pedigree.h 55.7 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
/* Spell-QTL  Software suite for the QTL analysis of modern datasets.
 * Copyright (C) 2016,2017  Damien Leroux <damien.leroux@inra.fr>, Sylvain Jasson <sylvain.jasson@inra.fr>
 *
 * This program is free software: you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation, either version 3 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
 * along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */

18
19
20
21
#ifndef _SPELL_BAYES_CSV_H_
#define _SPELL_BAYES_CSV_H_

#include <iostream>
22
#include "file.h"
23
24
25
26
27
#include <string>
#include <sstream>
#include <stdexcept>
#include <vector>
#include <utility>
28
#include <unordered_set>
29
30


Damien Leroux's avatar
Damien Leroux committed
31
32
/*#include "permutation.h"*/
/*#include "symmetry.h"*/
33
34
#include "geno_matrix.h"
#include "linear_combination.h"
Damien Leroux's avatar
Damien Leroux committed
35
#include "pedigree_tree.h"
36
#include "bayes/output.h"
37
38


39
40
41
42
43
44
45
46
47
48
49
struct bn_label_type {
    char first;
    char second;
    char first_allele;
    char second_allele;
    bn_label_type() : first(0), second(0), first_allele(0), second_allele(0) {}
    bn_label_type(int) : first(0), second(0), first_allele(0), second_allele(0) {}
    bn_label_type(char f, char s, char fa, char sa)
        : first(f), second(s), first_allele(fa), second_allele(sa)
    {}

50
51
52
    int compact() const { return *(int*) this; }
    int& compact() { return *(int*) this; }

53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
    friend std::ostream& operator << (std::ostream& os, const bn_label_type& bl)
    {
        if (bl.second != GAMETE_EMPTY) {
            return os << bl.first << ((int) bl.first_allele) << bl.second << ((int) bl.second_allele);
        } else {
            return os << bl.first << ((int) bl.first_allele);
        }
    }

    bool operator < (const bn_label_type& other) const
    {
        /*return first < other.first || (first == other.first*/
            /*&& (second < other.second || (second == other.second*/
            /*&& (first_allele < other.first_allele || (first_allele == other.first_allele*/
            /*&& second_allele < other.second_allele)))));*/
        return (*(int*)this) < (*(int*) &other);
    }

    bool operator == (const bn_label_type& other) const
    {
        return (*(int*)this) == (*(int*) &other);
    }
75
76
77
78
79

    bool operator != (const bn_label_type& other) const
    {
        return (*(int*)this) != (*(int*) &other);
    }
80
81
};

82
typedef combination_type<int, bn_label_type> genotype_comb_type;
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


template <typename Arg>
int read_field(std::stringstream& s, char sep, Arg& arg)
{
    std::string field;
    std::getline(s, field, sep);
    /*MSG_DEBUG("CSV FIELD |" << field << "|");*/
    std::stringstream ss(field);
    ss >> arg;
    return 0;
}


#define do_with_arg_pack(_expr) do { using _ = int[]; (void)_{0, ((_expr), void(), 0)...}; } while(0)

template <typename... Args>
void read_csv_line(std::istream& is, char sep, Args&... args)
{
    std::string line;
    std::getline(is, line);
    /*MSG_DEBUG("CSV LINE |" << line << "|");*/
    std::stringstream ss(line);
    do_with_arg_pack(read_field(ss, sep, args));
}





struct pedigree_item {
    std::string gen_name;
    int id;
    int p1;
    int p2;

119
120
    pedigree_item() : gen_name(), id(0), p1(0), p2(0) {}

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
    pedigree_item(const char* gn, int i, int a, int b)
        : gen_name(gn), id(i), p1(a), p2(b)
    {}

    pedigree_item(std::istream& is, char field_sep=';')
        : gen_name()
    {
        id = p1 = p2 = 0;
        read_csv_line(is, field_sep, gen_name, id, p1, p2);
        if (id == p1 && id == p2 && id == 0) {
            return;
        }
        if (id <= p1 || id <= p2) {
            throw std::runtime_error("Bad ID! ID must be greater than p1 AND p2");
            /*MSG_DEBUG("BAD ID!! " << id << " must be greater than " << p1 << " AND " << p2);*/
        }
    }

    bool is_ancestor() const { return p1 == 0 && p2 == 0; }
    bool is_self() const { return p1 > 0 && p1 == p2; }
    bool is_cross() const { return p1 > 0 && p2 > 0 && p1 != p2; }
    bool is_dh() const { return p1 != p2 && p1 >= 0 && p2 >= 0 && (p1 * p2) == 0; }
    bool is_bullshit() const { return !(is_ancestor() || is_self() || is_cross() || is_dh()); }
};


std::vector<pedigree_item>
read_csv(const std::string& pedigree_file, char field_sep=';');



typedef std::map<size_t, size_t> ancestor_list_type;


155
inline
156
157
158
159
160
161
162
163
164
165
166
167
ancestor_list_type reentrants(const ancestor_list_type& a)
{
    ancestor_list_type ret;
    for (const auto& kv: a) {
        if (kv.second > 1) {
            ret.emplace(kv);
        }
    }
    return ret;
}


168
inline
169
170
171
172
173
174
175
176
177
178
ancestor_list_type operator + (const ancestor_list_type& a1, const ancestor_list_type& a2)
{
    ancestor_list_type ret(a1);
    for (const auto& kv: a2) {
        ret[kv.first] += kv.second;
    }
    return ret;
}


179
inline
180
181
182
183
184
185
186
187
188
189
190
191
192
ancestor_list_type operator / (const ancestor_list_type& a, const ancestor_list_type& restr)
{
    ancestor_list_type ret;
    for (const auto& kv: a) {
        auto i = restr.find(kv.first);
        if (i != restr.end()) {
            ret.emplace(kv.first, std::min(kv.second, i->second));
        }
    }
    return ret;
}


193
inline
194
195
196
197
198
199
200
201
202
203
204
205
ancestor_list_type operator % (const ancestor_list_type& a, const ancestor_list_type& restr)
{
    ancestor_list_type ret;
    for (const auto& kv: a) {
        if (restr.find(kv.first) != restr.end()) {
            ret.emplace(kv);
        }
    }
    return ret;
}


206
inline
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
ancestor_list_type operator - (const ancestor_list_type& a, const ancestor_list_type& restr)
{
    ancestor_list_type ret;
    for (const auto& kv: a) {
        auto it = restr.find(kv.first);
        if (it == restr.end()) {
            ret.emplace(kv);
        } else if (kv.second > it->second) {
            ret.emplace(kv.first, kv.second - it->second);
        }
    }
    return ret;
}


222
inline
223
224
225
226
227
228
229
230
231
232
ancestor_list_type operator * (const ancestor_list_type& a, size_t weight)
{
    ancestor_list_type ret;
    for (const auto& kv: a) {
        ret.emplace(kv.first, kv.second * weight);
    }
    return ret;
}


233
inline
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
std::ostream& operator << (std::ostream& os, const ancestor_list_type& a)
{
    auto i = a.begin();
    auto j = a.end();
    if (i != j) {
        os << i->first << ':' << i->second;
        for (++i; i != j; ++i) {
            os << ' ' << i->first << ':' << i->second;
        }
    } else {
        os << "empty";
    }
    return os;
}


250
#if 0
251
252
253
254
255
256
257
258
259
260
261
262
label_type operator * (label_type a, label_type b)
{
    label_type ret;
    if (a.second == GAMETE_EMPTY) {
        ret = {a.first, b.first};
    } else {
        if (b.second == GAMETE_EMPTY) {
            ret = {SELECT(a, b.first), GAMETE_EMPTY};
        } else {
            ret = {SELECT(a, b.first), SELECT(a, b.second)};
        }
    }
263
264
265
    /*MSG_DEBUG("" << a << " * " << b << " = " << ret);*/
    return ret;
}
266
#endif
267
268
269
270


#define SELECT_A(__p, __b) ((__b) == GAMETE_R ? (__p).second_allele : (__p).first_allele)

271
inline
272
273
274
275
276
277
278
279
280
281
282
283
284
bn_label_type operator * (bn_label_type a, bn_label_type b)
{
    bn_label_type ret;
    if (a.second == GAMETE_EMPTY) {
        ret = {a.first, b.first, a.first_allele, b.first_allele};
    } else {
        if (b.second == GAMETE_EMPTY) {
            ret = {SELECT(a, b.first), GAMETE_EMPTY, SELECT_A(a, b.first), 0};
        } else {
            ret = {SELECT(a, b.first), SELECT(a, b.second), SELECT_A(a, b.first), SELECT_A(a, b.second)};
        }
    }
    /*MSG_DEBUG("" << a << " * " << b << " = " << ret);*/
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
    return ret;
}


template <typename F> struct make_one;
template <> struct make_one<MatrixXd> {
    static MatrixXd& _(bool der) {
        static MatrixXd one = MatrixXd::Ones(1, 1);
        static MatrixXd zero = MatrixXd::Zero(1, 1);
        return der ? zero : one;
    }
};
template <> struct make_one<VectorXd> {
    static VectorXd& _(bool der) {
        static VectorXd one = VectorXd::Ones(1);
        static VectorXd zero = VectorXd::Zero(1);
        return der ? zero : one;
    }
};


306
307
308
309
310
311
312
313
314
315
template <typename PARENT_TYPE, typename STATE_TYPE>
struct rw_comb : public rw_any<rw_comb<PARENT_TYPE, STATE_TYPE>> {
    typedef combination_type<PARENT_TYPE, STATE_TYPE> comb_type;

    virtual ~rw_comb() {}

    using rw_any<rw_comb<PARENT_TYPE, STATE_TYPE>>::fourcc;
    using rw_any<rw_comb<PARENT_TYPE, STATE_TYPE>>::ref;
    using rw_any<rw_comb<PARENT_TYPE, STATE_TYPE>>::operator ();

316
    void operator () (ifile& ifs, bn_label_type& l) { l.compact() = read_int(ifs); }
317
    void operator () (ofile& ofs, bn_label_type& l) { write_int(ofs, l.compact()); }
318

319
    void operator () (ifile& fs, typename comb_type::key_type& key) { ref() (fs, key.parent); ref() (fs, key.state); }
320
    void operator () (ofile& fs, typename comb_type::key_type& key) { ref() (fs, key.parent); ref() (fs, key.state); }
321

322
    void operator () (ifile& fs, typename comb_type::key_list& keys) { ref() (fs, keys.keys); }
323
    void operator () (ofile& fs, typename comb_type::key_list& keys) { ref() (fs, keys.keys); }
324

325
    void operator () (ifile& fs, typename comb_type::element_type& elem) { ref() (fs, elem.keys); ref() (fs, elem.coef); }
326
    void operator () (ofile& fs, typename comb_type::element_type& elem) { ref() (fs, elem.keys); ref() (fs, elem.coef); }
327

328
    void operator () (ifile& fs, comb_type& comb)
329
330
331
332
333
    {
        if (fourcc(fs, "COMB")) { return; }
        ref() (fs, comb.m_combination);
    }

334
    void operator () (ofile& fs, comb_type& comb)
335
336
337
338
339
340
341
342
343
344
345
346
347
348
    {
        if (fourcc(fs, "COMB")) { return; }
        ref() (fs, comb.m_combination);
    }
};


struct rw_tree : public rw_any<rw_tree> {
    virtual ~rw_tree() {}

    using rw_any<rw_tree>::fourcc;
    using rw_any<rw_tree>::ref;
    using rw_any<rw_tree>::operator ();

349
    void operator () (ifile& fs, pedigree_tree_type::node_type& node) { ref() (fs, node.p1); ref() (fs, node.p2); }
350
    void operator () (ofile& fs, pedigree_tree_type::node_type& node) { ref() (fs, node.p1); ref() (fs, node.p2); }
351

352
353
354
355
356
357
358
359
360
361
362
363
    template <typename STREAM_TYPE, typename TREE_TYPE>
        void tree_io_impl(STREAM_TYPE& fs, TREE_TYPE&& tree)
        {
            ref() (fs, tree.m_leaves);
            ref() (fs, tree.m_roots);
            ref() (fs, tree.m_nodes);
            ref() (fs, tree.m_must_recompute);
            ref() (fs, tree.m_node_number_to_ind_number);
            ref() (fs, tree.m_ind_number_to_node_number);
            ref() (fs, tree.m_original_ordering);
        }

364
    void operator () (ifile& fs, pedigree_tree_type& tree)
365
    {
366
        tree_io_impl(fs, tree);
367
368
    }

369
    void operator () (ofile& fs, pedigree_tree_type& tree)
370
    {
371
        tree_io_impl(fs, tree);
372
    }
373
374
375
376
377
378
379
380
381

    void operator () (ifile& fs, pedigree_item& pi)
    {
        (*this)(fs, pi.gen_name);
        (*this)(fs, pi.id);
        (*this)(fs, pi.p1);
        (*this)(fs, pi.p2);
    }

382
    void operator () (ofile& fs, pedigree_item& pi)
383
384
385
386
387
388
    {
        (*this)(fs, pi.gen_name);
        (*this)(fs, pi.id);
        (*this)(fs, pi.p1);
        (*this)(fs, pi.p2);
    }
389
390
391
};


392
393
394
395
396
397
398
/* TODO extraire l'arbre du pedigree
 * TODO opérations sur l'arbre :
 * TODO - insérer un nouveau noeud étant donné {P1, P2} (Pi étant soit néant soit un noeud existant)
 * TODO - extraire sous-arbre étant donné {RACINE, {FEUILLES}}
 * TODO - comparer deux arbres
 * TODO - pour deux arbres comparables, déterminer la rotation du second pour matcher le premier
 */
399
400
401
402

/*
 * pedigree_type: implements all facilities to compute proper geno_matrices for any pedigree, including reentrant individuals.
 */
403
struct pedigree_type {
404
405
406
407
408
    /*
     * original data
     */
    std::vector<pedigree_item> items;

409
410
411
    /*
     * pedigree tree implementation
     */
Damien Leroux's avatar
Damien Leroux committed
412
    pedigree_tree_type tree;
413
414
415
416
417

    /*
     * pedigree tree metadata
     */
    typedef size_t geno_matrix_index_type;
Damien Leroux's avatar
Damien Leroux committed
418
    typedef int individual_index_type;
419
420
421
    std::vector<std::shared_ptr<geno_matrix>> generations;
    std::vector<geno_matrix_index_type> node_generations;
    std::map<individual_index_type, char> ancestor_letters;
422
    std::map<char, std::string> ancestor_names;
423
    std::map<geno_matrix_index_type, std::string> generation_names;
424
    std::map<int, int> m_id;
425
    /*std::vector<VectorLC> LC;*/
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440

    /*
     * geno_matrix cache to avoid recomputing identical generations
     */
    std::map<geno_matrix_index_type, geno_matrix_index_type> cache_gamete;
    std::map<std::pair<geno_matrix_index_type, geno_matrix_index_type>, geno_matrix_index_type> cache_geno;

    /*
     * geno_matrix database
     */

    std::map<std::string, std::set<geno_matrix_index_type>> geno_matrix_by_generation_name;
    std::map<std::string, std::vector<individual_index_type>> individuals_by_generation_name;
    std::map<individual_index_type, const std::string*> generation_name_by_individual;

441
442
443
444
445
446
447
448
449
450
451
    /*
     * overlump control
     */

    size_t max_states;

    /*
     * BN metadata
     */
    size_t n_alleles;

452
453
454
455
456
457
458
459
460
461
462
463
464
    /*
     * Metadata for XML output and recreating command line
     */
    std::string filename;

    /*
     * Actual output: LC and factors for bayesian network
     */
    std::vector<std::vector<gencomb_type>> LC;
    std::vector<std::vector<std::map<bn_label_type, genotype_comb_type>>> factor_messages;
    std::vector<std::vector<size_t>> individuals_in_factors;
    /* i-th element means the i-th variable receives a message through this factor from variables in genotype_comb_type:keys */

465
466
467
468
    /*
     * default ctor
     */
    pedigree_type()
469
        : tree(), node_generations(), ancestor_letters(), ancestor_names(), generation_names(), m_id(),
Damien Leroux's avatar
Damien Leroux committed
470
471
          cache_gamete(), cache_geno(),
          max_states(NONE),
472
473
474
475
          n_alleles(1),
          filename("<none>"),
          LC(),
          factor_messages()
476
477
478
479
480
481
482
483
484
    {
        __init();
    }

    void __init()
    {
        generations.emplace_back();
    }

Damien Leroux's avatar
Damien Leroux committed
485
#if 0
486
    /*
487
     * prealloc ctor
488
489
490
     */
    pedigree_type(size_t n_ind)
    {
491
        n_alleles = 1;
Damien Leroux's avatar
Damien Leroux committed
492
        max_states = NONE;
493
494
495
496
497
498
        nodes.reserve(3 * n_ind);
        /*ind_generations.reserve(n_ind);*/
        ind_number_to_node_number.reserve(n_ind);
        __init();
    }

Damien Leroux's avatar
Damien Leroux committed
499
500
    size_t last_node_index() const { return tree.size() - 1; }
#endif
501

502
    individual_index_type spawn_gamete(const std::string&, int parent_node)
503
    {
Damien Leroux's avatar
Damien Leroux committed
504
        int n = tree.add_node(parent_node);
505
        node_generations.emplace_back(node_generations[parent_node]);
Damien Leroux's avatar
Damien Leroux committed
506
507
508
509
510
        /*MSG_DEBUG_INDENT_EXPR("[compute " << gamete_name << " gamete] ");*/
        /*compute_generation(n);*/
        /*compute_LC(n);*/
        /*MSG_DEBUG_DEDENT;*/
        return n;
511
512
    }

513
    individual_index_type spawn(const std::string& generation_name, std::initializer_list<individual_index_type> parents)
514
    {
Damien Leroux's avatar
Damien Leroux committed
515
        individual_index_type ind = tree.next_ind_idx();
516
517
518
519
520
        switch (parents.size()) {
            case 0: /* ancestor */
                {
                    MSG_DEBUG_INDENT_EXPR("[compute gen #" << ind << "] ");
                    MSG_DEBUG("ANCESTOR");
Damien Leroux's avatar
Damien Leroux committed
521
522
                    int n = tree.add_node();
                    MSG_DEBUG("node=" << n << " ind=" << ind);
523
                    compute_generation(generation_name, n);
524
                    compute_LC(n);
525
526
527
528
529
530
531
                    MSG_DEBUG_DEDENT;
                }
                break;
            case 1: /* doubling */
                {
                    MSG_DEBUG_INDENT_EXPR("[compute gen #" << ind << "] ");
                    MSG_DEBUG("DOUBLING");
Damien Leroux's avatar
Damien Leroux committed
532
533
534
535
                    individual_index_type p1 = *parents.begin();
                    int g1 = spawn_gamete("M", tree.ind2node(p1));
                    int n = tree.add_node(g1, g1);
                    MSG_DEBUG("node=" << n << " ind=" << ind);
536
537
                    compute_generation(generation_name, n);
                    compute_LC(n);
Damien Leroux's avatar
Damien Leroux committed
538
                    /*compute_data_for_bn(n);*/
539
540
541
542
543
544
545
546
547
                    MSG_DEBUG_DEDENT;
                }
                break;
            case 2: /* crossing & selfing */
                {
                    MSG_DEBUG_INDENT_EXPR("[compute gen #" << ind << "] ");
                    MSG_DEBUG("CROSSING/SELFING");
                    auto i = parents.begin();
                    individual_index_type p1 = *i++;
Damien Leroux's avatar
Damien Leroux committed
548
                    int n1 = tree.ind2node(p1);
549
                    individual_index_type p2 = *i;
Damien Leroux's avatar
Damien Leroux committed
550
                    int n2 = tree.ind2node(p2);
551
                    /*MSG_DEBUG("p1=" << p1 << " p2=" << p2 << " n1=" << n1 << " n2=" << n2);*/
Damien Leroux's avatar
Damien Leroux committed
552
553
554
555
                    int g1 = spawn_gamete("M", n1);
                    int g2 = spawn_gamete("F", n2);
                    int n = tree.add_node(g1, g2);
                    MSG_DEBUG("node=" << n << " ind=" << ind);
556
557
                    compute_generation(generation_name, n);
                    compute_LC(n);
Damien Leroux's avatar
Damien Leroux committed
558
                    /*compute_data_for_bn(n);*/
559
560
561
562
563
564
565
566
                    MSG_DEBUG_DEDENT;
                }
                break;
            default:;
        };
        return ind;
    }

567
568
569
570
    /*individual_index_type crossing(std::string& generation_name, individual_index_type p1, individual_index_type p2) { return spawn(generation_name, {p1, p2}); }*/
    /*individual_index_type selfing(std::string& generation_name, individual_index_type p1) { return spawn(generation_name, {p1, p1}); }*/
    /*individual_index_type dh(std::string& generation_name, individual_index_type p1) { return spawn(generation_name, {p1}); }*/
    /*individual_index_type ancestor(std::string& generation_name) { return spawn(generation_name, {}); }*/
571
572
573

    individual_index_type fill_db(const std::string& name, individual_index_type ind)
    {
574
        geno_matrix_by_generation_name[name].insert(get_gen_index(ind));
575
576
577
578
579
580
        individuals_by_generation_name[name].push_back(ind);
        auto it = individuals_by_generation_name.find(name);
        generation_name_by_individual[ind] = &it->first;
        return ind;
    }

581
582
583
584
    individual_index_type crossing(const std::string& name, individual_index_type p1, individual_index_type p2) { return fill_db(name, spawn(name, {p1, p2})); }
    individual_index_type selfing(const std::string& name, individual_index_type p1) { return fill_db(name, spawn(name, {p1, p1})); }
    individual_index_type dh(const std::string& name, individual_index_type p1) { return fill_db(name, spawn(name, {p1})); }
    individual_index_type ancestor(const std::string& name) { return fill_db(name, spawn(name, {})); }
585

Damien Leroux's avatar
Damien Leroux committed
586
    void propagate_symmetries(int n, geno_matrix& gen)
587
    {
588
        MSG_DEBUG_INDENT_EXPR("[propagate symmetries #" << n << "] ");
Damien Leroux's avatar
Damien Leroux committed
589
590
591
592
593
594
595
        std::vector<int> in, out;
        auto expr = tree.extract_expression(n, in, out);
        std::vector<pedigree_tree_type> input_trees;
        input_trees.reserve(in.size());
        for (int t: in) {
            input_trees.emplace_back(tree.extract_subtree(t));
        }
596
597
598
599
600
601
602
603
604
605
606
        /*auto recompute = tree.get_deep_recompute_vec(n);*/
        /*MSG_DEBUG("RECOMPUTE: " << recompute);*/
        auto get_lumper
            = [&, this] (int node) -> MatrixXb
            {
                /*if (tree[node].is_gamete() || recompute[node]) {*/
                    /*MSG_DEBUG("NIL lumper for node #" << node << " because" << (recompute[node] && tree[node].is_gamete() ? " recompute flag is set and it is a gamete" : tree[node].is_gamete() ? " it is a gamete" : " recompute flag is set"));*/
                    /*return {};*/
                /*}*/
                return get_node_gen(node)->collect.cast<bool>();
            };
Damien Leroux's avatar
Damien Leroux committed
607
        symmetry_propagator sp(expr);
608
        MSG_DEBUG_INDENT_EXPR("[SYMMETRIES] ");
Damien Leroux's avatar
Damien Leroux committed
609
610
611
612
613
        gen.symmetries = sp.compute_propagated_symmetries(
                [&] (int i) -> const pedigree_tree_type& { MSG_DEBUG("request subtree for input #" << i); return input_trees[i]; },
                [&] (int node) -> const symmetry_group_type& { MSG_DEBUG("request symmetries for node #" << node); return get_node_gen(node)->symmetries; },
                [&] (int node) -> const symmetry_group_type& { MSG_DEBUG("request latent symmetries for node #" << node); return get_node_gen(node)->latent_symmetries; },
                gen.labels, gen.inf_mat,
614
                get_lumper
Damien Leroux's avatar
Damien Leroux committed
615
                );
616
617
        MSG_DEBUG_DEDENT;
        MSG_DEBUG_INDENT_EXPR("[LATENT SYMMETRIES] ");
Damien Leroux's avatar
Damien Leroux committed
618
619
620
621
622
        auto temp = sp.compute_propagated_latent_symmetries(
                [&] (int i) -> const pedigree_tree_type& { MSG_DEBUG("request subtree for input #" << i); return input_trees[i]; },
                [&] (int node) -> const symmetry_group_type& { MSG_DEBUG("request symmetries for node #" << node); return get_node_gen(node)->symmetries; },
                [&] (int node) -> const symmetry_group_type& { MSG_DEBUG("request latent symmetries for node #" << node); return get_node_gen(node)->latent_symmetries; },
                gen.labels, gen.inf_mat,
623
                get_lumper
Damien Leroux's avatar
Damien Leroux committed
624
                );
625
        MSG_DEBUG_DEDENT;
Damien Leroux's avatar
Damien Leroux committed
626
627
        MSG_DEBUG(temp);
        gen.latent_symmetries = temp - gen.symmetries;
628
        MSG_DEBUG_INDENT_EXPR("[AFTER SYMMETRY PROPAGATION] ");
Damien Leroux's avatar
Damien Leroux committed
629
        MSG_DEBUG(gen);
630
        MSG_DEBUG_DEDENT;
631
        MSG_DEBUG_DEDENT;
632
633
    }

634
    void compute_generation(const std::string& generation_name, int n)
635
    {
Damien Leroux's avatar
Damien Leroux committed
636
        MSG_DEBUG("Computing generation for node " << tree.make_node_label(n));
637
        /*MSG_DEBUG(render_tree());*/
638

Damien Leroux's avatar
Damien Leroux committed
639
640
        int np1 = tree.get_p1(n);
        int np2 = tree.get_p2(n);
641
642
643
        node_generations.emplace_back(generations.size());
        geno_matrix new_gen;
        geno_matrix_index_type* cached_gen = NULL;
Damien Leroux's avatar
Damien Leroux committed
644
        if (np1 == NONE && np2 == NONE) {
645
646
647
            /* ancestor */
            char l = 'a' + ancestor_letters.size();
            ancestor_letters[n] = l;
648
            ancestor_names[l] = generation_name;
649
            new_gen = ancestor_matrix(generation_name, l);
Damien Leroux's avatar
Damien Leroux committed
650
        } else if (np2 == NONE) {
651
652
            /* gamete */
            auto gp = node_generations[np1];
Damien Leroux's avatar
Damien Leroux committed
653
            auto& cache = cache_gamete;
654
655
656
657
658
659
            cached_gen = &cache[gp];
            if (*cached_gen) {
                MSG_DEBUG("GENERATION HAS ALREADY BEEN COMPUTED");
                node_generations[n] = *cached_gen;
                return;
            }
Damien Leroux's avatar
Damien Leroux committed
660
            new_gen = kronecker(*generations[gp], gamete);
661
            new_gen.name = generation_name;
662
663
            /*MSG_DEBUG("TMP GAMETE GEN");*/
            /*MSG_DEBUG(new_gen);*/
664
        } else {
665
666
            /*auto ngp1 = node_generations[np1];*/
            /*auto ngp2 = node_generations[np2];*/
Damien Leroux's avatar
Damien Leroux committed
667
668
669
            /*auto gp1 = generations[ngp1];*/
            /*auto gp2 = generations[ngp2];*/
            MSG_DEBUG("Child of " << tree.node2ind(tree.get_p1(np1)) << " and " << tree.node2ind(tree.get_p1(np2)));
670

671
672
673
674
            /* use node's grandparents generations, not parents (no gamete generation) */
            size_t g1 = node_generations[tree.get_p1(np1)];
            size_t g2 = node_generations[tree.get_p1(np2)];
            cached_gen = &cache_geno[{g1, g2}];
675
676
677
678
679
680
            if (*cached_gen) {
                MSG_DEBUG("GENERATION HAS ALREADY BEEN COMPUTED");
                node_generations[n] = *cached_gen;
                return;
            }

Damien Leroux's avatar
Damien Leroux committed
681
682
            const auto& recompute = tree.get_recompute_vec(n);

683
684
685
            std::vector<bool> visited_clear(recompute.size(), false);
            std::vector<bool> visited;

686
687
            new_gen.name = generation_name;

688
689
690
691
692
693
            visited = visited_clear;
            MSG_DEBUG("COMPUTING INF_MAT");
            new_gen.inf_mat = eval(n, &geno_matrix::inf_mat, &pedigree_type::kron_d, recompute, visited);
            MSG_DEBUG(MATRIX_SIZE(new_gen.inf_mat));
            visited = visited_clear;
            MSG_DEBUG("COMPUTING DIAG");
694
            new_gen.diag = eval(n, &geno_matrix::diag, &pedigree_type::kron_d_diag, recompute, visited);
695
696
697
698
699
700
701
702
703
704
705
            visited = visited_clear;
            MSG_DEBUG("COMPUTING STAT_DIST");
            new_gen.stat_dist = eval(n, &geno_matrix::stat_dist, &pedigree_type::kron, recompute, visited);
            visited = visited_clear;
            MSG_DEBUG("COMPUTING P");
            new_gen.p = eval(n, &geno_matrix::p, &pedigree_type::kron, recompute, visited);
            visited = visited_clear;
            MSG_DEBUG("COMPUTING P_INV");
            new_gen.p_inv = eval(n, &geno_matrix::p_inv, &pedigree_type::kron_d, recompute, visited);
            /*new_gen.labels = eval_labels(n, recompute, visited_clear);*/
            new_gen.labels = eval_vector(n, recompute, &pedigree_type::get_labels, reentrant_label);
Damien Leroux's avatar
Damien Leroux committed
706
            new_gen.variant = (tree[n].is_genotype()
707
                               ? Geno
Damien Leroux's avatar
Damien Leroux committed
708
                               : Gamete);
709
710
711
712
713
714
715
716
            new_gen.dispatch = new_gen.collect = MatrixXd::Identity(new_gen.diag.size(), new_gen.diag.size());
            MSG_DEBUG(MATRIX_SIZE(new_gen.inf_mat));
            MSG_DEBUG(MATRIX_SIZE(new_gen.p));
            MSG_DEBUG(MATRIX_SIZE(new_gen.p_inv));
            MSG_DEBUG(MATRIX_SIZE(new_gen.diag));
            MSG_DEBUG("new_gen.labels.size()=" << new_gen.labels.size());
            MSG_QUEUE_FLUSH();

Damien Leroux's avatar
Damien Leroux committed
717
            /*if (!(ind_number_to_node_number.size() == 9 && tree.size() == 23)) {*/
718
                /*MSG_DEBUG("PROPAGATING SYMMETRIES");*/
719
                /*propagate_symmetries(new_gen, recompute, n);*/
720
721
                /*study_expression_symmetries(new_gen);*/
                /*complete_symmetries(new_gen);*/
722
                /*MSG_DEBUG("COMPUTING LATENT SYMMETRY");*/
723
            /*}*/
724
725
726
        }
        node_generations[n] = generations.size();
        generations.emplace_back(new geno_matrix());
727
728
        /*MSG_DEBUG("BEFORE LUMPING");*/
        /*MSG_DEBUG(new_gen);*/
729
        *generations.back() = lump(new_gen, max_states);
730
731
732
        /*if (tree[n].is_crossing()) {*/
            /*propagate_symmetries(n, *generations.back());*/
        /*} else if (tree[n].is_ancestor()) {*/
Damien Leroux's avatar
Damien Leroux committed
733
            generations.back()->symmetries = symmetry_group_type(generations.back()->labels);
734
        /*}*/
735
736
737
738
739
        /**node_generations[n] = lump(new_gen);*/
        if (cached_gen) {
            *cached_gen = node_generations[n];
        }
        MSG_DEBUG("DONE COMPUTING GENERATION FOR NODE #" << n);
740
        MSG_DEBUG_INDENT_EXPR("[RESULT " << tree.make_node_label(n) << " gen#" << node_generations[n] << "] ");
741
742
        MSG_DEBUG((*generations.back()));
        MSG_DEBUG_DEDENT;
743
        /*MSG_DEBUG((*generations[node_generations[n]]));*/
744
745
746
        MSG_DEBUG("=========================================================================");
    }

747
    std::map<genotype_comb_type::key_list, double> GLC_norm_factors(const std::vector<genotype_comb_type>& expanded)
748
    {
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
        std::map<genotype_comb_type::key_list, double> ret;
        for (const auto& e: expanded) {
            const auto& elem = e.m_combination.front();  /* all LC are singletons in the expanded vector */
            for (const auto& k: elem.keys) {
                auto sub_k = elem.keys - k;
                ret[sub_k] += elem.coef;
            }
        }
        return ret;
    }

    void compute_data_for_bn(int n)
    {
        compute_LC(n);

        factor_messages.resize(n + 1);
        for (auto& dest_f: compute_factors(n, true)) {
            factor_messages[dest_f.first].emplace_back(std::move(dest_f.second));
767
        }
768
769
770
771
772
773
774
775
776
777
778
779
780
        for (auto& dest_f: compute_factors(n, false)) {
            factor_messages[dest_f.first].emplace_back(std::move(dest_f.second));
        }
        individuals_in_factors.resize(n + 1);
        std::vector<size_t>& iif = individuals_in_factors.back();
        std::vector<int> in;
        std::vector<int> out;
        auto expr = tree.extract_expression(n, in, out);
        iif.reserve(expr.m_nodes.size());
        for (size_t i = 0; i < expr.m_nodes.size(); ++i) {
            if (expr.m_nodes[i].is_genotype()) {
                iif.push_back(expr.original_node_number(i));
            }
781
        }
782
783
        MSG_DEBUG("INDIVIDUALS IN FACTOR: " << iif);
    }
784

785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
    void
        compute_LC(int n)
        {
            int p1 = tree.get_p1(n);
            std::vector<gencomb_type> lc;
            const geno_matrix& m = *generations[node_generations[n]];
            if (p1 == NONE) {
                lc.emplace_back(1.);
            } else {
                lc = eval_vector(tree.size() - 1, tree.get_recompute_vec(n), &pedigree_type::get_LC, reentrant_LC);
            }
            VectorLC tmp(lc.size()), lumped;
            for (size_t i = 0; i < lc.size(); ++i) {
                tmp(i) = lc[i];
            }
            lumped = m.collect.cast<gencomb_type>() * tmp;
            LC.resize(n + 1);
            LC.back().assign(lumped.data(), lumped.data() + lumped.size());
            MSG_DEBUG("Computed new LC:");
            MSG_DEBUG("" << LC.back());
805
806
        }

807
808
809
810
811
    std::map<size_t, std::map<bn_label_type, genotype_comb_type>>
        compute_factors(int n, bool up)
        {
            std::vector<genotype_comb_type> glc;
            std::vector<bn_label_type> bn_labels;
812

813
814
815
            if (tree.get_p1(n) == NONE) {
                return {};
            }
816

817
818
819
820
821
822
823
824
825
826
            if (up) {
                std::vector<bool> recompute(n + 1, false);
                recompute[n] = true;
                recompute[tree.get_p1(n)] = true;
                recompute[tree.get_p2(n)] = true;
                glc = eval_vector(tree.size() - 1, recompute, &pedigree_type::get_GLC, reentrant_GLC);
                bn_labels = eval_vector(tree.size() - 1, recompute, &pedigree_type::get_bn_labels, reentrant_bn_label);
            } else {
                glc = eval_vector(tree.size() - 1, tree.get_recompute_vec(n), &pedigree_type::get_GLC, reentrant_GLC);
                bn_labels = eval_vector(tree.size() - 1, tree.get_recompute_vec(n), &pedigree_type::get_bn_labels, reentrant_bn_label);
827
828
            }

829
830
831
832
833
834
835
            MSG_DEBUG("glc.size = " << glc.size());
            MSG_DEBUG("bn_labels.size = " << bn_labels.size());

            std::map<bn_label_type, genotype_comb_type> glc_map;
            for (size_t i = 0; i < bn_labels.size(); ++i) {
                glc_map[bn_labels[i]] += glc[i];
            }
836

837
838
839
840
841
842
843
844
845
846
            MSG_DEBUG("Corresponding GLC:");
            std::vector<genotype_comb_type> expanded_glc;
            for (const auto& kv: glc_map) {
                MSG_DEBUG("" << kv.first << " = " << kv.second);
                for (const auto& elem: kv.second.m_combination) {
                    expanded_glc.emplace_back();
                    expanded_glc.back().m_combination.emplace_back(elem);
                    auto& keys = expanded_glc.back().m_combination.back().keys.keys;
                    keys.emplace_back(tree.size() - 1, kv.first);
                    std::sort(keys.begin(), keys.end());
847
                }
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
            }

            MSG_DEBUG("Sparse GLC coefs:");
            for (const auto& g: expanded_glc) {
                MSG_DEBUG("" << g);
            }

            auto glc_norm_factors = GLC_norm_factors(expanded_glc);
            for (const auto& kn: glc_norm_factors) {
                MSG_DEBUG('|' << kn.first << "| = " << kn.second);
            }

            std::map<size_t, std::map<bn_label_type, genotype_comb_type>> messages;
            size_t n_nodes = expanded_glc.front().m_combination.front().keys.keys.size();
            size_t first, last;
            if (up) {
                first = 0;
                last = n_nodes - 2;
            } else {
                first = 0;
                last = n_nodes - 1;
            }
            for (const auto& lc: expanded_glc) {
                const auto& elems = lc.m_combination.front();
                const auto& keys = elems.keys.keys;
                for (size_t ni = first; ni <= last; ++ni) {
                    genotype_comb_type tmp;
                    tmp.m_combination.emplace_back(elems.coef);
                    auto& msg_keys = tmp.m_combination.back().keys.keys;
                    size_t i;
                    for (i = 0; i < ni; ++i) {
                        msg_keys.emplace_back(keys[i]);
                    }
                    for (++i; i < n_nodes; ++i) {
                        msg_keys.emplace_back(keys[i]);
                    }
                    tmp.m_combination.front().coef /= glc_norm_factors[tmp.m_combination.back().keys];
                    messages[keys[ni].parent][keys[ni].state] += tmp;
886
887
888
                }
            }

889
890
891
892
893
            for (const auto& kv1: messages) {
                MSG_DEBUG("MESSAGE TOWARDS #" << kv1.first);
                for (const auto& kv2: kv1.second) {
                    MSG_DEBUG("" << kv2.first << " = " << kv2.second);
                }
894
            }
895
896

            return messages;
897
898
899
900
901
902
903
904
905
906
907
908
909
910
        }


    static MatrixXd kron(const MatrixXd& m1, const MatrixXd& m2)
    {
        return kroneckerProduct(m1, m2);
    }

    static MatrixXd kron_d(const MatrixXd& m1, const MatrixXd& m2)
    {
        return kroneckerProduct(m1, MatrixXd::Identity(m2.rows(), m2.cols()))
             + kroneckerProduct(MatrixXd::Identity(m1.rows(), m1.cols()), m2);
    }

911
912
913
914
915
916
    static MatrixXd kron_d_diag(const MatrixXd& m1, const MatrixXd& m2)
    {
        return kroneckerProduct(m1, MatrixXd::Ones(m2.rows(), m2.cols()))
             + kroneckerProduct(MatrixXd::Ones(m1.rows(), m1.cols()), m2);
    }

917
918
    char ancestor_letter(size_t a) const
    {
Damien Leroux's avatar
Damien Leroux committed
919
        auto i = ancestor_letters.find(tree.ind2node(a));
920
921
922
        return i == ancestor_letters.end() ? 0 : i->second;
    }

923
    geno_matrix_index_type get_gen_index(size_t ind) const
924
    {
Damien Leroux's avatar
Damien Leroux committed
925
        return node_generations[tree.ind2node(ind)];
926
927
928
929
930
931
932
    }

    const std::shared_ptr<geno_matrix> get_gen(size_t ind) const
    {
        return generations[get_gen_index(ind)];
    }

933
934
935
936
937
938
    int
        ind2id(int i) const
        {
            return items[i - 1].id;
        }

939
    const geno_matrix&
Damien Leroux's avatar
Damien Leroux committed
940
        get_geno_matrix_by_individual(size_t ind) const { return *generations[node_generations[tree.ind2node(ind)]]; }
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
    const std::set<geno_matrix_index_type>&
        get_geno_matrices_by_name(const std::string& name) const { return geno_matrix_by_generation_name.find(name)->second; }
    const std::string&
        get_generation_name_by_individual(size_t ind) const { return *generation_name_by_individual.find(ind)->second; }

    const std::shared_ptr<geno_matrix> get_node_gen(size_t node) const
    {
        return generations[node_generations[node]];
    }

    typedef MatrixXd (* KronFunc) (const MatrixXd&, const MatrixXd&);

    template <typename FIELD_TYPE>
    FIELD_TYPE eval(size_t node, FIELD_TYPE geno_matrix::* field, KronFunc func, const std::vector<bool>& recompute, std::vector<bool>& visited) const
    {
956
        scoped_indent _;
957
        /*MSG_DEBUG("eval node " << node);*/
958
        if (visited[node]) {
959
            /*MSG_DEBUG("already visited => 1");*/
960
            return make_one<FIELD_TYPE>::_(func == kron_d || func == kron_d_diag);
961
962
        }
        visited[node] = true;
Damien Leroux's avatar
Damien Leroux committed
963
        if (recompute[node] || tree[node].is_gamete()) {
964
            FIELD_TYPE ret, m1, m2;
Damien Leroux's avatar
Damien Leroux committed
965
966
967
968
969
970
971
972
            if (tree[node].is_genotype()) {
                m1 = eval(tree.get_p1(node), field, func, recompute, visited);
                m2 = eval(tree.get_p2(node), field, func, recompute, visited);
                ret = func(m1, m2);
            } else {
                m1 = eval(tree.get_p1(node), field, func, recompute, visited);
                ret = func(m1, gamete.*field);
            }
973
974
            return ret;
        } else {
975
976
            /*MSG_DEBUG("using actual matrix");*/
            /*MSG_DEBUG((*generations[node_generations[node]]).*field);*/
977
978
979
980
981
982
983
984
985
986
987
988
            return (*generations[node_generations[node]]).*field;
        }
        return {};
    }

    template <typename VALUE_TYPE>
    struct vector_iterator {
        typedef std::vector<VALUE_TYPE> vector_type;
        vector_type data;
        typename vector_type::const_iterator begin, end, cur;
        vector_iterator() : data(), begin(data.begin()), end(data.end()), cur(data.begin()) {}
        vector_iterator(const vector_type& l) : data(l), begin(data.begin()), end(data.end()), cur(data.begin()) {}
989
        void reset() { cur = begin = data.begin(); end = data.end(); }
990
991
992
993
994
        void start() { cur = begin; }
        bool next() { if (at_end()) return true; return ++cur == end; }
        bool at_end() const { return cur == end; }
        size_t size() const { return end - begin; }
        const VALUE_TYPE& operator * () const { return *cur; }
995
        const VALUE_TYPE* operator -> () const { return &*cur; }
996
997
998
999
1000
    };

    typedef vector_iterator<label_type> label_iterator;
    typedef vector_iterator<symmetry_table_type> symmetry_iterator;

For faster browsing, not all history is shown. View entire blame