PiCNN/LSoftmax.h at master · pms67/PiCNN · 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
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
#ifndef SOFTMAX_LAYER
#define SOFTMAX_LAYER

#include "Layer.h"
#include <cmath>

class LSoftmax : public Layer {

	private:

		//Weights ans Biases
		Tensor weights;
		Tensor bias;

		//Gradients
		Tensor dCdX;
		Tensor dCdW;
		Tensor dCdB;

		Tensor prev_dCdW;
		Tensor prev_dCdB;

	public:

		//Constructor
		LSoftmax( int in_size, int out_size ) {

			//Set dimensions
			this->in_dim = 1;
			this->in_rows = in_size;
			this->in_cols = 1;

			this->out_dim = 1;
			out_rows = out_size;
			out_cols = 1;

			//Redimension matrices
			in.resize(in_dim, in_rows, in_cols);
			in_w.resize(out_dim, out_rows, out_cols);
			out.resize(out_dim, out_rows, out_cols);

			weights.resize(in_dim, out_rows, in_rows);
			bias.resize(in_dim, out_rows, out_cols);

			dCdX.resize(in_dim, in_rows, in_cols);
			dCdW.resize(in_dim, out_rows, in_rows);
			dCdB.resize(in_dim, out_rows, out_cols);

			prev_dCdW.resize(in_dim, out_rows, in_rows);
			prev_dCdB.resize(in_dim, out_rows, out_cols);

		}

		//Properties
		char getType() { return 'd'; }
		void print() { weights.print();	bias.print(); return; }
		void printGrads() { dCdW.print(); dCdB.print(); return;	}

		//Functions
		//Initialise weights randomly according to a normal distribution
		void initweights( std::default_random_engine generator, double mean, double stddev ) {

			std::normal_distribution<double> distribution(mean, stddev);

			for (int i = 0; i < out_rows; i++) {

				bias(0, i, 0) = distribution(generator);

				for (int j = 0; j < in_rows; j++)
					weights(0, i, j) = distribution(generator);

			}

			return;

		}

		Tensor feedforward( Tensor in ) {

			//Copy inputs
			this->in = in.copy();
			double expsum = 0.0;

			for (int i = 0; i < out_rows; i++) {

				in_w(0, i, 0) = bias(0, i, 0);

				for (int j = 0; j < in_rows; j++) {

					in_w(0, i, 0) += weights(0, i, j) * in(0, j, 0);

				}

				//Exponentiate
				out(0, i, 0) = exp(in_w(0, i, 0));
				expsum += out(0, i, 0);

			}

			//Apply softmax normalisation
			for (int i = 0; i < out_rows; i++)
				out(0, i, 0) /= expsum;

			return out;

		}

		//Kronecker delta
		int kron_delta(int i, int j) {

			if (i == j)
				return 1;
			else
				return 0;

		}

		Tensor feedback( Tensor delta ) {

			prev_dCdW = dCdW.copy();
			prev_dCdB = dCdB.copy();

			dCdX.set(0);
			dCdW.set(0);
			dCdB.set(0);

			//Bias
			for (int i = 0; i < out_rows; i++)
				for (int j = 0; j < out_rows; j++)
					dCdB(0, i, 0) += delta(0, j, 0) * out(0, j, 0) * ( kron_delta(i, j) - out(0, i, 0) );

			//Weights
			for (int i = 0; i < out_rows; i++)
				for (int j = 0; j < in_rows; j++)
					for (int m = 0; m < out_rows; m++)
						for (int n = 0; n < out_rows; n++)
							dCdW(0, i, j) += delta(0, m, 0) * out(0, m, 0) * ( kron_delta(m, n) - out(0, n, 0) ) * in(0, j , 0) * kron_delta(n, i);

			//Deltas
			for (int i = 0; i < in_rows; i++)
				for (int j = 0; j < out_rows; j++)
					for (int k = 0; k < out_rows; k++)
						dCdX(0, i, 0) += delta(0, j, 0) * out(0, j, 0) * ( kron_delta(j, k) - out(0, k, 0) ) * weights(0, k, i);

			return dCdX;

		}

		void updateweights( float rate, float mom ) {

			for (int i = 0; i < out_rows; i++) {

				bias(0, i, 0) -= rate * dCdB(0, i, 0) + mom * (dCdB(0, i, 0) - prev_dCdB(0, i, 0));

				for (int j = 0; j < in_rows; j++) {

					weights(0, i, j) -= rate * dCdW(0, i, j) + mom * (dCdW(0, i, j) - prev_dCdW(0, i , j));

				}

			}

			return;

		}

};

#endif