1 | #region License Information
|
---|
2 | /* HeuristicLab
|
---|
3 | * Copyright (C) 2002-2018 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.Diagnostics;
|
---|
25 | using System.Linq;
|
---|
26 | using HeuristicLab.Analysis;
|
---|
27 | using HeuristicLab.Collections;
|
---|
28 | using HeuristicLab.Common;
|
---|
29 | using HeuristicLab.Core;
|
---|
30 | using HeuristicLab.Data;
|
---|
31 | using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
|
---|
32 | using HeuristicLab.Optimization;
|
---|
33 | using HeuristicLab.Parameters;
|
---|
34 | using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
|
---|
35 | using HeuristicLab.Problems.DataAnalysis;
|
---|
36 | using HeuristicLab.Problems.DataAnalysis.Symbolic;
|
---|
37 | using HeuristicLab.Problems.Instances;
|
---|
38 | using Variable = HeuristicLab.Problems.DataAnalysis.Symbolic.Variable;
|
---|
39 |
|
---|
40 | namespace HeuristicLab.Problems.DynamicalSystemsModelling {
|
---|
41 | public class Vector {
|
---|
42 | public readonly static Vector Zero = new Vector(new double[0]);
|
---|
43 |
|
---|
44 | public static Vector operator +(Vector a, Vector b) {
|
---|
45 | if (a == Zero) return b;
|
---|
46 | if (b == Zero) return a;
|
---|
47 | Debug.Assert(a.arr.Length == b.arr.Length);
|
---|
48 | var res = new double[a.arr.Length];
|
---|
49 | for (int i = 0; i < res.Length; i++)
|
---|
50 | res[i] = a.arr[i] + b.arr[i];
|
---|
51 | return new Vector(res);
|
---|
52 | }
|
---|
53 | public static Vector operator -(Vector a, Vector b) {
|
---|
54 | if (b == Zero) return a;
|
---|
55 | if (a == Zero) return -b;
|
---|
56 | Debug.Assert(a.arr.Length == b.arr.Length);
|
---|
57 | var res = new double[a.arr.Length];
|
---|
58 | for (int i = 0; i < res.Length; i++)
|
---|
59 | res[i] = a.arr[i] - b.arr[i];
|
---|
60 | return new Vector(res);
|
---|
61 | }
|
---|
62 | public static Vector operator -(Vector v) {
|
---|
63 | if (v == Zero) return Zero;
|
---|
64 | for (int i = 0; i < v.arr.Length; i++)
|
---|
65 | v.arr[i] = -v.arr[i];
|
---|
66 | return v;
|
---|
67 | }
|
---|
68 |
|
---|
69 | public static Vector operator *(double s, Vector v) {
|
---|
70 | if (v == Zero) return Zero;
|
---|
71 | if (s == 0.0) return Zero;
|
---|
72 | var res = new double[v.arr.Length];
|
---|
73 | for (int i = 0; i < res.Length; i++)
|
---|
74 | res[i] = s * v.arr[i];
|
---|
75 | return new Vector(res);
|
---|
76 | }
|
---|
77 |
|
---|
78 | public static Vector operator *(Vector v, double s) {
|
---|
79 | return s * v;
|
---|
80 | }
|
---|
81 | public static Vector operator *(Vector u, Vector v) {
|
---|
82 | if (v == Zero) return Zero;
|
---|
83 | if (u == Zero) return Zero;
|
---|
84 | var res = new double[v.arr.Length];
|
---|
85 | for (int i = 0; i < res.Length; i++)
|
---|
86 | res[i] = u.arr[i] * v.arr[i];
|
---|
87 | return new Vector(res);
|
---|
88 | }
|
---|
89 | public static Vector operator /(double s, Vector v) {
|
---|
90 | if (s == 0.0) return Zero;
|
---|
91 | if (v == Zero) throw new ArgumentException("Division by zero vector");
|
---|
92 | var res = new double[v.arr.Length];
|
---|
93 | for (int i = 0; i < res.Length; i++)
|
---|
94 | res[i] = 1.0 / v.arr[i];
|
---|
95 | return new Vector(res);
|
---|
96 | }
|
---|
97 | public static Vector operator /(Vector v, double s) {
|
---|
98 | return v * 1.0 / s;
|
---|
99 | }
|
---|
100 |
|
---|
101 | public static Vector Sin(Vector s) {
|
---|
102 | var res = new double[s.arr.Length];
|
---|
103 | for (int i = 0; i < res.Length; i++) res[i] = Math.Sin(s.arr[i]);
|
---|
104 | return new Vector(res);
|
---|
105 | }
|
---|
106 | public static Vector Cos(Vector s) {
|
---|
107 | var res = new double[s.arr.Length];
|
---|
108 | for (int i = 0; i < res.Length; i++) res[i] = Math.Cos(s.arr[i]);
|
---|
109 | return new Vector(res);
|
---|
110 | }
|
---|
111 |
|
---|
112 | private readonly double[] arr; // backing array;
|
---|
113 |
|
---|
114 | public Vector(double[] v) {
|
---|
115 | this.arr = v;
|
---|
116 | }
|
---|
117 |
|
---|
118 | public void CopyTo(double[] target) {
|
---|
119 | Debug.Assert(arr.Length <= target.Length);
|
---|
120 | Array.Copy(arr, target, arr.Length);
|
---|
121 | }
|
---|
122 | }
|
---|
123 |
|
---|
124 | [Item("Dynamical Systems Modelling Problem", "TODO")]
|
---|
125 | [Creatable(CreatableAttribute.Categories.GeneticProgrammingProblems, Priority = 900)]
|
---|
126 | [StorableClass]
|
---|
127 | public sealed class Problem : SingleObjectiveBasicProblem<MultiEncoding>, IRegressionProblem, IProblemInstanceConsumer<IRegressionProblemData>, IProblemInstanceExporter<IRegressionProblemData> {
|
---|
128 |
|
---|
129 | #region parameter names
|
---|
130 | private const string ProblemDataParameterName = "Data";
|
---|
131 | private const string TargetVariablesParameterName = "Target variables";
|
---|
132 | private const string FunctionSetParameterName = "Function set";
|
---|
133 | private const string MaximumLengthParameterName = "Size limit";
|
---|
134 | private const string MaximumParameterOptimizationIterationsParameterName = "Max. parameter optimization iterations";
|
---|
135 | private const string NumberOfLatentVariablesParameterName = "Number of latent variables";
|
---|
136 | private const string NumericIntegrationStepsParameterName = "Steps for numeric integration";
|
---|
137 | private const string TrainingEpisodesParameterName = "Training episodes";
|
---|
138 | private const string OptimizeParametersForEpisodesParameterName = "Optimize parameters for episodes";
|
---|
139 | #endregion
|
---|
140 |
|
---|
141 | #region Parameter Properties
|
---|
142 | IParameter IDataAnalysisProblem.ProblemDataParameter { get { return ProblemDataParameter; } }
|
---|
143 |
|
---|
144 | public IValueParameter<IRegressionProblemData> ProblemDataParameter {
|
---|
145 | get { return (IValueParameter<IRegressionProblemData>)Parameters[ProblemDataParameterName]; }
|
---|
146 | }
|
---|
147 | public IValueParameter<ReadOnlyCheckedItemCollection<StringValue>> TargetVariablesParameter {
|
---|
148 | get { return (IValueParameter<ReadOnlyCheckedItemCollection<StringValue>>)Parameters[TargetVariablesParameterName]; }
|
---|
149 | }
|
---|
150 | public IValueParameter<ReadOnlyCheckedItemCollection<StringValue>> FunctionSetParameter {
|
---|
151 | get { return (IValueParameter<ReadOnlyCheckedItemCollection<StringValue>>)Parameters[FunctionSetParameterName]; }
|
---|
152 | }
|
---|
153 | public IFixedValueParameter<IntValue> MaximumLengthParameter {
|
---|
154 | get { return (IFixedValueParameter<IntValue>)Parameters[MaximumLengthParameterName]; }
|
---|
155 | }
|
---|
156 | public IFixedValueParameter<IntValue> MaximumParameterOptimizationIterationsParameter {
|
---|
157 | get { return (IFixedValueParameter<IntValue>)Parameters[MaximumParameterOptimizationIterationsParameterName]; }
|
---|
158 | }
|
---|
159 | public IFixedValueParameter<IntValue> NumberOfLatentVariablesParameter {
|
---|
160 | get { return (IFixedValueParameter<IntValue>)Parameters[NumberOfLatentVariablesParameterName]; }
|
---|
161 | }
|
---|
162 | public IFixedValueParameter<IntValue> NumericIntegrationStepsParameter {
|
---|
163 | get { return (IFixedValueParameter<IntValue>)Parameters[NumericIntegrationStepsParameterName]; }
|
---|
164 | }
|
---|
165 | public IValueParameter<ItemList<IntRange>> TrainingEpisodesParameter {
|
---|
166 | get { return (IValueParameter<ItemList<IntRange>>)Parameters[TrainingEpisodesParameterName]; }
|
---|
167 | }
|
---|
168 | public IFixedValueParameter<BoolValue> OptimizeParametersForEpisodesParameter {
|
---|
169 | get { return (IFixedValueParameter<BoolValue>)Parameters[OptimizeParametersForEpisodesParameterName]; }
|
---|
170 | }
|
---|
171 | #endregion
|
---|
172 |
|
---|
173 | #region Properties
|
---|
174 | public IRegressionProblemData ProblemData {
|
---|
175 | get { return ProblemDataParameter.Value; }
|
---|
176 | set { ProblemDataParameter.Value = value; }
|
---|
177 | }
|
---|
178 | IDataAnalysisProblemData IDataAnalysisProblem.ProblemData { get { return ProblemData; } }
|
---|
179 |
|
---|
180 | public ReadOnlyCheckedItemCollection<StringValue> TargetVariables {
|
---|
181 | get { return TargetVariablesParameter.Value; }
|
---|
182 | }
|
---|
183 |
|
---|
184 | public ReadOnlyCheckedItemCollection<StringValue> FunctionSet {
|
---|
185 | get { return FunctionSetParameter.Value; }
|
---|
186 | }
|
---|
187 |
|
---|
188 | public int MaximumLength {
|
---|
189 | get { return MaximumLengthParameter.Value.Value; }
|
---|
190 | }
|
---|
191 | public int MaximumParameterOptimizationIterations {
|
---|
192 | get { return MaximumParameterOptimizationIterationsParameter.Value.Value; }
|
---|
193 | }
|
---|
194 | public int NumberOfLatentVariables {
|
---|
195 | get { return NumberOfLatentVariablesParameter.Value.Value; }
|
---|
196 | }
|
---|
197 | public int NumericIntegrationSteps {
|
---|
198 | get { return NumericIntegrationStepsParameter.Value.Value; }
|
---|
199 | }
|
---|
200 | public IEnumerable<IntRange> TrainingEpisodes {
|
---|
201 | get { return TrainingEpisodesParameter.Value; }
|
---|
202 | }
|
---|
203 | public bool OptimizeParametersForEpisodes {
|
---|
204 | get { return OptimizeParametersForEpisodesParameter.Value.Value; }
|
---|
205 | }
|
---|
206 |
|
---|
207 | #endregion
|
---|
208 |
|
---|
209 | public event EventHandler ProblemDataChanged;
|
---|
210 |
|
---|
211 | public override bool Maximization {
|
---|
212 | get { return false; } // we minimize NMSE
|
---|
213 | }
|
---|
214 |
|
---|
215 | #region item cloning and persistence
|
---|
216 | // persistence
|
---|
217 | [StorableConstructor]
|
---|
218 | private Problem(bool deserializing) : base(deserializing) { }
|
---|
219 | [StorableHook(HookType.AfterDeserialization)]
|
---|
220 | private void AfterDeserialization() {
|
---|
221 | if (!Parameters.ContainsKey(OptimizeParametersForEpisodesParameterName)) {
|
---|
222 | Parameters.Add(new FixedValueParameter<BoolValue>(OptimizeParametersForEpisodesParameterName, "Flag to select if parameters should be optimized globally or for each episode individually.", new BoolValue(false)));
|
---|
223 | }
|
---|
224 | RegisterEventHandlers();
|
---|
225 | }
|
---|
226 |
|
---|
227 | // cloning
|
---|
228 | private Problem(Problem original, Cloner cloner)
|
---|
229 | : base(original, cloner) {
|
---|
230 | RegisterEventHandlers();
|
---|
231 | }
|
---|
232 | public override IDeepCloneable Clone(Cloner cloner) { return new Problem(this, cloner); }
|
---|
233 | #endregion
|
---|
234 |
|
---|
235 | public Problem()
|
---|
236 | : base() {
|
---|
237 | var targetVariables = new CheckedItemCollection<StringValue>().AsReadOnly(); // HACK: it would be better to provide a new class derived from IDataAnalysisProblem
|
---|
238 | var functions = CreateFunctionSet();
|
---|
239 | Parameters.Add(new ValueParameter<IRegressionProblemData>(ProblemDataParameterName, "The data captured from the dynamical system. Use CSV import functionality to import data.", new RegressionProblemData()));
|
---|
240 | Parameters.Add(new ValueParameter<ReadOnlyCheckedItemCollection<StringValue>>(TargetVariablesParameterName, "Target variables (overrides setting in ProblemData)", targetVariables));
|
---|
241 | Parameters.Add(new ValueParameter<ReadOnlyCheckedItemCollection<StringValue>>(FunctionSetParameterName, "The list of allowed functions", functions));
|
---|
242 | Parameters.Add(new FixedValueParameter<IntValue>(MaximumLengthParameterName, "The maximally allowed length of each expression. Set to a small value (5 - 25). Default = 10", new IntValue(10)));
|
---|
243 | Parameters.Add(new FixedValueParameter<IntValue>(MaximumParameterOptimizationIterationsParameterName, "The maximum number of iterations for optimization of parameters (using L-BFGS). More iterations makes the algorithm slower, fewer iterations might prevent convergence in the optimization scheme. Default = 100", new IntValue(100)));
|
---|
244 | Parameters.Add(new FixedValueParameter<IntValue>(NumberOfLatentVariablesParameterName, "Latent variables (unobserved variables) allow us to produce expressions which are integrated up and can be used in other expressions. They are handled similarly to target variables in forward simulation / integration. The difference to target variables is that there are no data to which the calculated values of latent variables are compared. Set to a small value (0 .. 5) as necessary (default = 0)", new IntValue(0)));
|
---|
245 | Parameters.Add(new FixedValueParameter<IntValue>(NumericIntegrationStepsParameterName, "Number of steps in the numeric integration that are taken from one row to the next (set to 1 to 100). More steps makes the algorithm slower, less steps worsens the accuracy of the numeric integration scheme.", new IntValue(10)));
|
---|
246 | Parameters.Add(new ValueParameter<ItemList<IntRange>>(TrainingEpisodesParameterName, "A list of ranges that should be used for training, each range represents an independent episode. This overrides the TrainingSet parameter in ProblemData.", new ItemList<IntRange>()));
|
---|
247 | Parameters.Add(new FixedValueParameter<BoolValue>(OptimizeParametersForEpisodesParameterName, "Flag to select if parameters should be optimized globally or for each episode individually.", new BoolValue(false)));
|
---|
248 | RegisterEventHandlers();
|
---|
249 | InitAllParameters();
|
---|
250 |
|
---|
251 | // TODO: do not clear selection of target variables when the input variables are changed (keep selected target variables)
|
---|
252 | // TODO: UI hangs when selecting / deselecting input variables because the encoding is updated on each item
|
---|
253 | // TODO: use training range as default training episode
|
---|
254 |
|
---|
255 | }
|
---|
256 |
|
---|
257 | public override double Evaluate(Individual individual, IRandom random) {
|
---|
258 | var trees = individual.Values.Select(v => v.Value).OfType<ISymbolicExpressionTree>().ToArray(); // extract all trees from individual
|
---|
259 |
|
---|
260 | if (OptimizeParametersForEpisodes) {
|
---|
261 | int eIdx = 0;
|
---|
262 | double totalNMSE = 0.0;
|
---|
263 | int totalSize = 0;
|
---|
264 | foreach (var episode in TrainingEpisodes) {
|
---|
265 | double[] optTheta;
|
---|
266 | double nmse;
|
---|
267 | OptimizeForEpisodes(trees, random, new[] { episode }, out optTheta, out nmse);
|
---|
268 | individual["OptTheta_" + eIdx] = new DoubleArray(optTheta); // write back optimized parameters so that we can use them in the Analysis method
|
---|
269 | eIdx++;
|
---|
270 | totalNMSE += nmse * episode.Size;
|
---|
271 | totalSize += episode.Size;
|
---|
272 | }
|
---|
273 | return totalNMSE / totalSize;
|
---|
274 | } else {
|
---|
275 | double[] optTheta;
|
---|
276 | double nmse;
|
---|
277 | OptimizeForEpisodes(trees, random, TrainingEpisodes, out optTheta, out nmse);
|
---|
278 | individual["OptTheta"] = new DoubleArray(optTheta); // write back optimized parameters so that we can use them in the Analysis method
|
---|
279 | return nmse;
|
---|
280 | }
|
---|
281 | }
|
---|
282 |
|
---|
283 | private void OptimizeForEpisodes(ISymbolicExpressionTree[] trees, IRandom random, IEnumerable<IntRange> episodes, out double[] optTheta, out double nmse) {
|
---|
284 | var rows = episodes.SelectMany(e => Enumerable.Range(e.Start, e.End - e.Start)).ToArray();
|
---|
285 | var problemData = ProblemData;
|
---|
286 | var targetVars = TargetVariables.CheckedItems.Select(i => i.Value).ToArray();
|
---|
287 | var latentVariables = Enumerable.Range(1, NumberOfLatentVariables).Select(i => "λ" + i).ToArray(); // TODO: must coincide with the variables which are actually defined in the grammar and also for which we actually have trees
|
---|
288 | var targetValues = new double[rows.Length, targetVars.Length];
|
---|
289 |
|
---|
290 | // collect values of all target variables
|
---|
291 | var colIdx = 0;
|
---|
292 | foreach (var targetVar in targetVars) {
|
---|
293 | int rowIdx = 0;
|
---|
294 | foreach (var value in problemData.Dataset.GetDoubleValues(targetVar, rows)) {
|
---|
295 | targetValues[rowIdx, colIdx] = value;
|
---|
296 | rowIdx++;
|
---|
297 | }
|
---|
298 | colIdx++;
|
---|
299 | }
|
---|
300 |
|
---|
301 | var nodeIdx = new Dictionary<ISymbolicExpressionTreeNode, int>();
|
---|
302 |
|
---|
303 | foreach (var tree in trees) {
|
---|
304 | foreach (var node in tree.Root.IterateNodesPrefix().Where(n => IsConstantNode(n))) {
|
---|
305 | nodeIdx.Add(node, nodeIdx.Count);
|
---|
306 | }
|
---|
307 | }
|
---|
308 |
|
---|
309 | var theta = nodeIdx.Select(_ => random.NextDouble() * 2.0 - 1.0).ToArray(); // init params randomly from Unif(-1,1)
|
---|
310 |
|
---|
311 | optTheta = new double[0];
|
---|
312 | if (theta.Length > 0) {
|
---|
313 | alglib.minlbfgsstate state;
|
---|
314 | alglib.minlbfgsreport report;
|
---|
315 | alglib.minlbfgscreate(Math.Min(theta.Length, 5), theta, out state);
|
---|
316 | alglib.minlbfgssetcond(state, 0.0, 0.0, 0.0, MaximumParameterOptimizationIterations);
|
---|
317 | alglib.minlbfgsoptimize(state, EvaluateObjectiveAndGradient, null,
|
---|
318 | new object[] { trees, targetVars, problemData, nodeIdx, targetValues, episodes.ToArray(), NumericIntegrationSteps, latentVariables }); //TODO: create a type
|
---|
319 | alglib.minlbfgsresults(state, out optTheta, out report);
|
---|
320 |
|
---|
321 | /*
|
---|
322 | *
|
---|
323 | * L-BFGS algorithm results
|
---|
324 |
|
---|
325 | INPUT PARAMETERS:
|
---|
326 | State - algorithm state
|
---|
327 |
|
---|
328 | OUTPUT PARAMETERS:
|
---|
329 | X - array[0..N-1], solution
|
---|
330 | Rep - optimization report:
|
---|
331 | * Rep.TerminationType completetion code:
|
---|
332 | * -7 gradient verification failed.
|
---|
333 | See MinLBFGSSetGradientCheck() for more information.
|
---|
334 | * -2 rounding errors prevent further improvement.
|
---|
335 | X contains best point found.
|
---|
336 | * -1 incorrect parameters were specified
|
---|
337 | * 1 relative function improvement is no more than
|
---|
338 | EpsF.
|
---|
339 | * 2 relative step is no more than EpsX.
|
---|
340 | * 4 gradient norm is no more than EpsG
|
---|
341 | * 5 MaxIts steps was taken
|
---|
342 | * 7 stopping conditions are too stringent,
|
---|
343 | further improvement is impossible
|
---|
344 | * Rep.IterationsCount contains iterations count
|
---|
345 | * NFEV countains number of function calculations
|
---|
346 | */
|
---|
347 | if (report.terminationtype < 0) { nmse = 10E6; return; }
|
---|
348 | }
|
---|
349 |
|
---|
350 | // perform evaluation for optimal theta to get quality value
|
---|
351 | double[] grad = new double[optTheta.Length];
|
---|
352 | nmse = double.NaN;
|
---|
353 | EvaluateObjectiveAndGradient(optTheta, ref nmse, grad,
|
---|
354 | new object[] { trees, targetVars, problemData, nodeIdx, targetValues, episodes.ToArray(), NumericIntegrationSteps, latentVariables });
|
---|
355 | if (double.IsNaN(nmse) || double.IsInfinity(nmse)) { nmse = 10E6; return; } // return a large value (TODO: be consistent by using NMSE)
|
---|
356 | }
|
---|
357 |
|
---|
358 | private static void EvaluateObjectiveAndGradient(double[] x, ref double f, double[] grad, object obj) {
|
---|
359 | var trees = (ISymbolicExpressionTree[])((object[])obj)[0];
|
---|
360 | var targetVariables = (string[])((object[])obj)[1];
|
---|
361 | var problemData = (IRegressionProblemData)((object[])obj)[2];
|
---|
362 | var nodeIdx = (Dictionary<ISymbolicExpressionTreeNode, int>)((object[])obj)[3];
|
---|
363 | var targetValues = (double[,])((object[])obj)[4];
|
---|
364 | var episodes = (IntRange[])((object[])obj)[5];
|
---|
365 | var numericIntegrationSteps = (int)((object[])obj)[6];
|
---|
366 | var latentVariables = (string[])((object[])obj)[7];
|
---|
367 |
|
---|
368 | var predicted = Integrate(
|
---|
369 | trees, // we assume trees contain expressions for the change of each target variable over time y'(t)
|
---|
370 | problemData.Dataset,
|
---|
371 | problemData.AllowedInputVariables.ToArray(),
|
---|
372 | targetVariables,
|
---|
373 | latentVariables,
|
---|
374 | episodes,
|
---|
375 | nodeIdx,
|
---|
376 | x, numericIntegrationSteps).ToArray();
|
---|
377 |
|
---|
378 |
|
---|
379 | // for normalized MSE = 1/variance(t) * MSE(t, pred)
|
---|
380 | // TODO: Perf. (by standardization of target variables before evaluation of all trees)
|
---|
381 | var invVar = Enumerable.Range(0, targetVariables.Length)
|
---|
382 | .Select(c => Enumerable.Range(0, targetValues.GetLength(0)).Select(row => targetValues[row, c])) // column vectors
|
---|
383 | .Select(vec => vec.Variance())
|
---|
384 | .Select(v => 1.0 / v)
|
---|
385 | .ToArray();
|
---|
386 |
|
---|
387 | // objective function is NMSE
|
---|
388 | f = 0.0;
|
---|
389 | int n = predicted.Length;
|
---|
390 | double invN = 1.0 / n;
|
---|
391 | var g = Vector.Zero;
|
---|
392 | int r = 0;
|
---|
393 | foreach (var y_pred in predicted) {
|
---|
394 | for (int c = 0; c < y_pred.Length; c++) {
|
---|
395 |
|
---|
396 | var y_pred_f = y_pred[c].Item1;
|
---|
397 | var y = targetValues[r, c];
|
---|
398 |
|
---|
399 | var res = (y - y_pred_f);
|
---|
400 | var ressq = res * res;
|
---|
401 | f += ressq * invN * invVar[c];
|
---|
402 | g += -2.0 * res * y_pred[c].Item2 * invN * invVar[c];
|
---|
403 | }
|
---|
404 | r++;
|
---|
405 | }
|
---|
406 |
|
---|
407 | g.CopyTo(grad);
|
---|
408 | }
|
---|
409 |
|
---|
410 | public override void Analyze(Individual[] individuals, double[] qualities, ResultCollection results, IRandom random) {
|
---|
411 | base.Analyze(individuals, qualities, results, random);
|
---|
412 |
|
---|
413 | if (!results.ContainsKey("Prediction (training)")) {
|
---|
414 | results.Add(new Result("Prediction (training)", typeof(ReadOnlyItemList<DataTable>)));
|
---|
415 | }
|
---|
416 | if (!results.ContainsKey("Prediction (test)")) {
|
---|
417 | results.Add(new Result("Prediction (test)", typeof(ReadOnlyItemList<DataTable>)));
|
---|
418 | }
|
---|
419 | if (!results.ContainsKey("Models")) {
|
---|
420 | results.Add(new Result("Models", typeof(VariableCollection)));
|
---|
421 | }
|
---|
422 |
|
---|
423 | var bestIndividualAndQuality = this.GetBestIndividual(individuals, qualities);
|
---|
424 | var trees = bestIndividualAndQuality.Item1.Values.Select(v => v.Value).OfType<ISymbolicExpressionTree>().ToArray(); // extract all trees from individual
|
---|
425 |
|
---|
426 | // TODO extract common functionality from Evaluate and Analyze
|
---|
427 | var nodeIdx = new Dictionary<ISymbolicExpressionTreeNode, int>();
|
---|
428 | foreach (var tree in trees) {
|
---|
429 | foreach (var node in tree.Root.IterateNodesPrefix().Where(n => IsConstantNode(n))) {
|
---|
430 | nodeIdx.Add(node, nodeIdx.Count);
|
---|
431 | }
|
---|
432 | }
|
---|
433 | var problemData = ProblemData;
|
---|
434 | var targetVars = TargetVariables.CheckedItems.Select(i => i.Value).ToArray();
|
---|
435 | var latentVariables = Enumerable.Range(1, NumberOfLatentVariables).Select(i => "λ" + i).ToArray(); // TODO: must coincide with the variables which are actually defined in the grammar and also for which we actually have trees
|
---|
436 |
|
---|
437 | var trainingList = new ItemList<DataTable>();
|
---|
438 |
|
---|
439 | if (OptimizeParametersForEpisodes) {
|
---|
440 | var eIdx = 0;
|
---|
441 | var trainingPredictions = new List<Tuple<double, Vector>[][]>();
|
---|
442 | foreach (var episode in TrainingEpisodes) {
|
---|
443 | var episodes = new[] { episode };
|
---|
444 | var optTheta = ((DoubleArray)bestIndividualAndQuality.Item1["OptTheta_" + eIdx]).ToArray(); // see evaluate
|
---|
445 | var trainingPrediction = Integrate(
|
---|
446 | trees, // we assume trees contain expressions for the change of each target variable over time y'(t)
|
---|
447 | problemData.Dataset,
|
---|
448 | problemData.AllowedInputVariables.ToArray(),
|
---|
449 | targetVars,
|
---|
450 | latentVariables,
|
---|
451 | episodes,
|
---|
452 | nodeIdx,
|
---|
453 | optTheta,
|
---|
454 | NumericIntegrationSteps).ToArray();
|
---|
455 | trainingPredictions.Add(trainingPrediction);
|
---|
456 | eIdx++;
|
---|
457 | }
|
---|
458 |
|
---|
459 | // only for actual target values
|
---|
460 | var trainingRows = TrainingEpisodes.SelectMany(e => Enumerable.Range(e.Start, e.End - e.Start));
|
---|
461 | for (int colIdx = 0; colIdx < targetVars.Length; colIdx++) {
|
---|
462 | var targetVar = targetVars[colIdx];
|
---|
463 | var trainingDataTable = new DataTable(targetVar + " prediction (training)");
|
---|
464 | var actualValuesRow = new DataRow(targetVar, "The values of " + targetVar, problemData.Dataset.GetDoubleValues(targetVar, trainingRows));
|
---|
465 | var predictedValuesRow = new DataRow(targetVar + " pred.", "Predicted values for " + targetVar, trainingPredictions.SelectMany(arr => arr.Select(row => row[colIdx].Item1)).ToArray());
|
---|
466 | trainingDataTable.Rows.Add(actualValuesRow);
|
---|
467 | trainingDataTable.Rows.Add(predictedValuesRow);
|
---|
468 | trainingList.Add(trainingDataTable);
|
---|
469 | }
|
---|
470 | results["Prediction (training)"].Value = trainingList.AsReadOnly();
|
---|
471 |
|
---|
472 |
|
---|
473 | var models = new VariableCollection();
|
---|
474 |
|
---|
475 | foreach (var tup in targetVars.Zip(trees, Tuple.Create)) {
|
---|
476 | var targetVarName = tup.Item1;
|
---|
477 | var tree = tup.Item2;
|
---|
478 |
|
---|
479 | var origTreeVar = new HeuristicLab.Core.Variable(targetVarName + "(original)");
|
---|
480 | origTreeVar.Value = (ISymbolicExpressionTree)tree.Clone();
|
---|
481 | models.Add(origTreeVar);
|
---|
482 | }
|
---|
483 | results["Models"].Value = models;
|
---|
484 | } else {
|
---|
485 | var optTheta = ((DoubleArray)bestIndividualAndQuality.Item1["OptTheta"]).ToArray(); // see evaluate
|
---|
486 | var trainingPrediction = Integrate(
|
---|
487 | trees, // we assume trees contain expressions for the change of each target variable over time y'(t)
|
---|
488 | problemData.Dataset,
|
---|
489 | problemData.AllowedInputVariables.ToArray(),
|
---|
490 | targetVars,
|
---|
491 | latentVariables,
|
---|
492 | TrainingEpisodes,
|
---|
493 | nodeIdx,
|
---|
494 | optTheta,
|
---|
495 | NumericIntegrationSteps).ToArray();
|
---|
496 | // only for actual target values
|
---|
497 | var trainingRows = TrainingEpisodes.SelectMany(e => Enumerable.Range(e.Start, e.End - e.Start));
|
---|
498 | for (int colIdx = 0; colIdx < targetVars.Length; colIdx++) {
|
---|
499 | var targetVar = targetVars[colIdx];
|
---|
500 | var trainingDataTable = new DataTable(targetVar + " prediction (training)");
|
---|
501 | var actualValuesRow = new DataRow(targetVar, "The values of " + targetVar, problemData.Dataset.GetDoubleValues(targetVar, trainingRows));
|
---|
502 | var predictedValuesRow = new DataRow(targetVar + " pred.", "Predicted values for " + targetVar, trainingPrediction.Select(arr => arr[colIdx].Item1).ToArray());
|
---|
503 | trainingDataTable.Rows.Add(actualValuesRow);
|
---|
504 | trainingDataTable.Rows.Add(predictedValuesRow);
|
---|
505 | trainingList.Add(trainingDataTable);
|
---|
506 | }
|
---|
507 | // TODO: DRY for training and test
|
---|
508 | var testList = new ItemList<DataTable>();
|
---|
509 | var testRows = ProblemData.TestIndices.ToArray();
|
---|
510 | var testPrediction = Integrate(
|
---|
511 | trees, // we assume trees contain expressions for the change of each target variable over time y'(t)
|
---|
512 | problemData.Dataset,
|
---|
513 | problemData.AllowedInputVariables.ToArray(),
|
---|
514 | targetVars,
|
---|
515 | latentVariables,
|
---|
516 | new IntRange[] { ProblemData.TestPartition },
|
---|
517 | nodeIdx,
|
---|
518 | optTheta,
|
---|
519 | NumericIntegrationSteps).ToArray();
|
---|
520 |
|
---|
521 | for (int colIdx = 0; colIdx < targetVars.Length; colIdx++) {
|
---|
522 | var targetVar = targetVars[colIdx];
|
---|
523 | var testDataTable = new DataTable(targetVar + " prediction (test)");
|
---|
524 | var actualValuesRow = new DataRow(targetVar, "The values of " + targetVar, problemData.Dataset.GetDoubleValues(targetVar, testRows));
|
---|
525 | var predictedValuesRow = new DataRow(targetVar + " pred.", "Predicted values for " + targetVar, testPrediction.Select(arr => arr[colIdx].Item1).ToArray());
|
---|
526 | testDataTable.Rows.Add(actualValuesRow);
|
---|
527 | testDataTable.Rows.Add(predictedValuesRow);
|
---|
528 | testList.Add(testDataTable);
|
---|
529 | }
|
---|
530 |
|
---|
531 | results["Prediction (training)"].Value = trainingList.AsReadOnly();
|
---|
532 | results["Prediction (test)"].Value = testList.AsReadOnly();
|
---|
533 | #region simplification of models
|
---|
534 | // TODO the dependency of HeuristicLab.Problems.DataAnalysis.Symbolic is not ideal
|
---|
535 | var models = new VariableCollection(); // to store target var names and original version of tree
|
---|
536 |
|
---|
537 | foreach (var tup in targetVars.Zip(trees, Tuple.Create)) {
|
---|
538 | var targetVarName = tup.Item1;
|
---|
539 | var tree = tup.Item2;
|
---|
540 |
|
---|
541 | // when we reference HeuristicLab.Problems.DataAnalysis.Symbolic we can translate symbols
|
---|
542 | int nextParIdx = 0;
|
---|
543 | var shownTree = new SymbolicExpressionTree(TranslateTreeNode(tree.Root, optTheta, ref nextParIdx));
|
---|
544 |
|
---|
545 | // var shownTree = (SymbolicExpressionTree)tree.Clone();
|
---|
546 | // var constantsNodeOrig = tree.IterateNodesPrefix().Where(IsConstantNode);
|
---|
547 | // var constantsNodeShown = shownTree.IterateNodesPrefix().Where(IsConstantNode);
|
---|
548 | //
|
---|
549 | // foreach (var n in constantsNodeOrig.Zip(constantsNodeShown, (original, shown) => new { original, shown })) {
|
---|
550 | // double constantsVal = optTheta[nodeIdx[n.original]];
|
---|
551 | //
|
---|
552 | // ConstantTreeNode replacementNode = new ConstantTreeNode(new Constant()) { Value = constantsVal };
|
---|
553 | //
|
---|
554 | // var parentNode = n.shown.Parent;
|
---|
555 | // int replacementIndex = parentNode.IndexOfSubtree(n.shown);
|
---|
556 | // parentNode.RemoveSubtree(replacementIndex);
|
---|
557 | // parentNode.InsertSubtree(replacementIndex, replacementNode);
|
---|
558 | // }
|
---|
559 |
|
---|
560 | var origTreeVar = new HeuristicLab.Core.Variable(targetVarName + "(original)");
|
---|
561 | origTreeVar.Value = (ISymbolicExpressionTree)tree.Clone();
|
---|
562 | models.Add(origTreeVar);
|
---|
563 | var simplifiedTreeVar = new HeuristicLab.Core.Variable(targetVarName + "(simplified)");
|
---|
564 | simplifiedTreeVar.Value = TreeSimplifier.Simplify(shownTree);
|
---|
565 | models.Add(simplifiedTreeVar);
|
---|
566 |
|
---|
567 | }
|
---|
568 | results["Models"].Value = models;
|
---|
569 | #endregion
|
---|
570 | }
|
---|
571 | }
|
---|
572 |
|
---|
573 | private ISymbolicExpressionTreeNode TranslateTreeNode(ISymbolicExpressionTreeNode n, double[] parameterValues, ref int nextParIdx) {
|
---|
574 | ISymbolicExpressionTreeNode translatedNode = null;
|
---|
575 | if (n.Symbol is StartSymbol) {
|
---|
576 | translatedNode = new StartSymbol().CreateTreeNode();
|
---|
577 | } else if (n.Symbol is ProgramRootSymbol) {
|
---|
578 | translatedNode = new ProgramRootSymbol().CreateTreeNode();
|
---|
579 | } else if (n.Symbol.Name == "+") {
|
---|
580 | translatedNode = new Addition().CreateTreeNode();
|
---|
581 | } else if (n.Symbol.Name == "-") {
|
---|
582 | translatedNode = new Subtraction().CreateTreeNode();
|
---|
583 | } else if (n.Symbol.Name == "*") {
|
---|
584 | translatedNode = new Multiplication().CreateTreeNode();
|
---|
585 | } else if (n.Symbol.Name == "%") {
|
---|
586 | translatedNode = new Division().CreateTreeNode();
|
---|
587 | } else if (n.Symbol.Name == "sin") {
|
---|
588 | translatedNode = new Sine().CreateTreeNode();
|
---|
589 | } else if (n.Symbol.Name == "cos") {
|
---|
590 | translatedNode = new Cosine().CreateTreeNode();
|
---|
591 | } else if (IsConstantNode(n)) {
|
---|
592 | var constNode = (ConstantTreeNode)new Constant().CreateTreeNode();
|
---|
593 | constNode.Value = parameterValues[nextParIdx];
|
---|
594 | nextParIdx++;
|
---|
595 | translatedNode = constNode;
|
---|
596 | } else {
|
---|
597 | // assume a variable name
|
---|
598 | var varName = n.Symbol.Name;
|
---|
599 | var varNode = (VariableTreeNode)new Variable().CreateTreeNode();
|
---|
600 | varNode.Weight = 1.0;
|
---|
601 | varNode.VariableName = varName;
|
---|
602 | translatedNode = varNode;
|
---|
603 | }
|
---|
604 | foreach (var child in n.Subtrees) {
|
---|
605 | translatedNode.AddSubtree(TranslateTreeNode(child, parameterValues, ref nextParIdx));
|
---|
606 | }
|
---|
607 | return translatedNode;
|
---|
608 | }
|
---|
609 |
|
---|
610 | #region interpretation
|
---|
611 | private static IEnumerable<Tuple<double, Vector>[]> Integrate(
|
---|
612 | ISymbolicExpressionTree[] trees, IDataset dataset, string[] inputVariables, string[] targetVariables, string[] latentVariables, IEnumerable<IntRange> episodes,
|
---|
613 | Dictionary<ISymbolicExpressionTreeNode, int> nodeIdx, double[] parameterValues, int numericIntegrationSteps = 100) {
|
---|
614 |
|
---|
615 | int NUM_STEPS = numericIntegrationSteps;
|
---|
616 | double h = 1.0 / NUM_STEPS;
|
---|
617 |
|
---|
618 | foreach (var episode in episodes) {
|
---|
619 | var rows = Enumerable.Range(episode.Start, episode.End - episode.Start);
|
---|
620 | // return first value as stored in the dataset
|
---|
621 | yield return targetVariables
|
---|
622 | .Select(targetVar => Tuple.Create(dataset.GetDoubleValue(targetVar, rows.First()), Vector.Zero))
|
---|
623 | .ToArray();
|
---|
624 |
|
---|
625 | // integrate forward starting with known values for the target in t0
|
---|
626 |
|
---|
627 | var variableValues = new Dictionary<string, Tuple<double, Vector>>();
|
---|
628 | var t0 = rows.First();
|
---|
629 | foreach (var varName in inputVariables) {
|
---|
630 | variableValues.Add(varName, Tuple.Create(dataset.GetDoubleValue(varName, t0), Vector.Zero));
|
---|
631 | }
|
---|
632 | foreach (var varName in targetVariables) {
|
---|
633 | variableValues.Add(varName, Tuple.Create(dataset.GetDoubleValue(varName, t0), Vector.Zero));
|
---|
634 | }
|
---|
635 | // add value entries for latent variables which are also integrated
|
---|
636 | foreach (var latentVar in latentVariables) {
|
---|
637 | variableValues.Add(latentVar, Tuple.Create(0.0, Vector.Zero)); // we don't have observations for latent variables -> assume zero as starting value
|
---|
638 | }
|
---|
639 | var calculatedVariables = targetVariables.Concat(latentVariables); // TODO: must conincide with the order of trees in the encoding
|
---|
640 |
|
---|
641 | foreach (var t in rows.Skip(1)) {
|
---|
642 | for (int step = 0; step < NUM_STEPS; step++) {
|
---|
643 | var deltaValues = new Dictionary<string, Tuple<double, Vector>>();
|
---|
644 | foreach (var tup in trees.Zip(calculatedVariables, Tuple.Create)) {
|
---|
645 | var tree = tup.Item1;
|
---|
646 | var targetVarName = tup.Item2;
|
---|
647 | // skip programRoot and startSymbol
|
---|
648 | var res = InterpretRec(tree.Root.GetSubtree(0).GetSubtree(0), variableValues, nodeIdx, parameterValues);
|
---|
649 | deltaValues.Add(targetVarName, res);
|
---|
650 | }
|
---|
651 |
|
---|
652 | // update variableValues for next step
|
---|
653 | foreach (var kvp in deltaValues) {
|
---|
654 | var oldVal = variableValues[kvp.Key];
|
---|
655 | variableValues[kvp.Key] = Tuple.Create(
|
---|
656 | oldVal.Item1 + h * kvp.Value.Item1,
|
---|
657 | oldVal.Item2 + h * kvp.Value.Item2
|
---|
658 | );
|
---|
659 | }
|
---|
660 | }
|
---|
661 |
|
---|
662 | // only return the target variables for calculation of errors
|
---|
663 | yield return targetVariables
|
---|
664 | .Select(targetVar => variableValues[targetVar])
|
---|
665 | .ToArray();
|
---|
666 |
|
---|
667 | // update for next time step
|
---|
668 | foreach (var varName in inputVariables) {
|
---|
669 | variableValues[varName] = Tuple.Create(dataset.GetDoubleValue(varName, t), Vector.Zero);
|
---|
670 | }
|
---|
671 | }
|
---|
672 | }
|
---|
673 | }
|
---|
674 |
|
---|
675 | private static Tuple<double, Vector> InterpretRec(
|
---|
676 | ISymbolicExpressionTreeNode node,
|
---|
677 | Dictionary<string, Tuple<double, Vector>> variableValues,
|
---|
678 | Dictionary<ISymbolicExpressionTreeNode, int> nodeIdx,
|
---|
679 | double[] parameterValues
|
---|
680 | ) {
|
---|
681 |
|
---|
682 | switch (node.Symbol.Name) {
|
---|
683 | case "+": {
|
---|
684 | var l = InterpretRec(node.GetSubtree(0), variableValues, nodeIdx, parameterValues); // TODO capture all parameters into a state type for interpretation
|
---|
685 | var r = InterpretRec(node.GetSubtree(1), variableValues, nodeIdx, parameterValues);
|
---|
686 |
|
---|
687 | return Tuple.Create(l.Item1 + r.Item1, l.Item2 + r.Item2);
|
---|
688 | }
|
---|
689 | case "*": {
|
---|
690 | var l = InterpretRec(node.GetSubtree(0), variableValues, nodeIdx, parameterValues);
|
---|
691 | var r = InterpretRec(node.GetSubtree(1), variableValues, nodeIdx, parameterValues);
|
---|
692 |
|
---|
693 | return Tuple.Create(l.Item1 * r.Item1, l.Item2 * r.Item1 + l.Item1 * r.Item2);
|
---|
694 | }
|
---|
695 |
|
---|
696 | case "-": {
|
---|
697 | var l = InterpretRec(node.GetSubtree(0), variableValues, nodeIdx, parameterValues);
|
---|
698 | var r = InterpretRec(node.GetSubtree(1), variableValues, nodeIdx, parameterValues);
|
---|
699 |
|
---|
700 | return Tuple.Create(l.Item1 - r.Item1, l.Item2 - r.Item2);
|
---|
701 | }
|
---|
702 | case "%": {
|
---|
703 | var l = InterpretRec(node.GetSubtree(0), variableValues, nodeIdx, parameterValues);
|
---|
704 | var r = InterpretRec(node.GetSubtree(1), variableValues, nodeIdx, parameterValues);
|
---|
705 |
|
---|
706 | // protected division
|
---|
707 | if (r.Item1.IsAlmost(0.0)) {
|
---|
708 | return Tuple.Create(0.0, Vector.Zero);
|
---|
709 | } else {
|
---|
710 | return Tuple.Create(
|
---|
711 | l.Item1 / r.Item1,
|
---|
712 | l.Item1 * -1.0 / (r.Item1 * r.Item1) * r.Item2 + 1.0 / r.Item1 * l.Item2 // (f/g)' = f * (1/g)' + 1/g * f' = f * -1/g² * g' + 1/g * f'
|
---|
713 | );
|
---|
714 | }
|
---|
715 | }
|
---|
716 | case "sin": {
|
---|
717 | var x = InterpretRec(node.GetSubtree(0), variableValues, nodeIdx, parameterValues);
|
---|
718 | return Tuple.Create(
|
---|
719 | Math.Sin(x.Item1),
|
---|
720 | Vector.Cos(x.Item2) * x.Item2
|
---|
721 | );
|
---|
722 | }
|
---|
723 | case "cos": {
|
---|
724 | var x = InterpretRec(node.GetSubtree(0), variableValues, nodeIdx, parameterValues);
|
---|
725 | return Tuple.Create(
|
---|
726 | Math.Cos(x.Item1),
|
---|
727 | -Vector.Sin(x.Item2) * x.Item2
|
---|
728 | );
|
---|
729 | }
|
---|
730 | default: {
|
---|
731 | // distinguish other cases
|
---|
732 | if (IsConstantNode(node)) {
|
---|
733 | var vArr = new double[parameterValues.Length]; // backing array for vector
|
---|
734 | vArr[nodeIdx[node]] = 1.0;
|
---|
735 | var g = new Vector(vArr);
|
---|
736 | return Tuple.Create(parameterValues[nodeIdx[node]], g);
|
---|
737 | } else {
|
---|
738 | // assume a variable name
|
---|
739 | var varName = node.Symbol.Name;
|
---|
740 | return variableValues[varName];
|
---|
741 | }
|
---|
742 | }
|
---|
743 | }
|
---|
744 | }
|
---|
745 | #endregion
|
---|
746 |
|
---|
747 | #region events
|
---|
748 | /*
|
---|
749 | * Dependencies between parameters:
|
---|
750 | *
|
---|
751 | * ProblemData
|
---|
752 | * |
|
---|
753 | * V
|
---|
754 | * TargetVariables FunctionSet MaximumLength NumberOfLatentVariables
|
---|
755 | * | | | |
|
---|
756 | * V V | |
|
---|
757 | * Grammar <---------------+-------------------
|
---|
758 | * |
|
---|
759 | * V
|
---|
760 | * Encoding
|
---|
761 | */
|
---|
762 | private void RegisterEventHandlers() {
|
---|
763 | ProblemDataParameter.ValueChanged += ProblemDataParameter_ValueChanged;
|
---|
764 | if (ProblemDataParameter.Value != null) ProblemDataParameter.Value.Changed += ProblemData_Changed;
|
---|
765 |
|
---|
766 | TargetVariablesParameter.ValueChanged += TargetVariablesParameter_ValueChanged;
|
---|
767 | if (TargetVariablesParameter.Value != null) TargetVariablesParameter.Value.CheckedItemsChanged += CheckedTargetVariablesChanged;
|
---|
768 |
|
---|
769 | FunctionSetParameter.ValueChanged += FunctionSetParameter_ValueChanged;
|
---|
770 | if (FunctionSetParameter.Value != null) FunctionSetParameter.Value.CheckedItemsChanged += CheckedFunctionsChanged;
|
---|
771 |
|
---|
772 | MaximumLengthParameter.Value.ValueChanged += MaximumLengthChanged;
|
---|
773 |
|
---|
774 | NumberOfLatentVariablesParameter.Value.ValueChanged += NumLatentVariablesChanged;
|
---|
775 | }
|
---|
776 |
|
---|
777 | private void NumLatentVariablesChanged(object sender, EventArgs e) {
|
---|
778 | UpdateGrammarAndEncoding();
|
---|
779 | }
|
---|
780 |
|
---|
781 | private void MaximumLengthChanged(object sender, EventArgs e) {
|
---|
782 | UpdateGrammarAndEncoding();
|
---|
783 | }
|
---|
784 |
|
---|
785 | private void FunctionSetParameter_ValueChanged(object sender, EventArgs e) {
|
---|
786 | FunctionSetParameter.Value.CheckedItemsChanged += CheckedFunctionsChanged;
|
---|
787 | }
|
---|
788 |
|
---|
789 | private void CheckedFunctionsChanged(object sender, CollectionItemsChangedEventArgs<StringValue> e) {
|
---|
790 | UpdateGrammarAndEncoding();
|
---|
791 | }
|
---|
792 |
|
---|
793 | private void TargetVariablesParameter_ValueChanged(object sender, EventArgs e) {
|
---|
794 | TargetVariablesParameter.Value.CheckedItemsChanged += CheckedTargetVariablesChanged;
|
---|
795 | }
|
---|
796 |
|
---|
797 | private void CheckedTargetVariablesChanged(object sender, CollectionItemsChangedEventArgs<StringValue> e) {
|
---|
798 | UpdateGrammarAndEncoding();
|
---|
799 | }
|
---|
800 |
|
---|
801 | private void ProblemDataParameter_ValueChanged(object sender, EventArgs e) {
|
---|
802 | ProblemDataParameter.Value.Changed += ProblemData_Changed;
|
---|
803 | OnProblemDataChanged();
|
---|
804 | OnReset();
|
---|
805 | }
|
---|
806 |
|
---|
807 | private void ProblemData_Changed(object sender, EventArgs e) {
|
---|
808 | OnProblemDataChanged();
|
---|
809 | OnReset();
|
---|
810 | }
|
---|
811 |
|
---|
812 | private void OnProblemDataChanged() {
|
---|
813 | UpdateTargetVariables(); // implicitly updates other dependent parameters
|
---|
814 | var handler = ProblemDataChanged;
|
---|
815 | if (handler != null) handler(this, EventArgs.Empty);
|
---|
816 | }
|
---|
817 |
|
---|
818 | #endregion
|
---|
819 |
|
---|
820 | #region helper
|
---|
821 |
|
---|
822 | private void InitAllParameters() {
|
---|
823 | UpdateTargetVariables(); // implicitly updates the grammar and the encoding
|
---|
824 | }
|
---|
825 |
|
---|
826 | private ReadOnlyCheckedItemCollection<StringValue> CreateFunctionSet() {
|
---|
827 | var l = new CheckedItemCollection<StringValue>();
|
---|
828 | l.Add(new StringValue("+").AsReadOnly());
|
---|
829 | l.Add(new StringValue("*").AsReadOnly());
|
---|
830 | l.Add(new StringValue("%").AsReadOnly());
|
---|
831 | l.Add(new StringValue("-").AsReadOnly());
|
---|
832 | l.Add(new StringValue("sin").AsReadOnly());
|
---|
833 | l.Add(new StringValue("cos").AsReadOnly());
|
---|
834 | return l.AsReadOnly();
|
---|
835 | }
|
---|
836 |
|
---|
837 | private static bool IsConstantNode(ISymbolicExpressionTreeNode n) {
|
---|
838 | return n.Symbol.Name.StartsWith("θ");
|
---|
839 | }
|
---|
840 | private static bool IsLatentVariableNode(ISymbolicExpressionTreeNode n) {
|
---|
841 | return n.Symbol.Name.StartsWith("λ");
|
---|
842 | }
|
---|
843 |
|
---|
844 |
|
---|
845 | private void UpdateTargetVariables() {
|
---|
846 | var currentlySelectedVariables = TargetVariables.CheckedItems.Select(i => i.Value).ToArray();
|
---|
847 |
|
---|
848 | var newVariablesList = new CheckedItemCollection<StringValue>(ProblemData.Dataset.VariableNames.Select(str => new StringValue(str).AsReadOnly()).ToArray()).AsReadOnly();
|
---|
849 | var matchingItems = newVariablesList.Where(item => currentlySelectedVariables.Contains(item.Value)).ToArray();
|
---|
850 | foreach (var matchingItem in matchingItems) {
|
---|
851 | newVariablesList.SetItemCheckedState(matchingItem, true);
|
---|
852 | }
|
---|
853 | TargetVariablesParameter.Value = newVariablesList;
|
---|
854 | }
|
---|
855 |
|
---|
856 | private void UpdateGrammarAndEncoding() {
|
---|
857 | var encoding = new MultiEncoding();
|
---|
858 | var g = CreateGrammar();
|
---|
859 | foreach (var targetVar in TargetVariables.CheckedItems) {
|
---|
860 | encoding = encoding.Add(new SymbolicExpressionTreeEncoding(targetVar + "_tree", g, MaximumLength, MaximumLength)); // only limit by length
|
---|
861 | }
|
---|
862 | for (int i = 1; i <= NumberOfLatentVariables; i++) {
|
---|
863 | encoding = encoding.Add(new SymbolicExpressionTreeEncoding("λ" + i + "_tree", g, MaximumLength, MaximumLength));
|
---|
864 | }
|
---|
865 | Encoding = encoding;
|
---|
866 | }
|
---|
867 |
|
---|
868 | private ISymbolicExpressionGrammar CreateGrammar() {
|
---|
869 | // whenever ProblemData is changed we create a new grammar with the necessary symbols
|
---|
870 | var g = new SimpleSymbolicExpressionGrammar();
|
---|
871 | g.AddSymbols(FunctionSet.CheckedItems.Select(i => i.Value).ToArray(), 2, 2);
|
---|
872 |
|
---|
873 | // TODO
|
---|
874 | //g.AddSymbols(new[] {
|
---|
875 | // "exp",
|
---|
876 | // "log", // log( <expr> ) // TODO: init a theta to ensure the value is always positive
|
---|
877 | // "exp_minus" // exp((-1) * <expr>
|
---|
878 | //}, 1, 1);
|
---|
879 |
|
---|
880 | foreach (var variableName in ProblemData.AllowedInputVariables.Union(TargetVariables.CheckedItems.Select(i => i.Value)))
|
---|
881 | g.AddTerminalSymbol(variableName);
|
---|
882 |
|
---|
883 | // generate symbols for numeric parameters for which the value is optimized using AutoDiff
|
---|
884 | // we generate multiple symbols to balance the probability for selecting a numeric parameter in the generation of random trees
|
---|
885 | var numericConstantsFactor = 2.0;
|
---|
886 | for (int i = 0; i < numericConstantsFactor * (ProblemData.AllowedInputVariables.Count() + TargetVariables.CheckedItems.Count()); i++) {
|
---|
887 | g.AddTerminalSymbol("θ" + i); // numeric parameter for which the value is optimized using AutoDiff
|
---|
888 | }
|
---|
889 |
|
---|
890 | // generate symbols for latent variables
|
---|
891 | for (int i = 1; i <= NumberOfLatentVariables; i++) {
|
---|
892 | g.AddTerminalSymbol("λ" + i); // numeric parameter for which the value is optimized using AutoDiff
|
---|
893 | }
|
---|
894 |
|
---|
895 | return g;
|
---|
896 | }
|
---|
897 |
|
---|
898 | #endregion
|
---|
899 |
|
---|
900 | #region Import & Export
|
---|
901 | public void Load(IRegressionProblemData data) {
|
---|
902 | Name = data.Name;
|
---|
903 | Description = data.Description;
|
---|
904 | ProblemData = data;
|
---|
905 | }
|
---|
906 |
|
---|
907 | public IRegressionProblemData Export() {
|
---|
908 | return ProblemData;
|
---|
909 | }
|
---|
910 | #endregion
|
---|
911 |
|
---|
912 | }
|
---|
913 | }
|
---|