model.h 64.1 KB
Newer Older
1
2
3
#ifndef _SPEL_MODEL_MODEL_H_
#define _SPEL_MODEL_MODEL_H_

4
5
#include <stdexcept>

6
7
8
/*#include <Eigen/SVD>*/
/*#include <Eigen/QR>*/
#include "eigen.h"
Damien Leroux's avatar
Damien Leroux committed
9
#include "beta_gamma.h"
10

11
12
#include <cmath>

13

14
#include "labelled_matrix.h"
15
#include "settings.h"
16
#include "print.h"
17
#include "data/chromosome.h"
18
19
20

/*#define COMPONENT_EPSILON (1.e-10)*/
#define COMPONENT_EPSILON (active_settings->tolerance)
21
22
23

typedef labelled_matrix<Eigen::Matrix<double, -1, -1>, int, std::vector<char>> model_block_type;

24
25
26
27
struct model_block_key_struc;
typedef std::shared_ptr<model_block_key_struc> model_block_key;


Damien Leroux's avatar
Damien Leroux committed
28
29
struct model;

30
31
32
static inline
void f_test(const model& model_current, const model& model_new, int col_num, MatrixXd* pvalue, MatrixXd* lod);

Damien Leroux's avatar
Damien Leroux committed
33
34
35
36
37
38
39
40
41
struct key_already_exists : public std::exception {
    std::string m_what;
    key_already_exists(const model&, const model_block_key&);
    const char* what() const throw()
    {
        return m_what.c_str();
    }
};

42
43
typedef enum { mbk_CI, mbk_POP, mbk_Dominance, mbk_Interaction } key_type;

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
inline
collection<model_block_type>
disassemble_block_multipop(const model_block_type& block, const collection<const qtl_pop_type*>& all_pops)
{
    VectorXi keep = VectorXi::Zero(block.cols());
    int r0 = 0;
    collection<model_block_type> ret;
    ret.reserve(all_pops.size());
    for (const auto& vpop: all_pops) {
        const qtl_pop_type* pop = *vpop;
        auto sub_block = block.data.middleRows(r0, pop->size());
        size_t ncols = 0;
        for (int col = 0; col < sub_block.cols(); ++col) {
            keep(col) = !sub_block.col(col).isZero(active_settings->tolerance);
            ncols += keep(col);
        }
        /*MSG_DEBUG("disasm block keep = " << keep.transpose());*/
        /*MSG_QUEUE_FLUSH();*/
        ret.emplace_back(model_block_type{});
        auto& pop_block = *ret.back();
        pop_block.column_labels.reserve(ncols);
        pop_block.data.resize(pop->size(), ncols);
        int pcol = 0;
        for (int col = 0; col < sub_block.cols(); ++col) {
            if (keep(col)) {
                pop_block.data.col(pcol++) = sub_block.col(col);
                pop_block.column_labels.push_back(block.column_labels[col]);
            }
        }
        r0 += pop->size();
    }
    return ret;
}

inline
value<model_block_type>
assemble_block_multipop(const collection<model_block_type>& pop_blocks)
{
    std::map<std::vector<char>, int> col_indices;
    for (const auto& pb: pop_blocks) {
        for (const auto& v: pb->column_labels) {
            col_indices[v] = 0;
        }
    }

    int index = 0;
    for (auto& kv: col_indices) {
        kv.second = index++;
    }

    int r0 = 0;

    for (const auto& vpb: pop_blocks) {
        r0 += vpb->rows();
    }

    value<model_block_type> ret = model_block_type{};
    ret->data = MatrixXd::Zero(r0, col_indices.size());
    ret->column_labels.reserve(col_indices.size());
    for (const auto& kv: col_indices) {
        ret->column_labels.push_back(kv.first);
    }

    r0 = 0;

    for (const auto& vpb: pop_blocks) {
        const auto& pb = *vpb;
        for (int i = 0; i < pb.cols(); ++i) {
            int col = col_indices[pb.column_labels[i]];
            ret->data.block(r0, col, pb.rows(), 1) = pb.data.col(i);
        }
        r0 += pb.rows();
    }

    return ret;
}


122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137

struct model_block_key_struc {
    key_type type;

    const chromosome* chr;
    locus_key loci;
    model_block_key left, right;

    model_block_key_struc(key_type kt, const chromosome* c, const locus_key& lk, const model_block_key& c1, const model_block_key& c2)
        : type(kt), chr(c), loci(lk), left(c1), right(c2)
    {}

    size_t
        order() const
        {
            switch (type) {
138
                case mbk_CI: return 1;
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
                case mbk_POP: return 1; // loci->depth();
                case mbk_Dominance: return 2; // 2 * loci->depth();
                case mbk_Interaction: return left->order() + right->order();
            };
            return 0;
        }

    static
        model_block_key
        cross_indicator()
        {
            model_block_key void_;
            locus_key void_lk;
            return std::make_shared<model_block_key_struc>(mbk_CI, (const chromosome*) NULL, void_lk, void_, void_);
        }

    static
        model_block_key
        pop(const chromosome* c, locus_key loci)
        {
            model_block_key void_;
            return std::make_shared<model_block_key_struc>(mbk_POP, c, loci, void_, void_);
        }

    static
        model_block_key
        dominance(const model_block_key& haplo)
        {
            model_block_key void_;
            locus_key void_lk;
            return std::make_shared<model_block_key_struc>(mbk_Dominance, (const chromosome*) NULL, void_lk, haplo, void_);
        }

    static
        model_block_key
        interaction(const model_block_key& l, const model_block_key& r)
        {
            locus_key void_lk;
            if (l < r) {
                return std::make_shared<model_block_key_struc>(mbk_Interaction, (const chromosome*) NULL, void_lk, l, r);
            } else {
                return std::make_shared<model_block_key_struc>(mbk_Interaction, (const chromosome*) NULL, void_lk, r, l);
            }
        }

    friend
        std::ostream&
        operator << (std::ostream& os, const model_block_key_struc& mbk)
        {
            switch (mbk.type) {
                case mbk_CI:
                    os << "Cross";
                    break;
                case mbk_POP:
                    os << mbk.chr->name << ':' << mbk.loci;
                    break;
                case mbk_Dominance:
                    os << "Dom(" << mbk.left << ')';
                    break;
                case mbk_Interaction:
199
                    os << '(' << mbk.left << ":" << mbk.right << ')';
200
201
202
203
                    break;
            };
            return os;
        }
204

205
206
207
208
209
210
211
212
213
214
    friend
        std::ostream&
        operator << (std::ostream& os, const model_block_key& mbk)
        {
            return os << (*mbk);
        }

    bool
        can_interact_with(const model_block_key& mbk) const
        {
215
216
217
            if (type == mbk_CI || mbk->type == mbk_CI) {
                return active_settings->cross_indicator_can_interact;
            }
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
            auto v1 = flatten();
            auto v2 = mbk->flatten();
            decltype(v1) inter(std::min(v1.size(), v2.size()));
            auto e = std::set_intersection(v1.begin(), v1.end(), v2.begin(), v2.end(), inter.begin());
            return e == inter.begin();
        }

    bool
        has_chromosome(const chromosome* c) const
        {
            switch (type) {
                case mbk_CI:
                    return false;
                case mbk_POP:
                    return chr == c;
                case mbk_Dominance:
                    return left->has_chromosome(c);
                case mbk_Interaction:
                    return left->has_chromosome(c) || right->has_chromosome(c);
            };
            return false;
        }

    bool
        has_locus(const chromosome* c, double l) const
        {
            switch (type) {
                case mbk_CI:
                    return false;
                case mbk_POP:
                    return chr == c && loci->has(l);
                case mbk_Dominance:
                    return left->has_locus(c, l);
                case mbk_Interaction:
                    return left->has_locus(c, l) || right->has_locus(c, l);
            };
            return false;
        }

    bool
        has(const model_block_key& mbk) const
        {
            return *this == *mbk
                || (left && left->has(mbk))
                || (right && right->has(mbk));
        }

    bool
        operator < (const model_block_key_struc& other) const
        {
            /* CI < POP < Dom < Inter
             * then sort on haplotypes
             * for interactions, sort on left&right children
             */
            switch (type) {
                case mbk_CI:
                    return other.type != mbk_CI;
                case mbk_POP:
                    switch (other.type) {
                        case mbk_CI:
                            return false;
                        case mbk_POP:
                            return chr < other.chr || (chr == other.chr && loci < other.loci);
                        default:
                            return true;
                    };
                case mbk_Dominance:
                    switch (other.type) {
                        case mbk_CI:
                        case mbk_POP:
                            return false;
                        case mbk_Dominance:
                            return (*left) < (*other.left);
                        case mbk_Interaction:
                            return true;
                    };
                case mbk_Interaction:
295
296
297
298
299
300
301
302
303
304
                    {
                        auto o1 = order();
                        auto o2 = other.order();
                        if (o1 < o2) {
                            return true;
                        }
                        if (o1 > o2) {
                            return false;
                        }
                    }
305
306
307
308
309
310
311
312
313
314
315
316
317
                    switch (other.type) {
                        case mbk_Interaction:
                            return (*left) < (*other.left) || ((*left) == (*other.left) && (*right) < (*other.right));
                        default:
                            return false;
                    };
            };
            return false;
        }

    friend
        inline
        bool
318
319
320
321
322
323
324
325
326
327
        operator < (const model_block_key& k1, const model_block_key& k2)
        {
            if (!k1) {
                return !!k2;
            }
            if (!k2) {
                return false;
            }
            return (*k1) < (*k2);
        }
328
329
330
331
332
333
334
335
336

    bool
        operator == (const model_block_key_struc& other) const
        {
            if (type != other.type) {
                return false;
            }
            switch (type) {
                case mbk_Interaction:
337
                    if (!((*right) == (*other.right))) {
338
339
340
                        return false;
                    }
                case mbk_Dominance:
341
                    if (!((*left) == (*other.left))) {
342
343
344
345
346
347
348
349
350
351
352
353
354
355
                        return false;
                    }
                    break;
                case mbk_POP:
                    return chr == other.chr && loci == other.loci;
                case mbk_CI:
                    return true;
            };
            return false;
        }

    friend
        inline
        bool
356
357
358
359
        operator == (const model_block_key& k1, const model_block_key& k2)
        {
            return k1 && k2 && (*k1) == (*k2);
        }
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

    operator std::string () const { std::stringstream ret; ret << (*this); return ret.str(); }

private:
    typedef std::pair<const chromosome*, locus_key> item;

    std::vector<item>
        flatten() const
        {
            std::vector<item> ret;
            flatten(ret);
            std::sort(ret.begin(), ret.end(), [](const item& a, const item& b) { return a.first < b.first || (a.first == b.first && a.second < b.second); });
            return ret;
        }

    void
        flatten(std::vector<item>& ret) const
        {
            switch (type) {
                case mbk_Interaction:
                    right->flatten(ret);
                case mbk_Dominance:
                    left->flatten(ret);
                    break;
                case mbk_POP:
                    ret.emplace_back(chr, loci);
                case mbk_CI:
                    ;
            };
        }

};


394
struct mbk_comp {
395
    bool
396
        operator () (const model_block_key& k1, const model_block_key& k2) const
397
        {
398
            return (*k1) < (*k2);
399
        }
400
401
};

402
403

typedef std::map<model_block_key, value<model_block_type>, mbk_comp> model_block_collection;
404
405
406
407
408
409

namespace std {
    template <>
        struct hash<model_block_key> {
            size_t operator () (const model_block_key& mbk) const
            {
410
411
412
413
                std::hash<const chromosome*> hc;
                std::hash<locus_key> hlk;
                size_t accum;
#if 0
414
415
416
417
418
419
420
                /* WARNING FIXME this MUST NOT be used to hash a function parameter
                 * in a disk-cached task, because a POINTER is HASHED and the order
                 * is not guaranteed to be the same in every run.
                 */
                for (const auto& kv: mbk.selection) {
                    accum = impl::ROTATE<7>(impl::ROTATE<7>(accum * hc(kv.first)) ^ hlk(kv.second));
                }
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
#endif
                accum = 0xbadc0def;
                switch (mbk->type) {
                    case mbk_CI:
                        return 0;
                    case mbk_POP:
                        accum = 0xdeadbe3f;
                        accum = impl::ROTATE<7>(impl::ROTATE<7>(accum * hc(mbk->chr)) ^ hlk(mbk->loci));
                        break;
                    case mbk_Dominance:
                        accum = impl::ROTATE<7>(accum * operator () (mbk->left));
                        break;
                    case mbk_Interaction:
                        accum = impl::ROTATE<7>(accum * operator () (mbk->left)) ^ impl::ROTATE<13>(accum * operator () (mbk->right));
                };
436
437
438
439
440
441
442
                return accum;
            }
        };
} // namespace std



443
static inline
444
bool
445
446
447
448
449
450
around_zero(double o)
{
    return o < COMPONENT_EPSILON && o > -COMPONENT_EPSILON;
}

static inline
451
bool
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
much_smaller_than(double a, double b)
{
    return a < (COMPONENT_EPSILON * b);
}

static inline
void
set_if_much_smaller_than(double& a, double b)
{
    double tmp = COMPONENT_EPSILON * b;
    if (a < tmp) {
        a = tmp;
    }
}

467
468
469

using namespace Eigen;

470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
static inline
MatrixXd concat_right(const std::vector<MatrixXd>& mat_vec)
{
    size_t full_size = 0;
    MatrixXd ret;
    for (auto& m: mat_vec) {
        full_size += m.outerSize();
        /*MSG_DEBUG("preparing concat_right with matrix(" << m->innerSize() << ',' << m->outerSize() << ')');*/
    }
    ret.resize(mat_vec.front().innerSize(), full_size);
    full_size = 0;
    for (auto& m: mat_vec) {
        /*MSG_DEBUG("concat_right in M(" << ret.innerSize() << ',' << ret.outerSize() << ") at col " << full_size << "matrix(" << m->innerSize() << ',' << m->outerSize() << ')');*/
        /*ret.block(0, full_size, ret.innerSize(), m->outerSize()) = *m;*/
        ret.middleCols(full_size, m.outerSize()) = m;
        full_size += m.outerSize();
    }
    return ret;
}

490
static inline
491
MatrixXd concat_right(const collection<model_block_type>& mat_vec)
492
493
494
{
    size_t full_size = 0;
    MatrixXd ret;
495
496
    for (auto m: mat_vec) {
        full_size += m->outerSize();
497
        /*MSG_DEBUG("preparing concat_right with matrix(" << m->innerSize() << ',' << m->outerSize() << ')');*/
498
    }
499
    ret.resize(mat_vec.front()->innerSize(), full_size);
500
    full_size = 0;
501
    for (auto m: mat_vec) {
502
503
        /*MSG_DEBUG("concat_right in M(" << ret.innerSize() << ',' << ret.outerSize() << ") at col " << full_size << "matrix(" << m->innerSize() << ',' << m->outerSize() << ')');*/
        /*ret.block(0, full_size, ret.innerSize(), m->outerSize()) = *m;*/
504
        ret.middleCols(full_size, m->outerSize()) = m->data;
505
        full_size += m->outerSize();
506
507
508
509
    }
    return ret;
}

510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530

static inline
MatrixXd concat_right(const model_block_collection& mat_map)
{
    size_t full_size = 0;
    MatrixXd ret;
    for (auto m: mat_map) {
        full_size += m.second->outerSize();
        /*MSG_DEBUG("preparing concat_right with matrix(" << m->innerSize() << ',' << m->outerSize() << ')');*/
    }
    ret.resize(mat_map.begin()->second->innerSize(), full_size);
    full_size = 0;
    for (auto m: mat_map) {
        /*MSG_DEBUG("concat_right in M(" << ret.innerSize() << ',' << ret.outerSize() << ") at col " << full_size << "matrix(" << m->innerSize() << ',' << m->outerSize() << ')');*/
        /*ret.block(0, full_size, ret.innerSize(), m->outerSize()) = *m;*/
        ret.middleCols(full_size, m.second->outerSize()) = m.second->data;
        full_size += m.second->outerSize();
    }
    return ret;
}

Damien Leroux's avatar
Damien Leroux committed
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
static inline
MatrixXd concat_down(const std::vector<MatrixXd>& mat_vec)
{
    size_t full_size = 0;
    MatrixXd ret;
    for (auto& m: mat_vec) {
        full_size += m.innerSize();
    }
    ret.resize(full_size, mat_vec.front().outerSize());
    full_size = 0;
    for (auto& m: mat_vec) {
        ret.middleRows(full_size, m.innerSize()) = m;
        full_size += m.innerSize();
    }
    return ret;
}

548
static inline
549
MatrixXd concat_down(const std::vector<const MatrixXd*>& mat_vec)
550
551
552
{
    size_t full_size = 0;
    MatrixXd ret;
553
554
    for (auto m: mat_vec) {
        full_size += m->innerSize();
555
    }
556
    ret.resize(full_size, mat_vec.front()->outerSize());
557
    full_size = 0;
558
    for (auto m: mat_vec) {
559
560
        /*ret.block(full_size, 0, m->innerSize(), ret.outerSize()) = *m;*/
        ret.middleRows(full_size, m->innerSize()) = *m;
561
        full_size += m->innerSize();
562
563
564
565
566
567
568
569
570
    }
    return ret;
}


static inline
std::pair<int, MatrixXd>
rank_and_components(const MatrixXd& M)
{
571
    JacobiSVD<MatrixXd> svd(M, ComputeThinU);
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591

    std::cout << "Singular values " << svd.singularValues().transpose() << std::endl;
    int nzsv = svd.nonzeroSingularValues();

    return {nzsv, svd.matrixU().leftCols(nzsv)};
}


static inline
MatrixXd components(const MatrixXd& M, const MatrixXd& P)
{
    MatrixXd pnorm(P.innerSize(), P.outerSize());
    for (int i = 0; i < P.outerSize(); ++i) {
        pnorm.col(i) = P.col(i).normalized();
    }
    MatrixXd orth = M - pnorm * pnorm.transpose() * M; /* feu ! */
    return rank_and_components(orth).second;
}


592
enum class SolverType { QR };
593

Damien Leroux's avatar
Damien Leroux committed
594
595
typedef std::vector<std::map<model_block_key, MatrixXd>> constraint_list;

596
597
598
599
600
601
602
603
604
605
606
607
inline
std::ostream& operator << (std::ostream& os, const constraint_list& cl)
{
    for (const auto& map: cl) {
        for (const auto& kv: map) {
            os << kv.first << std::endl << kv.second << std::endl;
        }
        os << "---" << std::endl;
    }
    return os;
}

Damien Leroux's avatar
Damien Leroux committed
608
609

MatrixXd
610
contrast_groups(const collection<const qtl_pop_type*>& all_pops, const locus_key& lk);
Damien Leroux's avatar
Damien Leroux committed
611

Damien Leroux's avatar
Damien Leroux committed
612

613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659

struct filename_stream {
    struct filtering_streambuf : public std::streambuf {
        std::streambuf* original;
        filtering_streambuf(std::streambuf* _) : original(_) {}
    protected:
        int overflow(int ch) override
        {
            switch (ch) {
                case ',':
                case '}':
                    return ch;
                case '{':
                    ch = '-';
                    break;
                case ' ':
                    ch = '_';
                default:;
            };
            return original->sputc(ch);
        }
    };

    std::ostringstream sstream;
    filtering_streambuf rdbuf;
    std::ostream ostream;

    filename_stream()
        : sstream(), rdbuf(sstream.rdbuf()), ostream(&rdbuf)
    {}

    template <typename T>
        friend
        filename_stream& operator << (filename_stream& fs, T x)
        {
            fs.ostream << x;
            return fs;
        }

    operator
        std::string () const
        {
            return sstream.str();
        }
};


Damien Leroux's avatar
Damien Leroux committed
660
/*#define ULTRA_MEGA_PARANOID*/
661
662
663
static inline
MatrixXd matrix_inverse(const MatrixXd& m)
{
Damien Leroux's avatar
Damien Leroux committed
664
665
    /*return m.fullPivHouseholderQr().inverse();*/
    /*return m.inverse();*/
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
    /* FIXME Should NOT use SVD (see note in compute() */
    JacobiSVD<MatrixXd> inverter(m, ComputeFullV);
    auto& V = inverter.matrixV();
    VectorXd inv_sv(inverter.singularValues());
    for (int i = 0; i < inv_sv.innerSize(); ++i) {
        if (!around_zero(inv_sv(i))) {
            inv_sv(i) = 1. / inv_sv(i);
        } else {
            inv_sv(i) = 0.;
        }
    }
#ifdef ULTRA_MEGA_PARANOID
    /* CHECK THAT THIS IS STABLE BY THE BEN ISRAEL SEQUENCE */
    MatrixXd svd_ret = V * inv_sv.asDiagonal() * V.transpose();
    MatrixXd ret = svd_ret;
    MatrixXd test = 2 * ret - ret * m * ret;
    if (!test.isApprox(ret, .00001)) {
        MSG_DEBUG("ret" << std::endl << ret << std::endl << "test" << std::endl << test);
    } else {
        MSG_DEBUG("m^-1 IS GOOD! Yeah man.");
    }
    return ret;
#else
    return V * inv_sv.asDiagonal() * V.transpose();
#endif
}


694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
struct gauss_elimination {
    MatrixXd block;
    std::vector<int> leading;

    gauss_elimination(const MatrixXd& mat)
        : block(mat)
        , leading(mat.rows())
    {
        for (int i = 0; i < block.rows(); ++i) {
            find_leading_coef(i);
        }

        sort_rows(0);
        for (int ref = 0; ref < block.rows(); ++ref) {
            if (is_null(ref)) {
                break;
            }
            anihilate_coef(ref, leading[ref]);
            sort_rows(ref);
        }
    }

    bool is_null(int row) const { return leading[row] == block.cols(); }
    int not_null(int row) const { return (int) (leading[row] != block.cols()); }

    int
        rank() const
        {
            /*MSG_DEBUG("echelon form" << std::endl << block);*/
            int count = 0;
            for (int i = 0; i < block.rows(); ++i) {
                count += not_null(i);
            }
            return count;
        }

    void
        sort_rows(int start)
        {
            std::vector<int> indices(block.rows() - start);
            int i = start;
            std::generate(indices.begin(), indices.end(), [&] () { return i++; });
            std::sort(indices.begin(), indices.end(), [&](int a, int b) { return leading[a] < leading[b]; });
            int sz = (int) indices.size();
            for (i = 0; i < sz; ++i) {
                if (i > indices[i]) {
                    swap_rows(i, indices[i]);
                }
            }
        }

    void
        swap_rows(int r1, int r2)
        {
            VectorXd tmp = block.row(r1);
            block.row(r1) = block.row(r2);
            block.row(r2) = tmp;
            std::swap(leading[r1], leading[r2]);
        }

    void
        anihilate_coef(int ref, int j)
        {
            for (int i = ref + 1; i < block.rows(); ++i) {
                if (leading[i] == j) {
                    anihilate_coef_in_row(ref, i, j);
                }
            }
        }

    void
        anihilate_coef_in_row(int ref, int r, int j)
        {
            block.row(r) *= 1. / block(r, j);
            block.row(r) -= block.row(ref);
            block(r, j) = 0;
            find_leading_coef(r);
            if (not_null(r)) {
                block.row(r) *= 1. / block(r, leading[r]);
            }
        }

    void
        find_leading_coef(int r)
        {
            int j;
            for (j = 0; j < block.cols() && block(r, j) == 0; ++j);
            leading[r] = j;
        }
};

785

786
787
struct model {
    model()
788
        : m_Y(), m_blocks(), m_X(), m_rank(), m_rss(), m_coefficients(), m_solver_type(), m_computed(false), m_with_constraints(false), m_ghost_constraint(), m_all_pops(), m_ancestor_names(), m_max_order(1), m_threshold(0)
789
790
791

    {}

792
    model(const value<MatrixXd>& y, double threshold, const collection<const qtl_pop_type*>& pops, const std::map<char, std::string>& anam, size_t mo=1, SolverType st=SolverType::QR)
793
794
795
796
797
798
799
800
801
        : m_Y(y)
        , m_blocks(), m_X()
		, m_rank(), m_rss(), m_coefficients(), m_residuals()
		, m_solver_type(st)
        , m_computed(false)
        , m_with_constraints(false)
        , m_ghost_constraint()
        , m_all_pops(pops)
        , m_ancestor_names(anam)
802
        , m_max_order(mo)
803
        , m_threshold(threshold)
804
    /*{ MSG_DEBUG("new model " << __LINE__ << " with Y(" << y.innerSize() << ',' << y.outerSize() << ')'); }*/
805
806
    {}

807
    model(const value<MatrixXd>& y, double threshold, const collection<const qtl_pop_type*>& pops, size_t mo=1, SolverType st=SolverType::QR)
808
        : m_Y(y)
809
        , m_blocks(), m_X()
810
		, m_rank(), m_rss(), m_coefficients(), m_residuals()
811
812
		, m_solver_type(st)
        , m_computed(false)
Damien Leroux's avatar
Damien Leroux committed
813
        , m_with_constraints(false)
814
        , m_ghost_constraint()
Damien Leroux's avatar
Damien Leroux committed
815
        , m_all_pops(pops)
816
        , m_ancestor_names((*pops.front())->ancestor_names)
817
        , m_max_order(mo)
818
        , m_threshold(threshold)
819
820
821
822
823
    /*{ MSG_DEBUG("new model " << __LINE__ << " with Y(" << y.innerSize() << ',' << y.outerSize() << ')'); }*/
    {}

    model(const model& mo)
        : m_Y(mo.m_Y)
824
        , m_blocks(mo.m_blocks), m_X(mo.m_X)
825
		, m_rank(mo.m_rank), m_rss(mo.m_rss), m_coefficients(mo.m_coefficients)
826
        , m_residuals(mo.m_residuals)
827
828
		, m_solver_type(mo.m_solver_type)
        , m_computed(mo.m_computed)
829
830
        , m_with_constraints(mo.m_with_constraints)
        , m_ghost_constraint()
Damien Leroux's avatar
Damien Leroux committed
831
        , m_all_pops(mo.m_all_pops)
832
        , m_ancestor_names(mo.m_ancestor_names)
833
        , m_max_order(mo.m_max_order)
834
        , m_threshold(mo.m_threshold)
835
836
837
    /*{ MSG_DEBUG("new model " << __LINE__ << " with Y(" << m_Y->innerSize() << ',' << m_Y->outerSize() << ')'); }*/
    {}

838
839
840
841
842
843
844
845
846
847
    model(const value<MatrixXd>& y, const model& mo)
        : m_Y(y)
        , m_blocks(mo.m_blocks), m_X(mo.m_X)
		, m_rank(), m_rss(), m_coefficients()
        , m_residuals()
		, m_solver_type(mo.m_solver_type)
        , m_computed(false)
        , m_with_constraints(mo.m_with_constraints)
        , m_ghost_constraint()
        , m_all_pops(mo.m_all_pops)
848
        , m_ancestor_names(mo.m_ancestor_names)
849
        , m_max_order(mo.m_max_order)
850
        , m_threshold(mo.m_threshold)
851
852
853
    /*{ MSG_DEBUG("new model " << __LINE__ << " with Y(" << m_Y->innerSize() << ',' << m_Y->outerSize() << ')'); }*/
    {}

854
# if 0
855
856
    model&
        operator = (const model& mo)
857
858
        = delete;
#else
859
860
861
862
863
864
865
866
867
868
869
870
871
872
    model&
        operator = (const model& mo)
        {
            m_Y = mo.m_Y;
            m_blocks = mo.m_blocks;
            m_computed = mo.m_computed;
            m_with_constraints = mo.m_with_constraints;
            m_ghost_constraint = mo.m_ghost_constraint;
            m_X = mo.m_X;
            m_rss = mo.m_rss;
            m_coefficients = mo.m_coefficients;
            m_residuals = mo.m_residuals;
            m_rank = mo.m_rank;
            m_all_pops = mo.m_all_pops;
873
            m_ancestor_names = mo.m_ancestor_names;
874
            m_max_order = mo.m_max_order;
875
            m_threshold = mo.m_threshold;
876
877
            return *this;
        }
878

879
880
881
882
883
884
885
886
887
888
889
890
891
892
    model&
        operator = (model&& mo)
        {
            m_Y = mo.m_Y;
            m_blocks.clear();
            m_blocks.swap(mo.m_blocks);
            m_computed = mo.m_computed;
            m_with_constraints = mo.m_with_constraints;
            m_ghost_constraint = mo.m_ghost_constraint;
            m_X = mo.m_X;
            m_rss = mo.m_rss;
            m_coefficients = mo.m_coefficients;
            m_residuals = mo.m_residuals;
            m_rank = mo.m_rank;
893
            m_ancestor_names.swap(mo.m_ancestor_names);
894
            m_max_order = mo.m_max_order;
895
            m_threshold = mo.m_threshold;
896
897
            return *this;
        }
898
#endif
899

Damien Leroux's avatar
Damien Leroux committed
900
    constraint_list
901
902
903
        compute_constraint(const model_block_key& mbk, const model_block_type& mb)
        {
            constraint_list ret;
904
            if (mbk->type == mbk_CI) {
905
906
                return ret;
            /*} else if (mbk.selection.size() == 1) {*/
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
            } else if (mbk->type == mbk_POP) {
                size_t n_qtl = mb.column_labels.front().size();
                if (n_qtl == 1) {
                    /* only one chromosome: do the contrast groups trick */
                    ret = {{{mbk, contrast_groups(m_all_pops, mbk->loci)}}};
                } else {
                    std::vector<std::set<char>> uniq_letters_per_qtl(mb.column_labels.front().size());
                    /* need the epistasis magic here */
                    for (const auto& vec: mb.column_labels) {
                        auto i = uniq_letters_per_qtl.begin();
                        for (char c: vec) { i->insert(c); ++i; }
                    }
                    std::vector<size_t> letter_counts(uniq_letters_per_qtl.size());
                    for (size_t i = 0; i < letter_counts.size(); ++i) { letter_counts[i] = uniq_letters_per_qtl[i].size(); }
                    size_t start = 1;
                    size_t finish = (1 << letter_counts.size()) - 1;
                    for (size_t variant = start; variant < finish; ++variant) {
                        MatrixXd constraint = MatrixXd::Identity(1, 1);
                        for (size_t i = 0; i < letter_counts.size(); ++i) {
                            bool flat = (variant >> i) & 1;
                            MatrixXd tmp;
                            if (flat) {
                                tmp = kroneckerProduct(constraint, MatrixXd::Identity(letter_counts[i], letter_counts[i]));
                            } else {
                                tmp = kroneckerProduct(constraint, MatrixXd::Identity(letter_counts[i], letter_counts[i]));
                            }
                            constraint = tmp;
934
                        }
935
                        ret.emplace_back();
936
937
                        /*MSG_DEBUG("Created constraint");*/
                        /*MSG_DEBUG("" << constraint);*/
938
939
940
941
                        ret.back().insert({{mbk, constraint}});
                    }
                }
            } else if (mbk->type == mbk_Interaction) {
942
943
                /*MSG_DEBUG("Computing constraint for interaction " << mbk);*/
                /*MSG_QUEUE_FLUSH();*/
944
945
946
947
948
949
950
                auto b1 = m_blocks[mbk->left];
                auto b2 = m_blocks[mbk->right];
                int cols1 = b1->data.cols();
                int cols2 = b2->data.cols();
                constraint_list
                    C1 = compute_constraint(mbk->left, *b1),
                    C2 = compute_constraint(mbk->right, *b2);
951
952
953
954
955
956
957
958
959
960
                if (C1.size() == 0) {
                    C1 = {{{{mbk->left, MatrixXd::Ones(1, b1->cols())}}}};
                }
                if (C2.size() == 0) {
                    C2 = {{{{mbk->left, MatrixXd::Ones(1, b2->cols())}}}};
                }
                /*MSG_DEBUG("Computing interaction constraints from" << std::endl*/
                        /*<< "C1" << std::endl << C1*/
                        /*<< "C2" << std::endl << C2*/
                        /*);*/
961
                MatrixXd constraint;
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
                MatrixXd ghost_columns;
                if (mb.cols() < b1->cols() * b2->cols()) {
                    ghost_columns = MatrixXd::Zero(b1->cols() * b2->cols(), mb.cols());
                    int i = 0;
                    int j = 0;
                    /*MSG_DEBUG("Computing ghost columns");*/
                    /*MSG_DEBUG("L1 " << b1->column_labels);*/
                    /*MSG_DEBUG("L2 " << b2->column_labels);*/
                    for (const auto& l1: b1->column_labels) {
                        for (const auto& l2: b2->column_labels) {
                            auto il = make_interaction_label(l1, l2);
                            /*MSG_DEBUG("-- compare " << il << " and " << mb.column_labels[j]);*/
                            if (il == mb.column_labels[j]) {
                                ghost_columns(i, j++) = 1;
                            }
                            i++;
                        }
                    }
                    /*MSG_DEBUG("ghost_columns redux" << std::endl << ghost_columns);*/
                }
982
983
984
                for (const auto& map1: C1) {
                    for (const auto& c1: map1) {
                        constraint = kroneckerProduct(c1.second, MatrixXd::Identity(cols2, cols2));
985
986
987
988
                        if (ghost_columns.size()) {
                            constraint = constraint * ghost_columns;
                        }
                        ret.emplace_back();
989
990
991
992
993
                        ret.back().insert({{mbk, constraint}});
                    }
                }
                for (const auto& map2: C2) {
                    for (const auto& c2: map2) {
994
995
996
997
998
                        constraint = kroneckerProduct(MatrixXd::Identity(cols1, cols1), c2.second);
                        if (ghost_columns.size()) {
                            constraint = constraint * ghost_columns;
                        }
                        ret.emplace_back();
999
                        ret.back().insert({{mbk, constraint}});
1000
                    }