-
Notifications
You must be signed in to change notification settings - Fork 0
/
bp.cpp
328 lines (299 loc) · 8.95 KB
/
bp.cpp
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
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
#include<cstdio>
#include<algorithm>
#include<cstring>
#include<cmath>
#include<vector>
#include<windows.h>
#include<time.h>
#define random(x) (rand()%x)
using namespace std; //妈妈说写程序要养成写注释的好习惯
vector <vector<double> > input; //没每一组数据的输入
vector <vector<double> > label; //每一组数据的标签
vector <int> layer_nodesnum; //每一层有几个节点
vector <vector<double> > output; //输出的结果,跟label相对应
int trainnum,inputnum,layernum,labelnum; //几组训练数据,每一组数据有几个输入,总共有几层,输出应该有几个神经元
vector <double> learningrate;
double wishloss;
double sigmod_d(double x)
{
double temp;
temp = exp(0.0-x);
return temp/((1.0+temp)*(1.0+temp));
}
double sigmod(double x)
{
double temp;
temp = exp(0.0-x);
return 1.0/(1.0+temp);
}
void readfile()
{
freopen("input.txt","r",stdin);
scanf("%d %d %d",&trainnum,&inputnum,&labelnum);
input.resize(trainnum);
label.resize(trainnum);
for(int i=0;i<trainnum;i++)
{
double temp;
for(int j=0;j<inputnum;j++)
{
scanf("%lf",&temp);
input[i].push_back(temp);
}
for(int j=0;j<labelnum;j++)
{
scanf("%lf",&temp);
label[i].push_back(temp);
}
}
scanf("%d",&layernum);
for(int i=0;i<layernum;i++)
{
int temp;
scanf("%d",&temp);
layer_nodesnum.push_back(temp);
}
for(int i=0;i<layernum;i++)
{
double temp;
scanf("%lf",&temp);
learningrate.push_back(temp);
}
scanf("%lf",&wishloss);
fclose(stdin);
}
enum type{inputlayer,hiddenlayer,outputlayer};
struct node
{
vector <double> w;
vector <double> w_change;
double d;
double d_change;
type nodetype;
double value;
};
vector <vector<node> > layer; //每一层的节点情况
void init()
{
output.resize(trainnum);
for(int i=0;i<trainnum;i++)
for(int j=0;j<labelnum;j++)
output[i].push_back(0.0);
layer.resize(layernum);
for(int i=0;i<layernum;i++)
{
for(int j=0;j<layer_nodesnum[i];j++)
{
node temp;
if(i==0)
{
temp.nodetype = inputlayer;
}
else if(i==layernum-1)
{
temp.nodetype = outputlayer;
temp.d = (double)random(100)/100.0;
for(int k=0;k<layer_nodesnum[i-1];k++)
{
temp.w.push_back((double)random(100)/100.0);
temp.w_change.push_back(0.0);
}
}
else
{
temp.nodetype = hiddenlayer;
temp.d = (double)random(100)/100.0;
for(int k=0;k<layer_nodesnum[i-1];k++)
{
temp.w.push_back((double)random(100)/100.0);
temp.w_change.push_back(0.0);
}
}
layer[i].push_back(temp);
}
}
}
double compute_loss_sum(void)
{
double sum = 0.0;
for(int i=0;i<trainnum;i++)
for(int j=0;j<labelnum;j++)
sum=sum+(label[i][j]-output[i][j])*(label[i][j]-output[i][j]);
sum = sum/2.0/trainnum;
return sum;
}
double compute_loss(int k)
{
double sum=0.0;
for(int i=0;i<labelnum;i++)
{
sum = sum + (label[k][i]-output[k][i])*(label[k][i]-output[k][i]);
}
return sum/2.0;
}
void fw(int k)
{
for(int i=0;i<layer[0].size();i++)
{
node temp = layer[0][i];
temp.value = input[k][i];
layer[0][i] = temp;
}
for(int i=1;i<layernum;i++)
{
for(int j=0;j<layer_nodesnum[i];j++)
{
node temp = layer[i][j];
double tempp = 0.0;
for(int p=0;p<layer_nodesnum[i-1];p++)
{
tempp = tempp + temp.w[p]*layer[i-1][p].value;
}
tempp = tempp + temp.d;
temp.value = sigmod(tempp);
layer[i][j] = temp;
}
}
/*printf("%d :",k);
for(int i=0;i<inputnum;i++)
printf("%lf ",input[k][i]);
printf("\n");
for(int i=0;i<labelnum;i++)
{
output[k][i]=layer[layernum-1][i].value;
printf("%lf ",output[k][i]);
}
printf("\n");
Sleep(1000);*/
for(int i=0;i<labelnum;i++)
output[k][i]=layer[layernum-1][i].value;
}
//反向传播写的简直吐血,我日你先人
void bw(int k)
{
int current = layernum-1;
for(int i=0;i<layer_nodesnum[current];i++) //先计算最后一层
{
double change1 = -1.0*(label[k][i]-output[k][i]); //总残差对最后一层第i个神经元的影响
double change2 = output[k][i]*(1.0-output[k][i]); //对sigmoid之前的影响
for(int j=0;j<layer_nodesnum[current-1];j++)
{
double change3 = layer[current-1][j].value; //对第j个参数的影响
double change = change1*change2*change3;
layer[current][i].w_change[j] = change;
}
layer[current][i].d_change = change1*change2; //理解为对上一层的总影响,也是用来修改阈值的
}
current--;
while(current>0) //处理到最后一个隐含层,第0层为输入层
{
for(int i=0;i<layer_nodesnum[current];i++)
{
double change1 = 0.0;
for(int p=0;p<layer_nodesnum[current+1];p++)
{
change1 = change1+layer[current+1][p].d_change*layer[current+1][p].w[i]; //把对我所有有影响的层对我的影响累加起来
}
double change2 = layer[current][i].value*(1.0-layer[current][i].value); //对sigmoid之前的影响
for(int j=0;j<layer_nodesnum[current-1];j++)
{
double change3 = layer[current-1][j].value; //上一层的输入值
double change = change1*change2*change3;
layer[current][i].w_change[j] = change;
}
layer[current][i].d_change = change1*change2;
}
current--;
}
for(int i=1;i<layernum;i++)
{
for(int j=0;j<layer_nodesnum[i];j++)
{
for(int p=0;p<layer_nodesnum[i-1];p++)
{
layer[i][j].w[p] = layer[i][j].w[p] - learningrate[i]*layer[i][j].w_change[p];
}
layer[i][j].d = layer[i][j].d - learningrate[i]*layer[i][j].d_change;
}
}
}
/*void rengong_fuck(void)
{
layer[1][0].w[0]=0.0543;
layer[1][0].w[1]=0.0579;
layer[1][0].d=-0.0703;
layer[1][1].w[0]=-0.0291;
layer[1][1].w[1]=0.0999;
layer[1][1].d=-0.0939;
layer[2][0].w[0]=0.0801;
layer[2][0].w[1]=-0.0605;
layer[2][0].d=-0.0109;
}*/
int main(void)
{
srand((int)time(0));
readfile();
init();
//rengong_fuck(); //手动输入初始权
int ans=0;
do
{
ans++;
for(int i=0;i<trainnum;i++)
{
fw(i);
bw(i);
/* if(i==0&&ans==1) //为了看一下第一次的梯度情况
{
for(int i=1;i<layernum;i++)
{
printf("the %d layer:\n",i);
for(int j=0;j<layer_nodesnum[i];j++)
{
for(int k=0;k<layer_nodesnum[i-1];k++)
{
printf("%lf ",layer[i][j].w_change[k]);
}
printf("%lf\n",layer[i][j].d_change);
}
}
}*/
}
}
while(compute_loss_sum()>wishloss);
/* for(int i=1;i<layernum;i++) //输出最后神经元的参数
{
printf("the %d layer:\n",i);
for(int j=0;j<layer_nodesnum[i];j++)
{
for(int k=0;k<layer_nodesnum[i-1];k++)
{
printf("%lf ",layer[i][j].w[k]);
}
printf("%lf\n",layer[i][j].d);
}
}*/
/*for(int i=0;i<trainnum;i++) //用训练完的网络测试数据分析结果
{
fw(i);
for(int j=0;j<inputnum;j++)
printf("%lf ",input[i][j]);
printf("\n");
for(int j=0;j<labelnum;j++)
printf("%lf ",output[i][j]);
printf("\n");
}*/
/*for(int i=0;i<trainnum;i++) //老子用来测试输入的
{
for(int j=0;j<input[i].size();j++)
{
printf("%lf ",input[i][j]);
}
printf("\n");
printf("%lf\n",label[i]);
}
for(int i=0;i<layer_nodesnum.size();i++)
printf("%d " ,layer_nodesnum[i]);*/
printf("%dtimes\n",ans);
return 0;
}