Repository navigation
Expand file tree
/
Copy pathbranch.m
More file actions
executable file
·213 lines (183 loc) · 8.81 KB
/
Copy pathbranch.m
File metadata and controls
executable file
·213 lines (183 loc) · 8.81 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
%%% Author: Ruobai Feng, Zhaoming Zhang, Takeshi Kondoh
classdef branch
properties
Qx
depth
nNode %size of subtree
cp %cut probability
par %exist only split found
magnitude
entropy
BL %exist only split found
BR %exist only split found
PQ %probability distribution
end
methods
function obj=branch(Qx,depth,magnitude,entropy,PQ)
obj.Qx=Qx;
obj.depth=depth;
obj.magnitude=magnitude;
obj.entropy=entropy;
obj.PQ=PQ;
end
%%% Author: Ruobai Feng
function parent=grow(parent,data,maxdepth,clmax,train_function)
[QLx_, QRx_, par_, entropyQL_, entropyQR_, PQL_, PQR_, split_found] = ...
train_function(data, parent.Qx, parent.entropy, clmax);
% if no split is found, stop at current node. else
if(split_found==1 && parent.depth<=maxdepth)
parent.par = par_;
parent.BL=branch(QLx_,parent.depth+1,length(QLx_),entropyQL_,PQL_);
parent.BR=branch(QRx_,parent.depth+1,length(QRx_),entropyQR_,PQR_);
% display(parent)
parent.BL=grow(parent.BL,data,maxdepth,clmax,train_function);
parent.BR=grow(parent.BR,data,maxdepth,clmax,train_function);
end
end
%%% Author: Ruobai Feng, Zhaoming Zhang
function [parent, cut_max] = ULS(parent, data, Qx, clmax, maxdepth, cut_max, train_func, test_func)
if isempty(parent.par)
return;
end
[QLx, QRx] = test_func(data, Qx, parent.par);
parent.BL.Qx = [parent.BL.Qx, QLx];
parent.BL.magnitude = length(parent.BL.Qx);
parent.BR.Qx = [parent.BR.Qx, QRx];
parent.BR.magnitude = length(parent.BR.Qx);
[~, parent.BL.PQ, parent.BR.PQ, parent.BL.entropy, parent.BR.entropy] = ...
gain_entropy(parent.entropy, parent.BL.Qx, parent.BR.Qx, data, clmax);
[parent.BL, ~] = ULS(parent.BL, data, QLx, clmax, maxdepth, cut_max, train_func, test_func);
[parent.BR, ~] = ULS(parent.BR, data, QRx, clmax, maxdepth, cut_max, train_func, test_func);
end
%%% Author: Zhaoming Zhang, Takeshi Kondoh
function [parent, cut_max] = IGT(parent, data, Qx, clmax, maxdepth, cut_max, train_func, test_func)
if isempty(parent.par)
parent = grow(parent,data,maxdepth,clmax,train_func);
return;
end
[QLx, QRx] = test_func(data, Qx, parent.par);
parent.BL.Qx = [parent.BL.Qx, QLx];
parent.BL.magnitude = length(parent.BL.Qx);
parent.BR.Qx = [parent.BR.Qx, QRx];
parent.BR.magnitude = length(parent.BR.Qx);
[~, parent.BL.PQ, parent.BR.PQ, parent.BL.entropy, parent.BR.entropy] = ...
gain_entropy(parent.entropy, parent.BL.Qx, parent.BR.Qx, data, clmax);
[parent.BL, ~] = IGT(parent.BL, data, QLx, clmax, maxdepth, cut_max, train_func, test_func);
[parent.BR, ~] = IGT(parent.BR, data, QRx, clmax, maxdepth, cut_max, train_func, test_func);
end
%%% Author: Takeshi Kondoh
function [parent,cut_max] = RTST(parent, data, Qx, clmax, maxdepth, cut_max, train_func, test_func)
% nNode is the number of node in a tree
% cut_max is the maximum number of nodes which will be retrained
if parent.nNode <= cut_max
if rand(1) <= parent.cp;
cut_max = cut_max - parent.nNode;
parent = grow(parent,data,maxdepth,clmax,train_func);
return;
end
end
if isempty(parent.par)
parent = grow(parent,data,maxdepth,clmax,train_func);
return;
end
[QLx, QRx] = test_func(data, Qx, parent.par);
parent.BL.Qx = [parent.BL.Qx, QLx];
parent.BL.magnitude = length(parent.BL.Qx);
parent.BR.Qx = [parent.BR.Qx, QRx];
parent.BR.magnitude = length(parent.BR.Qx);
[~, parent.BL.PQ, parent.BR.PQ, parent.BL.entropy, parent.BR.entropy] = ...
gain_entropy(parent.entropy, parent.BL.Qx, parent.BR.Qx, data, clmax);
[parent.BL,cut_max] = RTST(parent.BL, data, QLx, clmax, maxdepth, cut_max, train_func, test_func);
[parent.BR,cut_max] = RTST(parent.BR, data, QRx, clmax, maxdepth, cut_max, train_func, test_func);
end
%%% Author: Zhaoming Zhang
function [parent,cut_max] = RTSTQ(parent, data, Qx, clmax, maxdepth, cut_max, train_func, test_func)
% nNode is the number of node in a subtree
% cut_max is the maximum number of nodes which will be retrained
if parent.nNode <= cut_max
if rand(1) <= parent.cp;
cut_max = cut_max - parent.nNode;
parent = grow(parent,data,maxdepth,clmax,train_func);
return;
end
end
if isempty(parent.par)
parent = grow(parent,data,maxdepth,clmax,train_func);
return;
end
[QLx, QRx] = test_func(data, Qx, parent.par);
parent.BL.Qx = [parent.BL.Qx, QLx];
parent.BL.magnitude = length(parent.BL.Qx);
parent.BR.Qx = [parent.BR.Qx, QRx];
parent.BR.magnitude = length(parent.BR.Qx);
[~, parent.BL.PQ, parent.BR.PQ, parent.BL.entropy, parent.BR.entropy] = ...
gain_entropy(parent.entropy, parent.BL.Qx, parent.BR.Qx, data, clmax);
[parent.BL,cut_max] = RTSTQ(parent.BL, data, QLx, clmax, maxdepth, cut_max, train_func, test_func);
[parent.BR,cut_max] = RTSTQ(parent.BR, data, QRx, clmax, maxdepth, cut_max, train_func, test_func);
end
%%% Author: Zhaoming Zhang
function parent = subtree_size(parent)
if isempty(parent.par)
parent.nNode = 1;
return
end
parent.BL = subtree_size(parent.BL);
parent.BR = subtree_size(parent.BR);
parent.nNode = parent.BL.nNode + parent.BR.nNode;
end
%%% Author: Zhaoming Zhang
function parent = uniform(parent, uniform_p)
parent.cp = uniform_p;
if isempty(parent.par)
return
end
parent.BL = uniform(parent.BL, uniform_p);
parent.BR = uniform(parent.BR, uniform_p);
end
%%% Author: Zhaoming Zhang
function parent = weighted_uniform(parent, uniform_p)
if isempty(parent.par)
parent.cp = uniform_p;
return
end
parent.BL = weighted_uniform(parent.BL, uniform_p);
parent.BR = weighted_uniform(parent.BR, uniform_p);
parent.cp = uniform_p + (parent.BL.nNode * parent.BL.cp ...
+ parent.BR.nNode * parent.BR.cp) / parent.nNode;
end
%%% Author: Zhaoming Zhang
function [parent, quality_sum] = quality(parent, quality_sum)
if isempty(parent.par)
parent.cp = parent.entropy;
quality_sum = quality_sum + parent.cp;
return
end
[parent.BL, quality_sum] = quality(parent.BL, quality_sum);
[parent.BR, quality_sum] = quality(parent.BR, quality_sum);
parent.cp = parent.entropy + (parent.BL.magnitude * (parent.BL.cp - parent.BL.entropy) ...
+ parent.BR.magnitude * (parent.BR.cp - parent.BR.entropy)) / parent.magnitude;
quality_sum = quality_sum + parent.cp;
end
%%% Author: Zhaoming Zhang
function parent = normalize_cp(parent, quality_sum)
parent.cp = parent.cp / quality_sum;
if isempty(parent.par)
return
end
parent.BL = normalize_cp(parent.BL, quality_sum);
parent.BR = normalize_cp(parent.BR, quality_sum);
end
%%% Author: Ruobai Feng, Zhaoming Zhang
function PQ_out = prediction(parent, data, Qx_in, PQ_out, test_func)
if isempty(parent.par)
for i = 1:length(parent.PQ);
PQ_out(Qx_in,i) = PQ_out(Qx_in,i)+parent.PQ(i);
end
return;
end
[QLx, QRx] = test_func(data, Qx_in, parent.par);
PQ_out = prediction(parent.BL, data, QLx, PQ_out, test_func);
PQ_out = prediction(parent.BR, data, QRx, PQ_out, test_func);
end
end
end