new ClusteringProblemData((Dataset)data, new[] {target});
|
|---|
| 531 | contourCluster = KMeansClustering.CreateKMeansSolution(cpd, contours, 3).Model;
|
|---|
| 532 |
|
|---|
| 533 | borders = Enumerable.Range(0, contours).Select(x => new[] {double.MaxValue, double.MinValue}).ToArray();
|
|---|
| 534 | var clusters = contourCluster.GetClusterValues(cpd.Dataset, cpd.AllIndices).ToArray();
|
|---|
| 535 | var targetvalues = cpd.Dataset.GetDoubleValues(target).ToArray();
|
|---|
| 536 | foreach (var i in cpd.AllIndices) {
|
|---|
| 537 | var cl = clusters[i] - 1;
|
|---|
| 538 | var clv = targetvalues[i];
|
|---|
| 539 | if (borders[cl][0] > clv) borders[cl][0] = clv;
|
|---|
| 540 | if (borders[cl][1] < clv) borders[cl][1] = clv;
|
|---|
| 541 | }
|
|---|
| 542 |
|
|---|
| 543 | contourNames = new Dictionary<int, string>();
|
|---|
| 544 | for (var i = 0; i < contours; i++)
|
|---|
| 545 | contourNames.Add(i, "[" + borders[i][0] + ";" + borders[i][1] + "]");
|
|---|
| 546 | }
|
|---|
| 547 |
|
|---|
| 548 | private static Color ConvertTotalToRgb(double low, double high, double cell) {
|
|---|
| 549 | var colorGradient = ColorGradient.Colors;
|
|---|
| 550 | var range = high - low;
|
|---|
| 551 | var h = Math.Min(cell / range * colorGradient.Count, colorGradient.Count - 1);
|
|---|
| 552 | return colorGradient[(int)h];
|
|---|
| 553 | }
|
|---|
| 554 | #endregion
|
|---|
| 555 | }
|
|---|
| 556 | } |
|---|
