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
|
/**
* ml - a neural network processor written with C
* Copyright (C) 2023 jvech
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program. If not, see <https://www.gnu.org/licenses/>.
*/
#include <stdio.h>
#include <stdint.h>
#include <stdbool.h>
#include <string.h>
#include <stdarg.h>
#include <errno.h>
#include <getopt.h>
#include <json-c/json.h>
#include "util.h"
#include "parse.h"
#include "nn.h"
#define MAX_FILE_SIZE 536870912 //1<<29; 0.5 GiB
#define ARRAY_SIZE(x, type) sizeof(x) / sizeof(type)
void load_config(struct Configs *cfg, int n_args, ...)
{
char *filepath;
va_list ap;
va_start(ap, n_args);
int i;
for (i = 0; i < n_args; i++) {
filepath = va_arg(ap, char *);
util_load_config(cfg, filepath);
if (errno == 0) {
va_end(ap);
return;
} else if (errno == ENOENT && i < n_args - 1) {
errno = 0;
} else break;
}
va_end(ap);
die("load_config('%s') Error:", filepath);
}
Layer * load_network(struct Configs cfg)
{
extern struct Activation NN_RELU;
extern struct Activation NN_SOFTPLUS;
extern struct Activation NN_SIGMOID;
extern struct Activation NN_LEAKY_RELU;
extern struct Activation NN_LINEAR;
extern struct Activation NN_TANH;
Layer *network = ecalloc(cfg.network_size, sizeof(Layer));
for (size_t i = 0; i < cfg.network_size; i++) {
if (!strcmp("relu", cfg.activations[i])) network[i].activation = NN_RELU;
else if (!strcmp("sigmoid", cfg.activations[i])) network[i].activation = NN_SIGMOID;
else if (!strcmp("softplus", cfg.activations[i])) network[i].activation = NN_SOFTPLUS;
else if (!strcmp("leaky_relu", cfg.activations[i])) network[i].activation = NN_LEAKY_RELU;
else if (!strcmp("linear", cfg.activations[i])) network[i].activation = NN_LINEAR;
else if (!strcmp("tanh", cfg.activations[i])) network[i].activation = NN_TANH;
else die("load_network() Error: Unknown '%s' activation", cfg.activations[i]);
network[i].neurons = cfg.neurons[i];
}
return network;
}
struct Cost load_loss(struct Configs cfg)
{
extern struct Cost NN_SQUARE;
if (!strcmp("square", cfg.loss)) return NN_SQUARE;
die("load_loss() Error: Unknown '%s' loss function", cfg.loss);
exit(1);
}
int main(int argc, char *argv[]) {
char default_config_path[512], *env_config_path;
struct Configs ml_configs = {
.epochs = 100,
.batch_size = 32,
.alpha = 1e-5,
.shuffle = true,
.config_filepath = "",
.network_size = 0,
.only_out = false,
.decimal_precision = -1,
.file_format = NULL,
.out_filepath = NULL,
};
// First past to check if --config option was put
util_load_cli(&ml_configs, argc, argv);
optind = 1;
// Load configs with different possible paths
sprintf(default_config_path, "%s/%s", getenv("HOME"), ".config/ml/ml.cfg");
env_config_path = (getenv("ML_CONFIG_PATH"))? getenv("ML_CONFIG_PATH"):"";
load_config(&ml_configs, 3,
ml_configs.config_filepath,
env_config_path,
default_config_path);
// re-read cli options again, to overwrite file configuration options
util_load_cli(&ml_configs, argc, argv);
argc -= optind;
argv += optind;
Layer *network = load_network(ml_configs);
Array X, y;
if (!strcmp("train", argv[0]) || !strcmp("retrain", argv[0])) {
file_read(argv[1], &X, &y,
ml_configs.input_keys, ml_configs.n_input_keys,
ml_configs.label_keys, ml_configs.n_label_keys,
true, ml_configs.file_format);
if (!strcmp("train", argv[0])) {
nn_network_init_weights(network, ml_configs.network_size, X.shape[1], true);
} else if (!strcmp("retrain", argv[0])) {
nn_network_init_weights(network, ml_configs.network_size, X.shape[1], false);
nn_network_read_weights(ml_configs.weights_filepath, network, ml_configs.network_size);
}
nn_network_train(
network, ml_configs.network_size,
X.data, X.shape,
y.data, y.shape,
load_loss(ml_configs),
ml_configs.epochs,
ml_configs.batch_size,
ml_configs.alpha,
ml_configs.shuffle);
nn_network_write_weights(ml_configs.weights_filepath, network, ml_configs.network_size);
fprintf(stderr, "weights saved on '%s'\n", ml_configs.weights_filepath);
} else if (!strcmp("predict", argv[0])) {
file_read(argv[1], &X, &y,
ml_configs.input_keys, ml_configs.n_input_keys,
ml_configs.label_keys, ml_configs.n_label_keys,
false, ml_configs.file_format);
nn_network_init_weights(network, ml_configs.network_size, X.shape[1], false);
nn_network_read_weights(ml_configs.weights_filepath, network, ml_configs.network_size);
nn_network_predict(y.data, y.shape, X.data, X.shape, network, ml_configs.network_size);
// If neither output and file_format defined use input to define the output format
if (!ml_configs.file_format && !ml_configs.out_filepath) {
ml_configs.file_format = file_format_infer(ml_configs.in_filepath);
}
file_write(ml_configs.out_filepath, X, y,
ml_configs.input_keys, ml_configs.n_input_keys,
ml_configs.label_keys, ml_configs.n_label_keys,
!ml_configs.only_out, ml_configs.file_format,
ml_configs.decimal_precision);
} else usage(1);
nn_network_free_weights(network, ml_configs.network_size);
free(network);
util_free_config(&ml_configs);
return 0;
}
|
