DeepJava/reduce_sum3d.java at master · dido1998/DeepJava · GitHub

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
packagetensordef;
importbasicops.*;
publicclassreduce_sum3dextendssuperopdef
{
tensorgraphgraph;
backpropagationstructure<reduce_sum3d> curstruct;
tensorarray3darr;
tensorarray3deval[];
addops[];

publicreduce_sum3d(tensorarray3darr,tensorgraphgraph)
	{
this.arr=arr;
this.graph=graph;
ops=newadd[arr.dim1*arr.dim2*arr.dim3];
eval=newtensorarray3d[arr.dim1*arr.dim2*arr.dim3+1];
for(inti=0;i<=arr.dim1*arr.dim2*arr.dim3;i++)
 {
eval[i]=newtensorarray3d(1,1,1,arr.trainable);
 }
curstruct=newbackpropagationstructure<>(this,null,eval[arr.dim1*arr.dim2*arr.dim3]);
graph.addtolist(curstruct);
intc=0;
for(inti=0;i<arr.dim1;i++)
 {
for(intj=0;j<arr.dim2;j++)
 {
for(intk=0;k<arr.dim3;k++)
 {

ops[c]=newadd(arr.arr[i][j][k],eval[c].arr[0][0][0]);
c++;
 }
 }
 }

	}

publictensorarray3dforwardconv()
	{
for(inti=0;i<arr.dim1*arr.dim2*arr.dim3;i++)
 {
eval[i+1].arr[0][0][0].data=ops[i].forward().data;
 }
returneval[arr.dim1*arr.dim2*arr.dim3];
	}

publicvoidbackwardconv(tensorarray3dbackflow)
	{
//System.out.println(backflow.arr[0][0][0].grad);
ops[arr.dim1*arr.dim2*arr.dim3-1].backward(backflow.arr[0][0][0]);

for(inti=arr.dim1*arr.dim2*arr.dim3-2;i>=0;i--)
 {

ops[i].backward(eval[i+1].arr[0][0][0]);
 }

graph.removefromlist(curstruct);
	}

}