-
Notifications
You must be signed in to change notification settings - Fork 2
/
ks_cpp.c
184 lines (177 loc) · 6.75 KB
/
ks_cpp.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
#include <stdio.h>
#include <malloc.h>
#include <assert.h>
#include <memory.h>
#include <math.h>
// C bool
typedef enum {
true=1, false=0
} bool;
inline void update_min(double* p1, double v2) {
if (v2 < *p1) *p1 = v2;
}
// https://stackoverflow.com/questions/28258590/using-openmp-to-get-the-index-of-minimum-element-parallelly
struct Compare { double val; size_t index; };
#pragma omp declare reduction(maximum : struct Compare : omp_out = omp_in.val > omp_out.val ? omp_in : omp_out)
void kennard_stone(double* cdist, size_t* seed, size_t* result, double* v_dist, size_t n_sample, size_t n_seed, size_t n_result) {
// 00. Assertions and Result Vector Initialization
struct Compare sup;
if (n_seed == 2) v_dist[0] = cdist[seed[0] * n_sample + seed[1]];
if (n_seed == 0) {
size_t n_sample_2 = n_sample * n_sample;
sup.val = -1.;
sup.index = -1;
#pragma omp parallel for reduction(maximum:sup)
for (size_t i = 0; i < n_sample_2; ++i) {
if (cdist[i] > sup.val) {
sup.val = cdist[i];
sup.index = i;
}
}
if (sup.index == 0) { // Threading Safety Check
sup.val = -1.;
sup.index = -1;
for (size_t i = 0; i < n_sample_2; ++i) {
if (cdist[i] > sup.val) {
sup.val = cdist[i];
sup.index = i;
}
}
}
seed[0] = sup.index / n_sample;
seed[1] = sup.index % n_sample;
n_seed = 2;
v_dist[0] = sup.val;
}
n_result = n_result == 0 ? n_sample : n_result;
assert(n_result <= n_sample);
assert(n_seed <= n_sample);
memcpy(result, seed, n_seed * sizeof(size_t));
memset(result + n_seed, 0, (n_result - n_seed) * sizeof(size_t));
// 01. Scratch Area Initialization
bool* selected = (bool*)malloc(n_sample * sizeof(bool));
memset(selected, false, n_sample * sizeof(bool));
for (size_t i = 0; i < n_seed; ++i)
selected[result[i]] = true;
// 02. Minimum Out-of-Group Initialization
double* min_vals = (double*)malloc(n_sample * sizeof(double));
memcpy(min_vals, cdist + n_sample * result[0], n_sample * sizeof(double));
for (size_t n = 1; n < n_seed; ++n) {
size_t idx_starting = result[n] * n_sample;
#pragma omp parallel for
for (size_t i = 0; i < n_sample; ++i) {
if (selected[i]) continue;
update_min(&min_vals[i], cdist[idx_starting + i]);
}
}
// 03. Main Algorithm
for (size_t n = n_seed; n < n_result; ++n) {
// Find sup of the minimum
sup.val = -1.;
sup.index = 0;
#pragma omp parallel for reduction(maximum:sup)
for (size_t i = 0; i < n_sample; ++i) {
if (selected[i]) continue;
if (min_vals[i] > sup.val) {
sup.index = i;
sup.val = min_vals[i];
}
}
if (sup.index == 0) { // Threading Safety Check
sup.val = -1.;
sup.index = n_sample + 1;
for (size_t i = 0; i < n_sample; ++i) {
if (selected[i]) continue;
if (min_vals[i] > sup.val) {
sup.index = i;
sup.val = min_vals[i];
}
}
}
v_dist[n - 1] = sup.val;
selected[sup.index] = true;
result[n] = sup.index;
size_t idx_starting = sup.index * n_sample;
#pragma omp parallel for
for (size_t i = 0; i < n_sample; ++i) {
if (selected[i]) continue;
update_min(&min_vals[i], cdist[idx_starting + i]);
}
}
free(selected);
free(min_vals);
}
double euclid_distance_vector(double* x1, double* x2, size_t n_feature) {
double res = 0.;
do {
res += (*x1 - *x2) * (*x1 - *x2);
++x1, ++x2;
} while (--n_feature);
return sqrt(res);
}
void kennard_stone_mem(double* X, size_t* seed, size_t* result, double* v_dist, size_t n_sample, size_t n_feature, size_t n_seed, size_t n_result) {
// 00. Assertions and Result Vector Initialization
struct Compare sup;
if (n_seed == 2) v_dist[0] = euclid_distance_vector(X + n_feature * seed[0], X + n_feature * seed[1], n_feature);
assert(n_seed != 0); // Seed should be supplied from outer program.
assert(n_result <= n_sample);
assert(n_seed <= n_sample);
memcpy(result, seed, n_seed * sizeof(size_t));
memset(result + n_seed, 0, (n_result - n_seed) * sizeof(size_t));
// 01. Scratch Area Initialization
bool* selected = (bool*)malloc(n_sample * sizeof(bool));
memset(selected, false, n_sample * sizeof(bool));
for (size_t i = 0; i < n_seed; ++i)
selected[result[i]] = true;
// 02. Minimum Out-of-Group Initialization
double* min_vals = (double*)malloc(n_sample * sizeof(double));
#pragma omp parallel for
for (size_t i = 0; i < n_sample; ++i) {
if (selected[i]) continue;
min_vals[i] = euclid_distance_vector(X + n_feature * result[0], X + n_feature * i, n_feature);
}
for (size_t n = 1; n < n_seed; ++n) {
double* p_starting = X + result[n] * n_feature;
#pragma omp parallel for
for (size_t i = 0; i < n_sample; ++i) {
if (selected[i]) continue;
update_min(&min_vals[i], euclid_distance_vector(p_starting, X + n_feature * i, n_feature));
}
}
// 03. Main Algorithm
for (size_t n = n_seed; n < n_result; ++n) {
// Find sup of the minimum
sup.val = -1.;
sup.index = 0;
#pragma omp parallel for reduction(maximum:sup)
for (size_t i = 0; i < n_sample; ++i) {
if (selected[i]) continue;
if (min_vals[i] > sup.val) {
sup.index = i;
sup.val = min_vals[i];
}
}
if (sup.index == 0) { // Threading Safety Check
sup.val = -1.;
sup.index = 0;
for (size_t i = 0; i < n_sample; ++i) {
if (selected[i]) continue;
if (min_vals[i] > sup.val) {
sup.index = i;
sup.val = min_vals[i];
}
}
}
v_dist[n - 1] = sup.val;
selected[sup.index] = true;
result[n] = sup.index;
double* p_starting = X + sup.index * n_feature;
#pragma omp parallel for
for (size_t i = 0; i < n_sample; ++i) {
if (selected[i]) continue;
update_min(&min_vals[i], euclid_distance_vector(p_starting, X + n_feature * i, n_feature));
}
}
free(selected);
free(min_vals);
}