1 | using System;
|
---|
2 | using System.Collections.Generic;
|
---|
3 | using System.Linq;
|
---|
4 | using HeuristicLab.Common;
|
---|
5 | using HeuristicLab.Core;
|
---|
6 | using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
|
---|
7 | using HeuristicLab.Problems.DataAnalysis;
|
---|
8 |
|
---|
9 | namespace GradientBoostedTrees {
|
---|
10 | [StorableClass]
|
---|
11 | [Item("RegressionTreeModel", "Represents a decision tree for regression.")]
|
---|
12 | // TODO: Implement a view for this
|
---|
13 | public class RegressionTreeModel : NamedItem, IRegressionModel {
|
---|
14 |
|
---|
15 | // trees are represented as a flat array
|
---|
16 | // object-graph-travesal has problems if this is defined as a struct. TODO investigate...
|
---|
17 | //[StorableClass]
|
---|
18 | public struct TreeNode {
|
---|
19 | public readonly static string NO_VARIABLE = string.Empty;
|
---|
20 | //[Storable]
|
---|
21 | public string varName; // name of the variable for splitting or -1 if terminal node
|
---|
22 | //[Storable]
|
---|
23 | public double val; // threshold
|
---|
24 | //[Storable]
|
---|
25 | public int leftIdx;
|
---|
26 | //[Storable]
|
---|
27 | public int rightIdx;
|
---|
28 |
|
---|
29 | //public TreeNode() {
|
---|
30 | // varName = NO_VARIABLE;
|
---|
31 | // leftIdx = -1;
|
---|
32 | // rightIdx = -1;
|
---|
33 | //}
|
---|
34 | //[StorableConstructor]
|
---|
35 | //private TreeNode(bool deserializing) { }
|
---|
36 | public override int GetHashCode()
|
---|
37 | {
|
---|
38 | return (leftIdx * rightIdx) ^ val.GetHashCode();
|
---|
39 | }
|
---|
40 | }
|
---|
41 |
|
---|
42 | [Storable]
|
---|
43 | public readonly TreeNode[] tree;
|
---|
44 |
|
---|
45 | [StorableConstructor]
|
---|
46 | private RegressionTreeModel(bool serializing) : base(serializing) { }
|
---|
47 | // cloning ctor
|
---|
48 | public RegressionTreeModel(RegressionTreeModel original, Cloner cloner)
|
---|
49 | : base(original, cloner) {
|
---|
50 | this.tree = original.tree; // shallow clone, tree must be readonly
|
---|
51 | }
|
---|
52 |
|
---|
53 | public RegressionTreeModel(TreeNode[] tree)
|
---|
54 | : base("RegressionTreeModel", "Represents a decision tree for regression.") {
|
---|
55 | this.tree = tree;
|
---|
56 | }
|
---|
57 |
|
---|
58 | private static double GetPredictionForRow(TreeNode[] t, int nodeIdx, IDataset ds, int row) {
|
---|
59 | var node = t[nodeIdx];
|
---|
60 | if (node.varName == TreeNode.NO_VARIABLE)
|
---|
61 | return node.val;
|
---|
62 | else if (ds.GetDoubleValue(node.varName, row) <= node.val)
|
---|
63 | return GetPredictionForRow(t, node.leftIdx, ds, row);
|
---|
64 | else
|
---|
65 | return GetPredictionForRow(t, node.rightIdx, ds, row);
|
---|
66 | }
|
---|
67 |
|
---|
68 | public override IDeepCloneable Clone(Cloner cloner) {
|
---|
69 | return new RegressionTreeModel(this, cloner);
|
---|
70 | }
|
---|
71 |
|
---|
72 | public IEnumerable<double> GetEstimatedValues(IDataset ds, IEnumerable<int> rows) {
|
---|
73 | return rows.Select(r => GetPredictionForRow(tree, 0, ds, r));
|
---|
74 | }
|
---|
75 |
|
---|
76 | public IRegressionSolution CreateRegressionSolution(IRegressionProblemData problemData) {
|
---|
77 | return new RegressionSolution(this, new RegressionProblemData(problemData));
|
---|
78 | }
|
---|
79 | }
|
---|
80 |
|
---|
81 | }
|
---|