-
Notifications
You must be signed in to change notification settings - Fork 0
/
util.c
185 lines (159 loc) · 4.49 KB
/
util.c
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
#include <math.h>
#include <stdio.h>
#include <stdlib.h>
#include "structs.h"
#include "meth.h"
#include "util.h"
// log a matrix to console
void printMatrix(Matrix * m) {
for (int r = 0; r < m->rows; r++) {
for (int c = 0; c < m->cols; c++) {
printf("%f ", m->at[r][c]);
}
printf("\n");
}
printf("\n");
}
// print vectors side by side (good for comparison)
void printSideBySide(Matrix * a, Matrix * b) {
for (int i = 0; i < a->rows; i++) {
printf("%f --> %f\n", a->at[i][0], b->at[i][0]);
}
}
// pass an input vector through a network
Matrix * forwardPass(NeuralNetwork * n, Matrix * input) {
int l;
Matrix *weighted, *plusBias;
// copy input as first layer activation
Matrix * act = initMatrix(input->rows, input->cols);
for (int l = 0; l < act->rows; l++) {
act->at[l][0] = input->at[l][0];
}
// pass activation through network
for (l = 0; l < n->numberOfLayers - 1; l++) {
weighted = dot(n->w[l], act);
plusBias = add(weighted, n->b[l]);
// activate, apply softmax at last layer
if (l < n->numberOfLayers - 2)
act = sig(plusBias);
else
act = softMax(plusBias);
freeMatrix(weighted);
freeMatrix(plusBias);
}
return act;
}
// determine accuracy of classification over a test set
float accuracy(NeuralNetwork * n, DataSet * test) {
int numCorrect = 0, i, j, max;
// for every pair in the test set
for (i = 0; i < test->size; i++) {
// compute network output
Matrix * output = forwardPass(n, test->inputs[i]);
// determine network's most confident classification
max = 0;
for (j = 0; j < output->rows; j++) {
if (output->at[j][0] > output->at[max][0]) {
max = j;
}
}
if (test->outputs[i]->at[max][0] == 1.0)
numCorrect++;
freeMatrix(output);
}
return numCorrect / (float) test->size;
}
// free a matrix
void freeMatrix(Matrix * m) {
// free every row
for (int i = 0; i < m->rows; i++) {
free(m->at[i]);
}
free(m->at); // free pointer to float[][] array
free(m); // free pointer to struct
}
// fully free an entire network
void freeNetwork(NeuralNetwork * n) {
// free all weight matrices / bias vectors
for (int i = 0; i < n->numberOfLayers - 1; i++) {
freeMatrix(n->w[i]);
freeMatrix(n->b[i]);
}
free(n->w); // free pointer to weight matrices
free(n->b); // free pointer to bias vectors
free(n->params); // free pointer to network parameters
free(n); // free pointer to struct
}
// set entire matrix to 0
void zero(Matrix * m) {
for (int i = 0; i < m->rows; i++) {
for (int j = 0; j < m->cols; j++) {
m->at[i][j] = 0.0f;
}
}
}
// initialize a matrix to random values in a range
void randomize(Matrix * m, float min, float max) {
for (int i = 0; i < m->rows; i++) {
for (int j = 0; j < m->cols; j++) {
m->at[i][j] = ((float) rand() / RAND_MAX) * (max - min) + min;
}
}
}
// generate random integer in range
int randInt(int min, int max) {
return rand() % (max - min) + min;
}
// randomize weights and biases uniformly within ranges
void randomizeNet(NeuralNetwork * n, float wMin, float wMax, float bMin, float bMax) {
for (int l = 0; l < n->numberOfLayers - 1; l++) {
randomize(n->w[l], wMin, wMax);
randomize(n->b[l], bMin, bMax);
}
}
// generate a gaussian variable with given mean and standard deviation
float gaussian(double mean, double std_dev) {
double u1 = ((double) rand() / RAND_MAX);
double u2 = ((double) rand() / RAND_MAX);
return u1 == 0.0 ? 0.0f : (float) (mean + (std_dev * sqrt(-2 * log(u1)) * cos(2 * M_PI * u2)));
}
// randomize weights & biases with Gaussian distributions
void gaussianRandomizeNet(NeuralNetwork * n) {
double std_dev, u1, u2;
int l, j, k;
// for each layer
for (l = 0; l < n->numberOfLayers - 1; l++) {
std_dev = 1.0 / sqrt(n->params[l]);
// for each weight / bias
for (j = 0; j < n->w[l]->rows; j++) {
for (k = 0; k < n->w[l]->cols; k++) {
n->w[l]->at[j][k] = gaussian(0.0, std_dev);
}
n->b[l]->at[j][0] = gaussian(0.0, 1.0);
}
}
}
// copy network parameters
int * paramCopy(int * params, int size) {
int * copy = malloc(size * sizeof(int));
for (int i = 0; i < size; i++) {
copy[i] = params[i];
}
return copy;
}
// swap a training pair within a dataset
void swapPair(DataSet * d, int p1, int p2) {
Matrix * inp = d->inputs[p1];
Matrix * out = d->outputs[p1];
d->inputs[p1] = d->inputs[p2];
d->outputs[p1] = d->outputs[p2];
d->inputs[p2] = inp;
d->outputs[p2] = out;
}
// randomize the order of pairs in a dataset
void shuffle(DataSet * d) {
// swap each pair with a randomly chosen pair
for (int n = 0; n < d->size; n++) {
swapPair(d, n, randInt(0, d->size));
}
}