1 | #region License Information
|
---|
2 | /* HeuristicLab
|
---|
3 | * Copyright (C) 2002-2017 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
|
---|
4 | *
|
---|
5 | * This file is part of HeuristicLab.
|
---|
6 | *
|
---|
7 | * HeuristicLab is free software: you can redistribute it and/or modify
|
---|
8 | * it under the terms of the GNU General Public License as published by
|
---|
9 | * the Free Software Foundation, either version 3 of the License, or
|
---|
10 | * (at your option) any later version.
|
---|
11 | *
|
---|
12 | * HeuristicLab is distributed in the hope that it will be useful,
|
---|
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
---|
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
---|
15 | * GNU General Public License for more details.
|
---|
16 | *
|
---|
17 | * You should have received a copy of the GNU General Public License
|
---|
18 | * along with HeuristicLab. If not, see <http://www.gnu.org/licenses/>.
|
---|
19 | */
|
---|
20 | #endregion
|
---|
21 |
|
---|
22 | using System.Collections.Generic;
|
---|
23 | using System.Linq;
|
---|
24 | using HeuristicLab.Analysis;
|
---|
25 | using HeuristicLab.Common;
|
---|
26 | using HeuristicLab.Core;
|
---|
27 | using HeuristicLab.Data;
|
---|
28 | using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
|
---|
29 | using HeuristicLab.Optimization;
|
---|
30 | using HeuristicLab.Problems.DataAnalysis;
|
---|
31 | using HEAL.Attic;
|
---|
32 |
|
---|
33 | namespace HeuristicLab.Algorithms.DataAnalysis {
|
---|
34 | public static class RegressionTreeAnalyzer {
|
---|
35 | private const string ConditionResultName = "Condition";
|
---|
36 | private const string CoverResultName = "Covered Instances";
|
---|
37 | private const string CoverageDiagramResultName = "Coverage";
|
---|
38 | private const string RuleModelResultName = "RuleModel";
|
---|
39 |
|
---|
40 | public static Dictionary<string, int> GetRuleVariableFrequences(RegressionRuleSetModel ruleSetModel) {
|
---|
41 | var res = ruleSetModel.VariablesUsedForPrediction.ToDictionary(x => x, x => 0);
|
---|
42 | foreach (var rule in ruleSetModel.Rules)
|
---|
43 | foreach (var att in rule.SplitAttributes)
|
---|
44 | res[att]++;
|
---|
45 | return res;
|
---|
46 | }
|
---|
47 |
|
---|
48 | public static Dictionary<string, int> GetTreeVariableFrequences(RegressionNodeTreeModel treeModel) {
|
---|
49 | var res = treeModel.VariablesUsedForPrediction.ToDictionary(x => x, x => 0);
|
---|
50 | var root = treeModel.Root;
|
---|
51 | foreach (var cur in root.EnumerateNodes().Where(x => !x.IsLeaf))
|
---|
52 | res[cur.SplitAttribute]++;
|
---|
53 | return res;
|
---|
54 | }
|
---|
55 |
|
---|
56 | public static Result CreateLeafDepthHistogram(RegressionNodeTreeModel treeModel) {
|
---|
57 | var list = new List<int>();
|
---|
58 | GetLeafDepths(treeModel.Root, 0, list);
|
---|
59 | var row = new DataRow("Depths", "", list.Select(x => (double)x)) {
|
---|
60 | VisualProperties = {ChartType = DataRowVisualProperties.DataRowChartType.Histogram}
|
---|
61 | };
|
---|
62 | var hist = new DataTable("LeafDepths");
|
---|
63 | hist.Rows.Add(row);
|
---|
64 | return new Result(hist.Name, hist);
|
---|
65 | }
|
---|
66 |
|
---|
67 | public static Result CreateRulesResult(RegressionRuleSetModel ruleSetModel, IRegressionProblemData pd, string resultName, bool displayModels) {
|
---|
68 | var res = new ResultCollection();
|
---|
69 | var i = 0;
|
---|
70 | foreach (var rule in ruleSetModel.Rules)
|
---|
71 | res.Add(new Result("Rule" + i++, CreateRulesResult(rule, pd, displayModels, out pd)));
|
---|
72 | return new Result(resultName, res);
|
---|
73 | }
|
---|
74 |
|
---|
75 | public static IResult CreateCoverageDiagram(RegressionRuleSetModel setModel, IRegressionProblemData problemData) {
|
---|
76 | var res = new DataTable(CoverageDiagramResultName);
|
---|
77 | var training = CountCoverage(setModel, problemData.Dataset, problemData.TrainingIndices);
|
---|
78 | var test = CountCoverage(setModel, problemData.Dataset, problemData.TestIndices);
|
---|
79 | res.Rows.Add(new DataRow("Training", "", training));
|
---|
80 | res.Rows.Add(new DataRow("Test", "", test));
|
---|
81 |
|
---|
82 | foreach (var row in res.Rows)
|
---|
83 | row.VisualProperties.ChartType = DataRowVisualProperties.DataRowChartType.Columns;
|
---|
84 | res.VisualProperties.XAxisMaximumFixedValue = training.Count + 1;
|
---|
85 | res.VisualProperties.XAxisMaximumAuto = false;
|
---|
86 | res.VisualProperties.XAxisMinimumFixedValue = 0;
|
---|
87 | res.VisualProperties.XAxisMinimumAuto = false;
|
---|
88 | res.VisualProperties.XAxisTitle = "Rule";
|
---|
89 | res.VisualProperties.YAxisTitle = "Covered Instances";
|
---|
90 |
|
---|
91 | return new Result(CoverageDiagramResultName, res);
|
---|
92 | }
|
---|
93 |
|
---|
94 | private static void GetLeafDepths(RegressionNodeModel n, int depth, ICollection<int> res) {
|
---|
95 | if (n == null) return;
|
---|
96 | if (n.Left == null && n.Right == null) res.Add(depth);
|
---|
97 | else {
|
---|
98 | GetLeafDepths(n.Left, depth + 1, res);
|
---|
99 | GetLeafDepths(n.Right, depth + 1, res);
|
---|
100 | }
|
---|
101 | }
|
---|
102 |
|
---|
103 | private static IScope CreateRulesResult(RegressionRuleModel regressionRuleModel, IRegressionProblemData pd, bool displayModels, out IRegressionProblemData notCovered) {
|
---|
104 | var training = pd.TrainingIndices.Where(x => !regressionRuleModel.Covers(pd.Dataset, x)).ToArray();
|
---|
105 | var test = pd.TestIndices.Where(x => !regressionRuleModel.Covers(pd.Dataset, x)).ToArray();
|
---|
106 | if (training.Length > 0 || test.Length > 0) {
|
---|
107 | var data = new Dataset(pd.Dataset.DoubleVariables, pd.Dataset.DoubleVariables.Select(v => pd.Dataset.GetDoubleValues(v, training.Concat(test)).ToArray()));
|
---|
108 | notCovered = new RegressionProblemData(data, pd.AllowedInputVariables, pd.TargetVariable);
|
---|
109 | notCovered.TestPartition.Start = notCovered.TrainingPartition.End = training.Length;
|
---|
110 | notCovered.TestPartition.End = training.Length + test.Length;
|
---|
111 | } else notCovered = null;
|
---|
112 |
|
---|
113 | var training2 = pd.TrainingIndices.Where(x => regressionRuleModel.Covers(pd.Dataset, x)).ToArray();
|
---|
114 | var test2 = pd.TestIndices.Where(x => regressionRuleModel.Covers(pd.Dataset, x)).ToArray();
|
---|
115 | var data2 = new Dataset(pd.Dataset.DoubleVariables, pd.Dataset.DoubleVariables.Select(v => pd.Dataset.GetDoubleValues(v, training2.Concat(test2)).ToArray()));
|
---|
116 | var covered = new RegressionProblemData(data2, pd.AllowedInputVariables, pd.TargetVariable);
|
---|
117 | covered.TestPartition.Start = covered.TrainingPartition.End = training2.Length;
|
---|
118 | covered.TestPartition.End = training2.Length + test2.Length;
|
---|
119 |
|
---|
120 | var res2 = new Scope("RuleModels");
|
---|
121 | res2.Variables.Add(new Variable(ConditionResultName, new StringValue(regressionRuleModel.ToCompactString())));
|
---|
122 | res2.Variables.Add(new Variable(CoverResultName, new IntValue(pd.TrainingIndices.Count() - training.Length)));
|
---|
123 | if (displayModels)
|
---|
124 | res2.Variables.Add(new Variable(RuleModelResultName, regressionRuleModel.CreateRegressionSolution(covered)));
|
---|
125 | return res2;
|
---|
126 | }
|
---|
127 |
|
---|
128 | private static IReadOnlyList<double> CountCoverage(RegressionRuleSetModel setModel, IDataset data, IEnumerable<int> rows) {
|
---|
129 | var rules = setModel.Rules.ToArray();
|
---|
130 | var res = new double[rules.Length];
|
---|
131 | foreach (var row in rows)
|
---|
132 | for (var i = 0; i < rules.Length; i++)
|
---|
133 | if (rules[i].Covers(data, row)) {
|
---|
134 | res[i]++;
|
---|
135 | break;
|
---|
136 | }
|
---|
137 | return res;
|
---|
138 | }
|
---|
139 |
|
---|
140 | public static void AnalyzeNodes(RegressionNodeTreeModel tree, ResultCollection results, IRegressionProblemData pd) {
|
---|
141 | var dict = new Dictionary<int, RegressionNodeModel>();
|
---|
142 | var trainingLeafRows = new Dictionary<int, IReadOnlyList<int>>();
|
---|
143 | var testLeafRows = new Dictionary<int, IReadOnlyList<int>>();
|
---|
144 | var modelNumber = new IntValue(1);
|
---|
145 | var symtree = new SymbolicExpressionTree(MirrorTree(tree.Root, dict, trainingLeafRows, testLeafRows, modelNumber, pd.Dataset, pd.TrainingIndices.ToList(), pd.TestIndices.ToList()));
|
---|
146 | results.AddOrUpdateResult("DecisionTree", symtree);
|
---|
147 |
|
---|
148 | if (dict.Count > 200) return;
|
---|
149 | var models = new Scope("NodeModels");
|
---|
150 | results.AddOrUpdateResult("NodeModels", models);
|
---|
151 | foreach (var m in dict.Keys.OrderBy(x => x))
|
---|
152 | models.Variables.Add(new Variable("Model " + m, dict[m].CreateRegressionSolution(Subselect(pd, trainingLeafRows[m], testLeafRows[m]))));
|
---|
153 | }
|
---|
154 |
|
---|
155 | public static void PruningChart(RegressionNodeTreeModel tree, ComplexityPruning pruning, ResultCollection results) {
|
---|
156 | var nodes = new Queue<RegressionNodeModel>();
|
---|
157 | nodes.Enqueue(tree.Root);
|
---|
158 | var max = 0.0;
|
---|
159 | var strenghts = new SortedList<double, int>();
|
---|
160 | while (nodes.Count > 0) {
|
---|
161 | var n = nodes.Dequeue();
|
---|
162 |
|
---|
163 | if (n.IsLeaf) {
|
---|
164 | max++;
|
---|
165 | continue;
|
---|
166 | }
|
---|
167 |
|
---|
168 | if (!strenghts.ContainsKey(n.PruningStrength)) strenghts.Add(n.PruningStrength, 0);
|
---|
169 | strenghts[n.PruningStrength]++;
|
---|
170 | nodes.Enqueue(n.Left);
|
---|
171 | nodes.Enqueue(n.Right);
|
---|
172 | }
|
---|
173 | if (strenghts.Count == 0) return;
|
---|
174 |
|
---|
175 | var plot = new ScatterPlot("Pruned Sizes", "") {
|
---|
176 | VisualProperties = {
|
---|
177 | XAxisTitle = "Pruning Strength",
|
---|
178 | YAxisTitle = "Tree Size",
|
---|
179 | XAxisMinimumAuto = false,
|
---|
180 | XAxisMinimumFixedValue = 0
|
---|
181 | }
|
---|
182 | };
|
---|
183 | var row = new ScatterPlotDataRow("TreeSizes", "", new List<Point2D<double>>());
|
---|
184 | row.Points.Add(new Point2D<double>(pruning.PruningStrength, max));
|
---|
185 |
|
---|
186 | var fillerDots = new Queue<double>();
|
---|
187 | var minX = pruning.PruningStrength;
|
---|
188 | var maxX = strenghts.Last().Key;
|
---|
189 | var size = (maxX - minX) / 200;
|
---|
190 | for (var x = minX; x <= maxX; x += size) {
|
---|
191 | fillerDots.Enqueue(x);
|
---|
192 | }
|
---|
193 |
|
---|
194 | foreach (var strenght in strenghts.Keys) {
|
---|
195 | while (fillerDots.Count > 0 && strenght > fillerDots.Peek())
|
---|
196 | row.Points.Add(new Point2D<double>(fillerDots.Dequeue(), max));
|
---|
197 | max -= strenghts[strenght];
|
---|
198 | row.Points.Add(new Point2D<double>(strenght, max));
|
---|
199 | }
|
---|
200 |
|
---|
201 |
|
---|
202 | row.VisualProperties.PointSize = 6;
|
---|
203 | plot.Rows.Add(row);
|
---|
204 | results.AddOrUpdateResult("PruningSizes", plot);
|
---|
205 | }
|
---|
206 |
|
---|
207 |
|
---|
208 | private static IRegressionProblemData Subselect(IRegressionProblemData data, IReadOnlyList<int> training, IReadOnlyList<int> test) {
|
---|
209 | var dataset = RegressionTreeUtilities.ReduceDataset(data.Dataset, training.Concat(test).ToList(), data.AllowedInputVariables.ToList(), data.TargetVariable);
|
---|
210 | var res = new RegressionProblemData(dataset, data.AllowedInputVariables, data.TargetVariable);
|
---|
211 | res.TrainingPartition.Start = 0;
|
---|
212 | res.TrainingPartition.End = training.Count;
|
---|
213 | res.TestPartition.Start = training.Count;
|
---|
214 | res.TestPartition.End = training.Count + test.Count;
|
---|
215 | return res;
|
---|
216 | }
|
---|
217 |
|
---|
218 | private static SymbolicExpressionTreeNode MirrorTree(RegressionNodeModel regressionNode, IDictionary<int, RegressionNodeModel> dict,
|
---|
219 | IDictionary<int, IReadOnlyList<int>> trainingLeafRows,
|
---|
220 | IDictionary<int, IReadOnlyList<int>> testLeafRows,
|
---|
221 | IntValue nextId, IDataset data, IReadOnlyList<int> trainingRows, IReadOnlyList<int> testRows) {
|
---|
222 | if (regressionNode.IsLeaf) {
|
---|
223 | var i = nextId.Value++;
|
---|
224 | dict.Add(i, regressionNode);
|
---|
225 | trainingLeafRows.Add(i, trainingRows);
|
---|
226 | testLeafRows.Add(i, testRows);
|
---|
227 | return new SymbolicExpressionTreeNode(new TextSymbol("Model " + i + "\n(" + trainingRows.Count + "/" + testRows.Count + ")"));
|
---|
228 | }
|
---|
229 |
|
---|
230 | var pftext = "\npf = " + regressionNode.PruningStrength.ToString("0.###");
|
---|
231 | var text = regressionNode.SplitAttribute + " <= " + regressionNode.SplitValue.ToString("0.###");
|
---|
232 | if (!double.IsNaN(regressionNode.PruningStrength)) text += pftext;
|
---|
233 |
|
---|
234 | var textNode = new SymbolicExpressionTreeNode(new TextSymbol(text));
|
---|
235 | IReadOnlyList<int> lTrainingRows, rTrainingRows;
|
---|
236 | IReadOnlyList<int> lTestRows, rTestRows;
|
---|
237 | RegressionTreeUtilities.SplitRows(trainingRows, data, regressionNode.SplitAttribute, regressionNode.SplitValue, out lTrainingRows, out rTrainingRows);
|
---|
238 | RegressionTreeUtilities.SplitRows(testRows, data, regressionNode.SplitAttribute, regressionNode.SplitValue, out lTestRows, out rTestRows);
|
---|
239 |
|
---|
240 | textNode.AddSubtree(MirrorTree(regressionNode.Left, dict, trainingLeafRows, testLeafRows, nextId, data, lTrainingRows, lTestRows));
|
---|
241 | textNode.AddSubtree(MirrorTree(regressionNode.Right, dict, trainingLeafRows, testLeafRows, nextId, data, rTrainingRows, rTestRows));
|
---|
242 |
|
---|
243 | return textNode;
|
---|
244 | }
|
---|
245 |
|
---|
246 |
|
---|
247 | [StorableType("D5540C63-310B-4D6F-8A3D-6C1A08DE7F80")]
|
---|
248 | private sealed class TextSymbol : Symbol {
|
---|
249 | [StorableConstructor]
|
---|
250 | private TextSymbol(StorableConstructorFlag _) : base(_) { }
|
---|
251 | private TextSymbol(Symbol original, Cloner cloner) : base(original, cloner) { }
|
---|
252 | public TextSymbol(string name) : base(name, "") {
|
---|
253 | Name = name;
|
---|
254 | }
|
---|
255 | public override IDeepCloneable Clone(Cloner cloner) {
|
---|
256 | return new TextSymbol(this, cloner);
|
---|
257 | }
|
---|
258 | public override int MinimumArity {
|
---|
259 | get { return 0; }
|
---|
260 | }
|
---|
261 | public override int MaximumArity {
|
---|
262 | get { return int.MaxValue; }
|
---|
263 | }
|
---|
264 | }
|
---|
265 | }
|
---|
266 | } |
---|