1 | #region License Information
|
---|
2 | /* HeuristicLab
|
---|
3 | * Copyright (C) 2002-2008 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.Linq;
|
---|
25 | using System.Text;
|
---|
26 | using HeuristicLab.Core;
|
---|
27 | using System.Xml;
|
---|
28 | using System.Diagnostics;
|
---|
29 | using HeuristicLab.DataAnalysis;
|
---|
30 |
|
---|
31 | namespace HeuristicLab.GP.StructureIdentification {
|
---|
32 | /// <summary>
|
---|
33 | /// Evaluates FunctionTrees recursively by interpretation of the function symbols in each node.
|
---|
34 | /// Not thread-safe!
|
---|
35 | /// </summary>
|
---|
36 | public class HL2TreeEvaluator : ItemBase, ITreeEvaluator {
|
---|
37 | private const double EPSILON = 1.0e-10;
|
---|
38 | private double estimatedValueMax;
|
---|
39 | private double estimatedValueMin;
|
---|
40 |
|
---|
41 | private class Instr {
|
---|
42 | public double d_arg0;
|
---|
43 | public short i_arg0;
|
---|
44 | public short i_arg1;
|
---|
45 | public byte arity;
|
---|
46 | public byte symbol;
|
---|
47 | public IFunction function;
|
---|
48 | }
|
---|
49 |
|
---|
50 | private Instr[] codeArr;
|
---|
51 | private int PC;
|
---|
52 | private Dataset dataset;
|
---|
53 | private int sampleIndex;
|
---|
54 |
|
---|
55 | public void ResetEvaluator(Dataset dataset, int targetVariable, int start, int end, double punishmentFactor) {
|
---|
56 | this.dataset = dataset;
|
---|
57 | double maximumPunishment = punishmentFactor * dataset.GetRange(targetVariable, start, end);
|
---|
58 |
|
---|
59 | // get the mean of the values of the target variable to determine the max and min bounds of the estimated value
|
---|
60 | double targetMean = dataset.GetMean(targetVariable, start, end);
|
---|
61 | estimatedValueMin = targetMean - maximumPunishment;
|
---|
62 | estimatedValueMax = targetMean + maximumPunishment;
|
---|
63 | }
|
---|
64 |
|
---|
65 | private Instr TranslateToInstr(LightWeightFunction f) {
|
---|
66 | Instr instr = new Instr();
|
---|
67 | instr.arity = f.arity;
|
---|
68 | instr.symbol = EvaluatorSymbolTable.MapFunction(f.functionType);
|
---|
69 | switch (instr.symbol) {
|
---|
70 | case EvaluatorSymbolTable.DIFFERENTIAL:
|
---|
71 | case EvaluatorSymbolTable.VARIABLE: {
|
---|
72 | instr.i_arg0 = (short)f.data[0]; // var
|
---|
73 | instr.d_arg0 = f.data[1]; // weight
|
---|
74 | instr.i_arg1 = (short)f.data[2]; // sample-offset
|
---|
75 | break;
|
---|
76 | }
|
---|
77 | case EvaluatorSymbolTable.CONSTANT: {
|
---|
78 | instr.d_arg0 = f.data[0]; // value
|
---|
79 | break;
|
---|
80 | }
|
---|
81 | case EvaluatorSymbolTable.UNKNOWN: {
|
---|
82 | instr.function = f.functionType;
|
---|
83 | break;
|
---|
84 | }
|
---|
85 | }
|
---|
86 | return instr;
|
---|
87 | }
|
---|
88 |
|
---|
89 | public double Evaluate(IFunctionTree functionTree, int sampleIndex) {
|
---|
90 | BakedFunctionTree bakedTree = functionTree as BakedFunctionTree;
|
---|
91 | if (bakedTree == null) throw new ArgumentException("HL2Evaluator can only evaluate BakedFunctionTrees");
|
---|
92 |
|
---|
93 | List<LightWeightFunction> linearRepresentation = bakedTree.LinearRepresentation;
|
---|
94 | codeArr = new Instr[linearRepresentation.Count];
|
---|
95 | int i = 0;
|
---|
96 | foreach (LightWeightFunction f in linearRepresentation) {
|
---|
97 | codeArr[i++] = TranslateToInstr(f);
|
---|
98 | }
|
---|
99 |
|
---|
100 | PC = 0;
|
---|
101 | this.sampleIndex = sampleIndex;
|
---|
102 |
|
---|
103 | double estimated = EvaluateBakedCode();
|
---|
104 | if (double.IsNaN(estimated) || double.IsInfinity(estimated)) {
|
---|
105 | estimated = estimatedValueMax;
|
---|
106 | } else if (estimated > estimatedValueMax) {
|
---|
107 | estimated = estimatedValueMax;
|
---|
108 | } else if (estimated < estimatedValueMin) {
|
---|
109 | estimated = estimatedValueMin;
|
---|
110 | }
|
---|
111 | return estimated;
|
---|
112 | }
|
---|
113 |
|
---|
114 | // skips a whole branch
|
---|
115 | private void SkipBakedCode() {
|
---|
116 | int i = 1;
|
---|
117 | while (i > 0) {
|
---|
118 | i += codeArr[PC++].arity;
|
---|
119 | i--;
|
---|
120 | }
|
---|
121 | }
|
---|
122 |
|
---|
123 | private double EvaluateBakedCode() {
|
---|
124 | Instr currInstr = codeArr[PC++];
|
---|
125 | switch (currInstr.symbol) {
|
---|
126 | case EvaluatorSymbolTable.VARIABLE: {
|
---|
127 | int row = sampleIndex + currInstr.i_arg1;
|
---|
128 | if (row < 0 || row >= dataset.Rows) return double.NaN;
|
---|
129 | else return currInstr.d_arg0 * dataset.GetValue(row, currInstr.i_arg0);
|
---|
130 | }
|
---|
131 | case EvaluatorSymbolTable.CONSTANT: {
|
---|
132 | return currInstr.d_arg0;
|
---|
133 | }
|
---|
134 | case EvaluatorSymbolTable.DIFFERENTIAL: {
|
---|
135 | int row = sampleIndex + currInstr.i_arg1;
|
---|
136 | if (row < 1 || row >= dataset.Rows) return double.NaN;
|
---|
137 | else {
|
---|
138 | double prevValue = dataset.GetValue(row - 1, currInstr.i_arg0);
|
---|
139 | return currInstr.d_arg0 * (dataset.GetValue(row, currInstr.i_arg0) - prevValue);
|
---|
140 | }
|
---|
141 | }
|
---|
142 | case EvaluatorSymbolTable.MULTIPLICATION: {
|
---|
143 | double result = EvaluateBakedCode();
|
---|
144 | for (int i = 1; i < currInstr.arity; i++) {
|
---|
145 | result *= EvaluateBakedCode();
|
---|
146 | }
|
---|
147 | return result;
|
---|
148 | }
|
---|
149 | case EvaluatorSymbolTable.ADDITION: {
|
---|
150 | double sum = EvaluateBakedCode();
|
---|
151 | for (int i = 1; i < currInstr.arity; i++) {
|
---|
152 | sum += EvaluateBakedCode();
|
---|
153 | }
|
---|
154 | return sum;
|
---|
155 | }
|
---|
156 | case EvaluatorSymbolTable.SUBTRACTION: {
|
---|
157 | return EvaluateBakedCode() - EvaluateBakedCode();
|
---|
158 | }
|
---|
159 | case EvaluatorSymbolTable.DIVISION: {
|
---|
160 | double arg0 = EvaluateBakedCode();
|
---|
161 | double arg1 = EvaluateBakedCode();
|
---|
162 | if (double.IsNaN(arg0) || double.IsNaN(arg1)) return double.NaN;
|
---|
163 | if (Math.Abs(arg1) < (10e-20)) return 0.0; else return (arg0 / arg1);
|
---|
164 | }
|
---|
165 | case EvaluatorSymbolTable.COSINUS: {
|
---|
166 | return Math.Cos(EvaluateBakedCode());
|
---|
167 | }
|
---|
168 | case EvaluatorSymbolTable.SINUS: {
|
---|
169 | return Math.Sin(EvaluateBakedCode());
|
---|
170 | }
|
---|
171 | case EvaluatorSymbolTable.EXP: {
|
---|
172 | return Math.Exp(EvaluateBakedCode());
|
---|
173 | }
|
---|
174 | case EvaluatorSymbolTable.LOG: {
|
---|
175 | return Math.Log(EvaluateBakedCode());
|
---|
176 | }
|
---|
177 | case EvaluatorSymbolTable.POWER: {
|
---|
178 | double x = EvaluateBakedCode();
|
---|
179 | double p = EvaluateBakedCode();
|
---|
180 | return Math.Pow(x, p);
|
---|
181 | }
|
---|
182 | case EvaluatorSymbolTable.SIGNUM: {
|
---|
183 | double value = EvaluateBakedCode();
|
---|
184 | if (double.IsNaN(value)) return double.NaN;
|
---|
185 | if (value < 0.0) return -1.0;
|
---|
186 | if (value > 0.0) return 1.0;
|
---|
187 | return 0.0;
|
---|
188 | }
|
---|
189 | case EvaluatorSymbolTable.SQRT: {
|
---|
190 | return Math.Sqrt(EvaluateBakedCode());
|
---|
191 | }
|
---|
192 | case EvaluatorSymbolTable.TANGENS: {
|
---|
193 | return Math.Tan(EvaluateBakedCode());
|
---|
194 | }
|
---|
195 | case EvaluatorSymbolTable.AND: { // only defined for inputs 1 and 0
|
---|
196 | double result = EvaluateBakedCode();
|
---|
197 | bool hasNaNBranch = false;
|
---|
198 | for (int i = 1; i < currInstr.arity; i++) {
|
---|
199 | if (result < 0.5 || double.IsNaN(result)) hasNaNBranch |= double.IsNaN(EvaluateBakedCode());
|
---|
200 | else {
|
---|
201 | result = EvaluateBakedCode();
|
---|
202 | }
|
---|
203 | }
|
---|
204 | if (hasNaNBranch || double.IsNaN(result)) return double.NaN;
|
---|
205 | if (result < 0.5) return 0.0;
|
---|
206 | return 1.0;
|
---|
207 | }
|
---|
208 | case EvaluatorSymbolTable.EQU: {
|
---|
209 | double x = EvaluateBakedCode();
|
---|
210 | double y = EvaluateBakedCode();
|
---|
211 | if (double.IsNaN(x) || double.IsNaN(y)) return double.NaN;
|
---|
212 | // direct comparison of double values is most likely incorrect but
|
---|
213 | // that's the way how it is implemented in the standard HL2 function library
|
---|
214 | if (x == y) return 1.0; else return 0.0;
|
---|
215 | }
|
---|
216 | case EvaluatorSymbolTable.GT: {
|
---|
217 | double x = EvaluateBakedCode();
|
---|
218 | double y = EvaluateBakedCode();
|
---|
219 | if (double.IsNaN(x) || double.IsNaN(y)) return double.NaN;
|
---|
220 | if (x > y) return 1.0;
|
---|
221 | return 0.0;
|
---|
222 | }
|
---|
223 | case EvaluatorSymbolTable.IFTE: { // only defined for condition 0 or 1
|
---|
224 | double condition = EvaluateBakedCode();
|
---|
225 | double result;
|
---|
226 | bool hasNaNBranch = false;
|
---|
227 | if (double.IsNaN(condition)) return double.NaN;
|
---|
228 | if (condition > 0.5) {
|
---|
229 | result = EvaluateBakedCode(); hasNaNBranch = double.IsNaN(EvaluateBakedCode());
|
---|
230 | } else {
|
---|
231 | hasNaNBranch = double.IsNaN(EvaluateBakedCode()); result = EvaluateBakedCode();
|
---|
232 | }
|
---|
233 | if (hasNaNBranch) return double.NaN;
|
---|
234 | return result;
|
---|
235 | }
|
---|
236 | case EvaluatorSymbolTable.LT: {
|
---|
237 | double x = EvaluateBakedCode();
|
---|
238 | double y = EvaluateBakedCode();
|
---|
239 | if (double.IsNaN(x) || double.IsNaN(y)) return double.NaN;
|
---|
240 | if (x < y) return 1.0;
|
---|
241 | return 0.0;
|
---|
242 | }
|
---|
243 | case EvaluatorSymbolTable.NOT: { // only defined for inputs 0 or 1
|
---|
244 | double result = EvaluateBakedCode();
|
---|
245 | if (double.IsNaN(result)) return double.NaN;
|
---|
246 | if (result < 0.5) return 1.0;
|
---|
247 | return 0.0;
|
---|
248 | }
|
---|
249 | case EvaluatorSymbolTable.OR: { // only defined for inputs 0 or 1
|
---|
250 | double result = EvaluateBakedCode();
|
---|
251 | bool hasNaNBranch = false;
|
---|
252 | for (int i = 1; i < currInstr.arity; i++) {
|
---|
253 | if (double.IsNaN(result) || result > 0.5) hasNaNBranch |= double.IsNaN(EvaluateBakedCode());
|
---|
254 | else
|
---|
255 | result = EvaluateBakedCode();
|
---|
256 | }
|
---|
257 | if (hasNaNBranch || double.IsNaN(result)) return double.NaN;
|
---|
258 | if (result > 0.5) return 1.0;
|
---|
259 | return 0.0;
|
---|
260 | }
|
---|
261 | default: {
|
---|
262 | throw new NotImplementedException();
|
---|
263 | }
|
---|
264 | }
|
---|
265 | }
|
---|
266 | }
|
---|
267 | }
|
---|