-
Notifications
You must be signed in to change notification settings - Fork 27
/
dbscan.cpp
421 lines (344 loc) · 11.5 KB
/
dbscan.cpp
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
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
#include <iostream>
#include <boost/numeric/ublas/matrix.hpp>
#include <boost/numeric/ublas/matrix_proxy.hpp>
#include <boost/numeric/ublas/io.hpp>
#include <boost/algorithm/minmax.hpp>
#include <boost/tokenizer.hpp>
#include <boost/lexical_cast.hpp>
#include <vector>
#include <map>
#include <algorithm> // std::unique, std::distance
#include <omp.h>
#include "dbscan.h"
namespace clustering
{
void ClusterData::init(size_t elements_num, size_t features_num)
{
m_elements_num = elements_num;
m_features_num = features_num;
}
ClusterData::ClusterData(size_t elements_num, size_t features_num)
: m_elements_num( elements_num )
, m_features_num( features_num )
, m( elements_num, features_num )
{
reset();
}
void ClusterData::reset()
{
m_ids.clear();
m_cl.clear();
m.clear();
}
size_t ClusterData::getElementsNum()
{
return m_elements_num;
}
size_t ClusterData::getFeaturesNum()
{
return m_features_num;
}
ClusterData::~ClusterData()
{
}
//------------------------------------------------------------------
void DBSCAN::init(double eps, size_t min_elems, int num_threads)
{
m_eps = eps;
m_min_elems = min_elems;
m_num_threads = num_threads;
}
DBSCAN::DBSCAN(double eps, size_t min_elems, int num_threads)
: m_eps( eps )
, m_min_elems( min_elems )
, m_num_threads( num_threads )
, m_dmin(0.0)
, m_dmax(0.0)
{
reset();
}
DBSCAN::~DBSCAN()
{
}
ClusterData::Data DBSCAN::gen_cluster_data( size_t features_num, size_t elements_num )
{
ClusterData::Data cl_d( elements_num, features_num );
ulong64 random;
//elements -> by row
for (size_t i = 0; i < elements_num; ++i)
{
//features -> by column
for (size_t j = 0; j < features_num; ++j)
{
random =
(((ulong64) rand() << 0) & 0x000000000000FFFFull) |
(((ulong64) rand() << 16) & 0x00000000FFFF0000ull) |
(((ulong64) rand() << 32) & 0x0000FFFF00000000ull) |
(((ulong64) rand() << 48) & 0xFFFF000000000000ull);
cl_d(i, j) = random;
}
}
return cl_d;
}
ClusterData DBSCAN::read_cluster_data(std::istream & is, size_t features_num, size_t elements_num )
{
/**
* Read data from stream.
*
* N = features_num
* M = elements_num
*
* Expected format is:
*
* ClassElement1 stringIdElement1 ulong64feature1 ulong64feature2 ... ulong64featureN
* ClassElement2 stringidelement2 ulong64feature1 ulong64feature2 ... ulong64featureN
* ClassElement3 stringidelement3 ulong64feature1 ulong64feature2 ... ulong64featureN
* .
* .
* .
* ClassElementM stringidelement3 ulong64feature1 ulong64feature2 ... ulong64featureN
*/
ClusterData cl_d( elements_num, features_num );
std::vector<std::string> vec;
std::string line;
typedef boost::tokenizer<boost::escaped_list_separator<char> > Tokenizer;
Tokenizer::iterator it;
for (size_t i = 0; std::getline(is, line) && i < elements_num; ++i)
{
size_t j = 0;
Tokenizer tok(line);
it = tok.begin();
cl_d.m_cl.push_back(boost::lexical_cast<ClusterData::Klass>(*it));
++it;
cl_d.m_ids.push_back(boost::lexical_cast<ClusterData::Id>(*it));
++it;
for (; j < features_num && it != tok.end(); ++j)
{
cl_d.m(i, j) = boost::lexical_cast<ulong64>(*it);
}
}
return cl_d;
}
DBSCAN::FeaturesWeights DBSCAN::std_weights( size_t s )
{
// num cols
DBSCAN::FeaturesWeights ws( s );
for (size_t i = 0; i < s; ++i)
{
ws(i) = 1.0;
}
return ws;
}
void DBSCAN::reset()
{
m_labels.clear();
m_cluster_index.clear();
}
void DBSCAN::prepare_labels( size_t s )
{
m_labels.resize(s);
for( auto & l : m_labels)
{
l = -1; //noise by default
}
}
ublas::vector<DBSCAN::Distance> DBSCAN::distance(
ublas::matrix_row<ClusterData::Data> & rowA,
ublas::matrix_row<ClusterData::Data> & rowB)
{
ublas::vector<DBSCAN::Distance> d(rowA.size());
for (size_t i = 0; i < rowA.size(); ++i)
{
//Hamming
d(i) = (DBSCAN::Distance)ph_hamming_distance((ulong64)rowA(i), (ulong64)rowB(i));
//Euclidean
//d(i) = (DBSCAN::Distance)(rowA(i)-rowB(i));
}
return d;
}
const DBSCAN::DistanceMatrix DBSCAN::calc_dist_matrix(
const ClusterData::Data & C, const DBSCAN::FeaturesWeights & W, bool normalize)
{
ClusterData::Data cl_d = C;
// by-column normalization
if (normalize)
{
omp_set_dynamic(0);
omp_set_num_threads( m_num_threads );
#pragma omp parallel for
for (size_t i = 0; i < cl_d.size2(); ++i)
{
ublas::matrix_column<ClusterData::Data> col(cl_d, i);
const auto r = minmax_element( col.begin(), col.end() );
double data_min = *r.first;
double data_range = *r.second - *r.first;
if (data_range == 0.0) {
data_range = 1.0;
}
const double scale = 1/data_range;
const double min = -1.0 * data_min * scale;
col *= scale;
col.plus_assign(
ublas::scalar_vector< typename ublas::matrix_column<ClusterData::Data>::value_type >(col.size(), min) );
}
}
// rows x rows
DBSCAN::DistanceMatrix d_m( cl_d.size1(), cl_d.size1() );
ublas::vector<double> d_max( cl_d.size1() );
ublas::vector<double> d_min( cl_d.size1() );
omp_set_dynamic(0);
omp_set_num_threads( m_num_threads );
#pragma omp parallel for
for (size_t i = 0; i < cl_d.size1(); ++i)
{
for (size_t j = i; j < cl_d.size1(); ++j)
{
d_m(i, j) = 0.0;
if (i != j)
{
ublas::matrix_row<ClusterData::Data> U (cl_d, i);
ublas::matrix_row<ClusterData::Data> V (cl_d, j);
//distance computation between two elements (i.e., rows)
int k = 0;
for (const auto e : distance(U, V))
{
d_m(i, j) += fabs(e)*W[k++];
}
d_m(j, i) = d_m(i, j);
}
}
const auto cur_row = ublas::matrix_row<DBSCAN::DistanceMatrix>(d_m, i);
const auto mm = minmax_element( cur_row.begin(), cur_row.end() );
d_max(i) = *mm.second;
d_min(i) = *mm.first;
}
if (normalize)
{
m_dmin = *(min_element( d_min.begin(), d_min.end() ));
m_dmax = *(max_element( d_max.begin(), d_max.end() ));
m_eps = (m_dmax - m_dmin) * m_eps + m_dmin;
}
return d_m;
}
DBSCAN::Neighbors DBSCAN::find_neighbors(const DBSCAN::DistanceMatrix & D, uint32_t pid)
{
Neighbors ne;
DBSCAN::Distance d;
for (uint32_t j = 0; j < D.size1(); ++j)
{
d = D(pid, j);
if ( d <= m_eps )
{
ne.push_back(j);
}
}
return ne;
}
void DBSCAN::dbscan( const DBSCAN::DistanceMatrix & dm )
{
std::vector<uint8_t> visited( dm.size1() );
uint32_t cluster_id = 0;
//for each element
for (uint32_t pid = 0; pid < dm.size1(); ++pid)
{
//only if not visited yet
if ( !visited[pid] )
{
visited[pid] = 1;
//look for neighbors
Neighbors ne = find_neighbors(dm, pid );
//if enough neighbors
if (ne.size() >= m_min_elems)
{
//assign that element to the current cluster
m_labels[pid] = cluster_id;
//for each element in the neighborhood
for (uint32_t i = 0; i < ne.size(); ++i)
{
uint32_t nPid = ne[i];
//if not visited yet
if ( !visited[nPid] )
{
//mark it as visited
visited[nPid] = 1;
//find its neighbors
Neighbors ne1 = find_neighbors(dm, nPid);
//if enough neighbors
if ( ne1.size() >= m_min_elems )
{
//expand neighbors
for (const auto & n1 : ne1)
{
ne.push_back(n1);
}
}
}
//if the last element in the neighborhood was noise
// mark it as non-noise anymore and include it to the
// current cluster
if ( m_labels[nPid] == -1 )
{
m_labels[nPid] = cluster_id;
}
}
++cluster_id; //create new cluster
}
}
}
}
void DBSCAN::fit( const ClusterData::Data & C, bool normalize )
{
const DBSCAN::FeaturesWeights W = DBSCAN::std_weights( C.size2() );
wfit( C, W, normalize );
}
void DBSCAN::fit_precomputed( const DBSCAN::DistanceMatrix & D )
{
prepare_labels( D.size1() );
dbscan( D );
}
void DBSCAN::wfit( const ClusterData::Data & C, const DBSCAN::FeaturesWeights & W, bool normalize )
{
prepare_labels( C.size1() );
const DBSCAN::DistanceMatrix D = calc_dist_matrix( C, W, normalize );
dbscan( D );
}
const DBSCAN::Labels & DBSCAN::get_labels() const
{
return m_labels;
}
void DBSCAN::prepare_cluster_index()
{
if (m_cluster_index.size() == 0) {
//key->value (element_id -> label)
for (size_t element = 0; element < m_labels.size(); ++element)
{
m_cluster_index[m_labels[element]].insert(element);
}
}
}
const DBSCAN::ClusterIndex & DBSCAN::get_cluster_index() const
{
return m_cluster_index;
}
std::ostream& operator<<(std::ostream& o, DBSCAN & d)
{
o << "[";
for ( const auto & l : d.get_labels() )
{
o << " " << l;
}
o << "] " << std::endl;
return o;
}
std::ostream& operator<<(std::ostream& o, ClusterData & cl_d)
{
for (size_t i = 0; i < cl_d.getElementsNum(); ++i)
{
o << cl_d.m_cl[i] << " " << cl_d.m_ids[i];
for (size_t j = 0; j < cl_d.getFeaturesNum(); ++j)
o << " " << cl_d.m(i, j);
o << std::endl;
}
return o;
}
}