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
|
/*******************************************************
* Copyright (c) 2014, ArrayFire
* All rights reserved.
*
* This file is distributed under 3-clause BSD license.
* The complete license agreement can be obtained at:
* http://arrayfire.com/licenses/BSD-3-Clause
********************************************************/
#include <sstream>
#include <algorithm>
#include <utility>
#include "../common/idxio.h"
bool compare(const std::pair<float, int> l,
const std::pair<float, int> r)
{
return l.first >= r.first;
}
typedef std::pair<float, int> sort_type;
template<bool expand_labels>
std::string classify(af::array arr, int k)
{
std::stringstream ss;
if (expand_labels) {
af::array vec = arr(af::span, k).as(f32);
float *h_vec = vec.host<float>();
std::vector<sort_type> data;
for (int i = 0; i < (int)vec.elements(); i++)
data.push_back(std::make_pair(h_vec[i], i));
std::stable_sort(data.begin(), data.end(), compare);
ss << data[0].second;
} else {
ss << (int)(arr(k).as(f32).scalar<float>());
}
return ss.str();
}
template<bool expand_labels>
static void setup_mnist(int *num_classes, int *num_train, int *num_test,
af::array &train_images, af::array &test_images,
af::array &train_labels, af::array &test_labels, float frac)
{
std::vector<dim_t> idims;
std::vector<float > idata;
read_idx(idims, idata, ASSETS_DIR"/examples/data/mnist/images-subset");
std::vector<dim_t> ldims;
std::vector<unsigned> ldata;
read_idx(ldims, ldata, ASSETS_DIR"/examples/data/mnist/labels-subset");
std::reverse(idims.begin(), idims.end());
unsigned numdims = idims.size();
af::array images = af::array(af::dim4(numdims, &idims[0]), &idata[0]);
af::array R = af::randu(10000, 1);
af::array cond = R < std::min(frac, 0.8f);
af::array train_indices = where( cond);
af::array test_indices = where(!cond);
train_images = lookup(images, train_indices, 2) / 255;
test_images = lookup(images, test_indices , 2) / 255;
*num_classes = 10;
*num_train = train_images.dims(2);
*num_test = test_images.dims(2);
if (expand_labels) {
train_labels = af::constant(0, *num_classes, *num_train);
test_labels = af::constant(0, *num_classes, *num_test );
unsigned *h_train_idx = train_indices.host<unsigned>();
unsigned *h_test_idx = test_indices.host<unsigned>();
for (int ii = 0; ii < *num_train; ii++) {
train_labels(ldata[h_train_idx[ii]], ii) = 1;
}
for (int ii = 0; ii < *num_test; ii++) {
test_labels(ldata[ h_test_idx[ii]], ii) = 1;
}
delete[] h_train_idx;
delete[] h_test_idx;
} else {
af::array labels = af::array(ldims[0], &ldata[0]);
train_labels = labels(train_indices);
test_labels = labels( test_indices);
}
return;
}
#if 0
static af::array randidx(int num, int total)
{
af::array locs;
do {
locs = af::where(af::randu(total, 1) < float(num * 2) / total);
} while (locs.elements() < num);
return locs(af::seq(num));
}
#endif
template<bool expand_labels>
static void display_results(const af::array &test_images,
const af::array &test_output,
const af::array &test_actual,
int num_display)
{
#if 0
af::array locs = randidx(num_display, test_images.dims(2));
af::array disp_in = test_images(af::span, af::span, locs);
af::array disp_out = expand_labels ? test_output(af::span, locs) : test_output(locs);
for (int i = 0; i < 5; i++) {
int imgs_per_iter = num_display / 5;
for (int j = 0; j < imgs_per_iter; j++) {
int k = i * imgs_per_iter + j;
af::fig("sub", 2, imgs_per_iter / 2, j+1);
af::image(disp_in(af::span, af::span, k).T());
std::string pred_name = std::string("Predicted: ");
pred_name = pred_name + classify<expand_labels>(disp_out, k);
af::fig("title", pred_name.c_str());
}
printf("Press any key to see next set");
getchar();
}
#else
using namespace af;
for(int i = 0; i < num_display; i++) {
std::cout << "Predicted: " << classify<expand_labels>(test_output, i) << std::endl;
std::cout << "Actual: " << classify<expand_labels>(test_actual, i) << std::endl;
unsigned char* img = (test_images(span, span, i) > 0.1f).as(u8).host<unsigned char>();
for(int j = 0; j < 28; j++) {
for(int k = 0; k < 28; k++) {
std::cout << (img[j*28+k] ? "\u2588" : " ") << " ";
}
std::cout << std::endl;
}
delete[] img;
getchar();
}
#endif
}
|
