graph_generation.cc 18.1 KB
Newer Older
Tiago Peixoto's avatar
Tiago Peixoto committed
1
2
// graph-tool -- a general graph modification and manipulation thingy
//
Tiago Peixoto's avatar
Tiago Peixoto committed
3
// Copyright (C) 2007  Tiago de Paula Peixoto <tiago@forked.de>
Tiago Peixoto's avatar
Tiago Peixoto committed
4
5
6
//
// This program is free software; you can redistribute it and/or
// modify it under the terms of the GNU General Public License
Tiago Peixoto's avatar
Tiago Peixoto committed
7
// as published by the Free Software Foundation; either version 3
Tiago Peixoto's avatar
Tiago Peixoto committed
8
9
10
11
12
13
14
15
// 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
16
17
// along with this program. If not, see <http://www.gnu.org/licenses/>.

Tiago Peixoto's avatar
Tiago Peixoto committed
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

#include <algorithm>
#include <tr1/unordered_set>
#include <boost/lambda/lambda.hpp>
#include <boost/lambda/bind.hpp>
#include <boost/random.hpp>
#include <iomanip>
#include <map>

#include "graph.hh"
#include "histogram.hh"

using namespace std;
using namespace boost;
using namespace boost::lambda;
using namespace graph_tool;

typedef boost::mt19937 rng_t;

// this will sample a (j,k) pair from a pjk distribution given a ceil function
// and its inverse

template <class Distribution, class Ceil, class InvCeil>
struct sample_from_distribution
{
43
44
45
46
    sample_from_distribution(Distribution &dist, Ceil& ceil, InvCeil &inv_ceil, 
                             double bound, rng_t& rng)
        : _dist(dist), _ceil(ceil), _inv_ceil(inv_ceil), _bound(bound), 
          _rng(rng), _uniform_p(0.0, 1.0) {}
Tiago Peixoto's avatar
Tiago Peixoto committed
47
48
49
    
    pair<size_t, size_t> operator()()
    {
50
51
52
53
54
55
56
57
58
59
        // sample j,k from ceil
        size_t j,k;
        double u;
        do
        {
            tie(j,k) = _inv_ceil(_uniform_p(_rng), _uniform_p(_rng));
            u = _uniform_p(_rng);
        }
        while (u > _dist(j,k)/(_bound*_ceil(j,k)));
        return make_pair(j,k);
Tiago Peixoto's avatar
Tiago Peixoto committed
60
61
62
63
64
65
66
67
68
69
    }

    Distribution& _dist;
    Ceil& _ceil;
    InvCeil& _inv_ceil;
    double _bound;
    rng_t &_rng;
    boost::uniform_real<double> _uniform_p;
};

70
// desired vertex type, with desired j,k values and the index in the real graph
Tiago Peixoto's avatar
Tiago Peixoto committed
71

72
struct dvertex_t 
Tiago Peixoto's avatar
Tiago Peixoto committed
73
{
74
75
76
77
    dvertex_t() {}
    dvertex_t(size_t in, size_t out): in_degree(in), out_degree(out) {}
    dvertex_t(const pair<size_t,size_t>& deg): in_degree(deg.first), 
                                              out_degree(deg.second) {}
Tiago Peixoto's avatar
Tiago Peixoto committed
78
    size_t index, in_degree, out_degree;
79
    bool operator==(const dvertex_t& other) const {return other.index == index;}
Tiago Peixoto's avatar
Tiago Peixoto committed
80
81
};

82
inline std::size_t hash_value(const dvertex_t& v)
Tiago Peixoto's avatar
Tiago Peixoto committed
83
84
85
86
87
88
{
    size_t h = hash_value(v.in_degree);
    hash_combine(h, v.out_degree);
    return h;
}

89
inline size_t dist(const dvertex_t& a, const dvertex_t& b)
Tiago Peixoto's avatar
Tiago Peixoto committed
90
91
{
    return int(a.in_degree-b.in_degree)*int(a.in_degree-b.in_degree) + 
92
        int(a.out_degree-b.out_degree)*int(a.out_degree-b.out_degree);
Tiago Peixoto's avatar
Tiago Peixoto committed
93
94
95
96
97
98
}

struct total_deg_comp
{
    bool operator()(const pair<size_t,size_t>& a, const pair<size_t,size_t>& b)
    {
99
        return a.first + a.second < b.first + b.second;
Tiago Peixoto's avatar
Tiago Peixoto committed
100
101
102
103
104
105
    }
};

// this structure will keep the existing (j,k) pairs in the graph in a matrix,
// so that the nearest (j,k) to a given target can be found easily.

106
class degree_matrix_t
Tiago Peixoto's avatar
Tiago Peixoto committed
107
{
108
public:    
109
110
    degree_matrix_t(size_t N, size_t minj, size_t mink, size_t maxj, 
                    size_t maxk)
Tiago Peixoto's avatar
Tiago Peixoto committed
111
    {
112
113
114
115
116
117
118
119
120
        _L = max(size_t(pow(2,ceil(log2(sqrt(N))))),size_t(2));
        _minj = minj;
        _mink = mink;
        _maxj = max(maxj,_L);
        _maxk = max(maxk,_L);
        _bins.resize(_L, vector<vector<pair<size_t,size_t> > >(_L));
        _high_bins.resize(size_t(log2(_L)));
        for(size_t i = 0; i < _high_bins.size(); ++i)
            _high_bins[i].resize(_L/(1<<(i+1)), vector<size_t>(_L/(1<<(i+1))));
Tiago Peixoto's avatar
Tiago Peixoto committed
121
    }
122

Tiago Peixoto's avatar
Tiago Peixoto committed
123
124
    void insert(const pair<size_t, size_t>& v)
    {
125
126
127
128
129
130
131
132
133
        size_t j_bin, k_bin;
        tie(j_bin, k_bin) = get_bin(v.first, v.second, 0);
        _bins[j_bin][k_bin].push_back(v);
        for (size_t i = 0; i < _high_bins.size(); ++i)
        {
            size_t hj,hk;
            tie(hj,hk) = get_bin(j_bin,k_bin, i+1);
            _high_bins[i][hj][hk]++;
        }
Tiago Peixoto's avatar
Tiago Peixoto committed
134
    }
135
    
Tiago Peixoto's avatar
Tiago Peixoto committed
136
137
    void erase(const pair<size_t,size_t>& v)
    {
138
139
140
141
142
143
        size_t j_bin, k_bin;
        tie(j_bin, k_bin) = get_bin(v.first, v.second, 0);
        for(size_t i = 0; i < _bins[j_bin][k_bin].size(); ++i)
        {
            if (_bins[j_bin][k_bin][i] == v)
            {
144
145
                _bins[j_bin][k_bin].erase(_bins[j_bin][k_bin].begin()+i);
                break;
146
147
148
149
150
151
152
153
154
155
            }
        }
        
        for (size_t i = 0; i < _high_bins.size(); ++i)
        {
            size_t hj,hk;
            tie(hj,hk) = get_bin(j_bin,k_bin, i+1);
            _high_bins[i][hj][hk]--;
        }
        
Tiago Peixoto's avatar
Tiago Peixoto committed
156
157
    }

158
    pair<size_t,size_t> find_closest(size_t j, size_t k, rng_t& rng)
Tiago Peixoto's avatar
Tiago Peixoto committed
159
    {
160
161
162
163
164
165
166
167
168
169
170
        vector<pair<size_t,size_t> > candidates;

        size_t level;

        // find the appropriate level on which to operate
        for (level = _high_bins.size(); level <= 0; --level)
        {
            size_t hj, hk;
            tie(hj,hk) = get_bin(j,k,level);
            if (get_bin_count(hj,hk,level) == 0)
            {
171
172
173
                if (level < _high_bins.size())
                    level++;
                break;
174
175
176
177
178
179
            }
        }

        size_t j_bin, k_bin;
        tie(j_bin, k_bin) = get_bin(j, k, level);

180
181
182
183
184
        for (size_t hj = ((j_bin>0)?j_bin-1:j_bin); 
             hj < j_bin + 1 && hj <= get_bin(_maxj, _maxk, level).first; ++hj)
            for (size_t hk = ((k_bin>0)?k_bin-1:k_bin); 
                 hk < k_bin + 1 && hk <= get_bin(_maxj, _maxk, level).second; 
                 ++hk)
185
                search_bin(hj,hk,j,k,level,candidates);
186
187
188
        
        uniform_int<size_t> sample(0, candidates.size() - 1);
        return candidates[sample(rng)];
Tiago Peixoto's avatar
Tiago Peixoto committed
189
190
    }

191
private:    
192
    pair<size_t,size_t> get_bin(size_t j, size_t k, size_t level) 
Tiago Peixoto's avatar
Tiago Peixoto committed
193
    {
194
        if (level == 0)
195
196
            return make_pair(((j-_minj)*(_L-1))/_maxj, 
                             ((k-_mink)*(_L-1))/_maxk);
197

198
199
200
201
        pair<size_t, size_t> bin = get_bin(j,k,0);
        bin.first /=  1 << level;
        bin.second /= 1 << level;
        return bin;
202
203
204
205
    }

    size_t get_bin_count(size_t bin_j, size_t bin_k, size_t level)
    {
206
207
208
209
        if (level == 0)
            return _bins[bin_j][bin_k].size();
        else
            return _high_bins[level-1][bin_j][bin_k];
Tiago Peixoto's avatar
Tiago Peixoto committed
210
    }
211

212
213
    void search_bin(size_t hj, size_t hk, size_t j, size_t k, size_t level, 
                    vector<pair<size_t,size_t> >& candidates)
Tiago Peixoto's avatar
Tiago Peixoto committed
214
    {
215
216
217
218
        size_t w = 1 << level;
        for (size_t j_bin = hj*w; j_bin < (hj+1)*w; ++j_bin)
            for (size_t k_bin = hk*w; k_bin < (hk+1)*w; ++k_bin)
            {
219
220
221
222
223
224
225
226
                for (size_t i = 0; i < _bins[j_bin][k_bin].size(); ++i)
                {
                    pair<size_t, size_t>& v = _bins[j_bin][k_bin][i];
                    if (candidates.empty())
                    {
                        candidates.push_back(v);
                        continue;
                    }
227
228
                    if (dist(dvertex_t(v), dvertex_t(j,k)) < 
                        dist(dvertex_t(candidates.front()),dvertex_t(j,k)))
229
230
231
232
                    {
                        candidates.clear();
                        candidates.push_back(v);
                    }
233
234
                    else if (dist(dvertex_t(v), dvertex_t(j,k)) == 
                             dist(dvertex_t(candidates.front()),dvertex_t(j,k)))
235
236
237
238
                    {
                        candidates.push_back(v);
                    }
                }
239
            }
Tiago Peixoto's avatar
Tiago Peixoto committed
240
241
242
    }

    size_t _L;
243
244
245
246
247
248
    vector<vector<vector<pair<size_t,size_t> > > > _bins;
    vector<vector<vector<size_t> > > _high_bins;
    size_t _minj;
    size_t _mink;
    size_t _maxj;
    size_t _maxk;
Tiago Peixoto's avatar
Tiago Peixoto committed
249
250
251
};

// generates a directed graph with given pjk and degree correlation
252
253
254
255
256
257

void GraphInterface::GenerateCorrelatedConfigurationalModel
    (size_t N, pjk_t pjk, pjk_t ceil_pjk, inv_ceil_t inv_ceil_pjk, 
     double ceil_pjk_bound, corr_t corr, corr_t ceil_corr, 
     inv_corr_t inv_ceil_corr, double ceil_corr_bound, 
     bool undirected_corr, size_t seed, bool verbose)
Tiago Peixoto's avatar
Tiago Peixoto committed
258
259
260
{
    _mg.clear();
    _properties = dynamic_properties();
261
    rng_t rng(static_cast<rng_t::result_type>(seed));
Tiago Peixoto's avatar
Tiago Peixoto committed
262
263
264

    // sample the N (j,k) pairs

265
266
267
    sample_from_distribution<pjk_t, pjk_t, inv_ceil_t> 
        pjk_sample(pjk, ceil_pjk, inv_ceil_pjk, ceil_pjk_bound, rng);
    vector<dvertex_t> vertices(N);
268
    size_t sum_j=0, sum_k=0, min_j=0, min_k=0, max_j=0, max_k=0;
Tiago Peixoto's avatar
Tiago Peixoto committed
269
270
    if (verbose)
    {
271
        cout << "adding vertices: " << flush;
Tiago Peixoto's avatar
Tiago Peixoto committed
272
273
274
    }
    for(size_t i = 0; i < N; ++i)
    {
275
276
277
278
        if (verbose)
        {
            static stringstream str;
            for (size_t j = 0; j < str.str().length(); ++j)
279
                cout << "\b";
280
281
282
283
            str.str("");
            str << i+1 << " of " << N << " (" << (i+1)*100/N << "%)";
            cout << str.str() << flush;
        }
284
        dvertex_t& v = vertices[i];
285
286
287
288
289
290
291
292
        v.index = _vertex_index[add_vertex(_mg)];
        tie(v.in_degree, v.out_degree) = pjk_sample();
        sum_j += v.in_degree;
        sum_k += v.out_degree;
        min_j = min(v.in_degree,min_j);
        min_k = min(v.out_degree,min_k);
        max_j = max(v.in_degree,max_j);
        max_k = max(v.out_degree,max_k); 
Tiago Peixoto's avatar
Tiago Peixoto committed
293
294
295
    }

    if (verbose)
296
        cout << "\nfixing average degrees: " << flush;
Tiago Peixoto's avatar
Tiago Peixoto committed
297
298
299
300
301

    // <j> and <k> must be the same. Resample random pairs until this holds.
    uniform_int<size_t> vertex_sample(0, N-1);
    while (sum_j != sum_k)
    {
302
        dvertex_t& v = vertices[vertex_sample(rng)];
303
304
305
306
        sum_j -= v.in_degree;
        sum_k -= v.out_degree;
        tie(v.in_degree, v.out_degree) = pjk_sample();
        sum_j += v.in_degree;
Tiago Peixoto's avatar
Tiago Peixoto committed
307
        sum_k +=  v.out_degree;
308
309
310
311
312
313
        max_j = max(v.in_degree,max_j);
        max_k = max(v.out_degree,max_k);
        if (verbose)
        {
            static stringstream str;
            for (size_t j = 0; j < str.str().length(); ++j)
314
                cout << "\b";
315
            for (size_t j = 0; j < str.str().length(); ++j)
316
                cout << " ";
317
            for (size_t j = 0; j < str.str().length(); ++j)
318
                cout << "\b";
319
320
321
322
            str.str("");
            str << min(sum_j-sum_k, sum_k-sum_j);
            cout << str.str() << flush;
        }
Tiago Peixoto's avatar
Tiago Peixoto committed
323
324
325
326
    }

    size_t E = sum_k;
 
327
328
329
    vector<dvertex_t> sources; // sources of edges
    typedef tr1::unordered_multimap<pair<size_t,size_t>, dvertex_t, 
                                    hash<pair<size_t,size_t> > > targets_t;
Tiago Peixoto's avatar
Tiago Peixoto committed
330
    targets_t targets; // vertices with j > 0
331
332
    typedef tr1::unordered_set<pair<size_t,size_t>, hash<pair<size_t,size_t> > > 
        target_degrees_t;
Tiago Peixoto's avatar
Tiago Peixoto committed
333
334
335
336
337
338
    target_degrees_t target_degrees; // existing (j,k) pairs
    
    // fill up sources, targets and target_degrees
    sources.reserve(E);
    for(size_t i = 0; i < N; ++i)
    {
339
340
341
342
        for(size_t k = 0; k < vertices[i].out_degree; ++k)
            sources.push_back(vertices[i]);
        if (vertices[i].in_degree > 0)
        {
343
344
345
346
347
            targets.insert(make_pair(make_pair(vertices[i].in_degree, 
                                               vertices[i].out_degree), 
                                     vertices[i]));
            target_degrees.insert(make_pair(vertices[i].in_degree, 
                                            vertices[i].out_degree));
348
        }
Tiago Peixoto's avatar
Tiago Peixoto committed
349
350
351
    }

    typedef multiset<pair<size_t,size_t>, total_deg_comp> ordered_degrees_t;
352
353
354
355
356
357
358
359
    ordered_degrees_t ordered_degrees; // (j,k) pairs ordered by (j+k), i.e,
                                       // total degree
    degree_matrix_t degree_matrix(target_degrees.size(), 
                                  min_j, min_k, 
                                  max_j, max_k); // (j,k) pairs layed out in a 2
                                                 // dimensional matrix
    for(typeof(target_degrees.begin()) iter = target_degrees.begin();
        iter != target_degrees.end(); ++iter)
360
361
362
363
        if (undirected_corr)
            ordered_degrees.insert(*iter);
        else
            degree_matrix.insert(*iter);
Tiago Peixoto's avatar
Tiago Peixoto committed
364
365
366
367
    
    // shuffle sources 
    for (size_t i = 0; i < sources.size(); ++i)
    {
368
369
        uniform_int<size_t> source_sample(i, sources.size()-1);
        swap(sources[i], sources[source_sample(rng)]);
Tiago Peixoto's avatar
Tiago Peixoto committed
370
371
372
    }

    if (verbose)
373
        cout << "\nadding edges: " << flush;
Tiago Peixoto's avatar
Tiago Peixoto committed
374
375
376
377
378

    // connect the sources to targets
    uniform_real<double> sample_probability(0.0, 1.0); 
    for (size_t i = 0; i < sources.size(); ++i)
    {
379
        dvertex_t source = sources[i], target;
380
381
382
383
384
385
386
        size_t j = source.in_degree;
        size_t k = source.out_degree;
        
        //choose the target vertex according to correlation
            
        pjk_t prob_func = lambda::bind(corr,lambda::_1,lambda::_2,j,k);
        pjk_t ceil = lambda::bind(ceil_corr,lambda::_1,lambda::_2,j,k);
387
388
389
390
        inv_ceil_t inv_ceil = lambda::bind(inv_ceil_corr,
                                           lambda::_1,lambda::_2,j,k);
        sample_from_distribution<pjk_t, pjk_t, inv_ceil_t> 
            corr_sample(prob_func, ceil, inv_ceil, ceil_corr_bound, rng);
391
392
393
394
395
396
397
        
        size_t jl,kl;
        tie(jl,kl) = corr_sample(); // target (j,k)
        
        target_degrees_t::iterator iter = target_degrees.find(make_pair(jl,kl));
        if (iter != target_degrees.end())
        {
398
399
            target = targets.find(*iter)->second; // if an (jl,kl) pair exists,
                                                  // just use that
400
401
        }
        else
402
        {        
403
404
405
            pair<size_t, size_t> deg;
            if (undirected_corr)
            {
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
                // select the (j,k) pair with the closest total degree (j+k)
                ordered_degrees_t::iterator upper;
                upper = ordered_degrees.upper_bound(make_pair(jl,kl));
                if (upper == ordered_degrees.end())
                {
                    --upper;
                    deg = *upper;
                }
                else if (upper == ordered_degrees.begin())
                {
                    deg = *upper;
                }
                else
                {
                    ordered_degrees_t::iterator lower = upper;
                    --lower;
422
423
                    if (jl + kl - (lower->first + lower->second) < 
                        upper->first + upper->second - (jl + kl))
424
                        deg = *lower;
425
426
                    else if (jl + kl - (lower->first + lower->second) != 
                             upper->first + upper->second - (jl + kl))
427
428
429
430
431
432
433
434
435
436
437
438
                        deg = *upper;
                    else
                    {
                        // if equal, choose randomly with equal probability
                        uniform_int<size_t> sample(0, 1);
                        if (sample(rng))
                            deg = *lower;
                        else
                            deg = *upper;
                    }
                }
                target = targets.find(deg)->second;
439
440
441
            }
            else
            {   
442
443
444
445
446
                // select the (j,k) which is the closest in the j,k plane.
                deg = degree_matrix.find_closest(jl, kl, rng);
                target = targets.find(deg)->second;
//                cerr << "wanted: " << jl << ", " << kl
//                     << " got: " << deg.first << ", " << deg.second << "\n";
447
               
448
            }            
449
450
451
452
        }

        //add edge
        graph_traits<multigraph_t>::edge_descriptor e;
453
454
        e = add_edge(vertex(source.index, _mg), 
                     vertex(target.index, _mg), _mg).first;
455
456
457
458
459
460
        _edge_index[e] = i;

        // if target received all the edges it should, remove it from target
        if (in_degree(vertex(target.index, _mg), _mg) == target.in_degree)
        {
            targets_t::iterator iter,end;
461
462
463
464
            for(tie(iter,end) = 
                    targets.equal_range(make_pair(target.in_degree, 
                                                  target.out_degree)); 
                iter != end; ++iter)
465
466
467
468
469
                if (iter->second == target)
                {
                    targets.erase(iter);
                    break;
                }
470

471
472
473
474
            // if there are no more targets with (jl,kl), remove pair from
            // target_degrees, etc.
            if (targets.find(make_pair(target.in_degree, target.out_degree)) ==
                targets.end())
475
            {
476
477
478
                target_degrees.erase(target_degrees.find(make_pair
                                                         (target.in_degree, 
                                                          target.out_degree)));
479
480
481
                if (target_degrees.bucket_count() > 2*target_degrees.size())
                {
                    target_degrees_t temp;
482
483
484
                    for(target_degrees_t::iterator iter = 
                            target_degrees.begin(); 
                        iter != target_degrees.end(); ++iter)
485
486
487
488
489
                        temp.insert(*iter);
                    target_degrees = temp;
                }
                if (undirected_corr)
                {
490
491
492
                    for(ordered_degrees_t::iterator iter = 
                            ordered_degrees.find(make_pair(target.in_degree, 
                                                           target.out_degree)); 
493
                        iter != ordered_degrees.end(); ++iter)
494
495
                        if (*iter == make_pair(target.in_degree, 
                                               target.out_degree))
496
497
498
499
500
501
502
                        {
                            ordered_degrees.erase(iter);
                            break;
                        }
                }
                else
                {
503
504
                    degree_matrix.erase(make_pair(target.in_degree, 
                                                  target.out_degree));
505
                }
506
507
508
509
510
511
            }
            
        }

        if (verbose)
        {
512
            static stringstream str;            
513
            for (size_t j = 0; j < str.str().length(); ++j)
514
                cout << "\b";
515
            for (size_t j = 0; j < str.str().length(); ++j)
516
                cout << " ";
517
            for (size_t j = 0; j < str.str().length(); ++j)
518
                cout << "\b";
519
520
521
522
523
            str.str("");
            str << (i+1) << " of " << E << " (" << (i+1)*100/E << "%)";
            cout << str.str() << flush;
        }
        
Tiago Peixoto's avatar
Tiago Peixoto committed
524
525
526
    }
    
    if (verbose)
527
        cout << "\n";
Tiago Peixoto's avatar
Tiago Peixoto committed
528
}