-
Notifications
You must be signed in to change notification settings - Fork 0
/
BPNN_epoch.m
153 lines (134 loc) · 5.01 KB
/
BPNN_epoch.m
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
clc,clear;
%========================Load Data ======================================
format long;
load('data.mat')
d=normolized_data(data(:,3),0.2,0.8);% map data -->range 0.2~0.8
% Input Data-->1(bias_input),x,y,z(normolized)
x=zeros(400,4);
x(:,1)=1;
x(:,2:3)=data(1:400,1:2);
x(:,4)=d(1:400);
%======================= Parameters ==============================
batch=300;
iter=300000; %iteration(epoch)
learning_rate=0.2;
alpha=0.9; % momentum
input_cell=3; % 1(bias_input),x,y
hidden_cell=15;
output_cell=1; % z=f(x,y)
%hidden layer to input layer:w_h2i
w_h2i=rand(hidden_cell,input_cell)-0.5; %-0.5~0.5
%output layer to hidden layer :w_o2h
w_o2h=rand(output_cell,hidden_cell)-0.5; %-0.5~0.5
%========================= Loss ================================
error=zeros(1,300);
E=zeros(1,300);
E_ave=zeros(1,iter);
%=========================Temp array=================================
yout=zeros(output_cell);
dw_o2h=zeros(output_cell,hidden_cell);
dw_o2h_epoch=zeros(output_cell,hidden_cell);
dw_h2i=zeros(hidden_cell,input_cell );
dw_h2i_epoch=zeros(hidden_cell,input_cell);
%=========================Training=============================
for epoch=1:iter
for num=1:batch % #1~#batch data
%=======================Forward===========================
% the value of x0*w0+x1*w1+...
v_hidden=zeros(hidden_cell,1); %hidden layer
v_output=zeros(output_cell,1); %output layer
% input layer to hidden layer
for j=1:hidden_cell
for i=1:input_cell
v_hidden(j,1)=v_hidden(j,1)+x(num,i)*w_h2i(j,i);
end
end
% hidden layer to output layer
for j=1:output_cell
for i=1:hidden_cell
v_output(j,1)=v_output(j,1)+sigmoid(v_hidden(i,1))*w_o2h(j,i);
end
yout(j)=sigmoid(v_output(j,1));
end
error(1,num)=d(num,1)-yout(j);
E(num)=1/2*(error(1,num))^2;
%===================== Back Propagation ==============================
for j=1:output_cell
delta_o2h=error(j,num)*(yout(j)*(1-yout(j)));
for i=1:hidden_cell
dw_o2h(j,i)=alpha*dw_o2h(j,i)+learning_rate*delta_o2h*sigmoid(v_hidden(i,1));
%w_o2h_new(j,i)=w_o2h(j,i)+dw_o2h(j,i);
end
end
for j=1:hidden_cell
delta_h2i=sigmoid(v_hidden(j,1))*(1-sigmoid(v_hidden(j,1)))*delta_o2h*w_o2h(1,j);
for i=1:input_cell
dw_h2i(j,i)=alpha*dw_h2i(j,i)+learning_rate*delta_h2i*sigmoid(x(num,i));
%w_h2i_new(j,i)=w_h2i(j,i)+dw_h2i(j,i);
end
end
dw_o2h_epoch = dw_o2h_epoch + dw_o2h;
dw_h2i_epoch = dw_h2i_epoch + dw_h2i;
end
dw_o2h_epoch = dw_o2h_epoch / batch;
dw_h2i_epoch = dw_h2i_epoch / batch;
w_o2h = w_o2h + dw_o2h_epoch;
w_h2i = w_h2i + dw_h2i_epoch;
E_ave(epoch)=mean(E);
end
%%
%======================== Predict data ==========================
% output cell data=1; % z=f(x,y)
predict_data=zeros(100,1);
for num=301:400
% the value of x0*w0+x1*w1+...
v_hidden=zeros(hidden_cell,1); %hidden layer
v_output=zeros(output_cell,1); %output layer
% input layer to hidden layer
for j=1:hidden_cell
for i=1:input_cell
v_hidden(j,1)=v_hidden(j,1)+x(num,i)*w_h2i(j,i);
end
end
% hidden layer to output layer
for j=1:output_cell
for i=1:hidden_cell
v_output(j,1)=v_output(j,1)+sigmoid(v_hidden(i,1))*w_o2h(j,i);
end
yout(j)=sigmoid(v_output(j,1));
predict_data(num-300,1)=yout(j);
end
error(1,num)=d(num,1)-yout(j);
E(num)=1/2*(error(1,num))^2;
end
%%
figure(1)
subplot(211),plot(1:epoch,E_ave),
title(['Training Loss:',num2str(E_ave(epoch)) ,' Hidden cell :',num2str(hidden_cell),' \eta:',num2str(learning_rate),' \alpha:',num2str(alpha)]);
xlabel('epoch'),ylabel('E ave')
subplot(212),stem(301:400,E(301:400),'.'),
title(['test Loss :',num2str(mean(E(301:400)))]);
xlabel('number of data'),ylabel('E')
suptitle(['E test - E train = ',num2str(mean(E(301:400))-E_ave(epoch))]);
% figure(2)
% subplot(121)
% a=data(301:400,1);
% b=data(301:400,2);
% c=normolized_data(predict_data(1:100,1),min(data(:,3)),max(data(:,3)));% z--> map to original range
% scatter3(a,b,c,'.','r');
% xlabel('x'),ylabel('y'),zlabel('z'),title('100 predict data');
%
% subplot(122)
% a=data(1:300,1);
% b=data(1:300,2);
% c=data(1:300,3);
% scatter3(a,b,c,'.','b')
% xlabel('x'),ylabel('y'),zlabel('z'),title('300 training data')
figure(3)
a=data(301:400,1);
b=data(301:400,2);
c=data(301:400,3);
scatter3(a,b,c,'.','r');
xlabel('x'),ylabel('y'),zlabel('z');
title('100 testing true data')
% Reference--> http://darren1231.pixnet.net/blog/post/338810666-%E9%A1%9E%E7%A5%9E%E7%B6%93%E7%B6%B2%E8%B7%AF%28backpropagation%29-%E7%AD%86%E8%A8%98