-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathPCfrSolver.cpp
1187 lines (1042 loc) · 48.3 KB
/
PCfrSolver.cpp
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
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
660
661
662
663
664
665
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
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
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
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
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
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
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
//
// Created by Xuefeng Huang on 2020/1/31.
//
#include <include/solver/BestResponse.h>
#include "include/solver/PCfrSolver.h"
#include <QtCore>
#include <QObject>
#include <QTranslator>
//#define DEBUG;
PCfrSolver::~PCfrSolver(){
//cout << "Pcfr destroyed" << endl;
}
PCfrSolver::PCfrSolver(shared_ptr<GameTree> tree, vector<PrivateCards> range1, vector<PrivateCards> range2,
vector<int> initial_board, shared_ptr<Compairer> compairer, Deck deck, int iteration_number, bool debug,
int print_interval, string logfile, string trainer, Solver::MonteCarolAlg monteCarolAlg,int warmup,float accuracy,bool use_isomorphism,int use_halffloats,int num_threads) :Solver(tree){
this->initial_board = initial_board;
this->initial_board_long = Card::boardInts2long(initial_board);
this->logfile = logfile;
this->trainer = trainer;
this->warmup = warmup;
range1 = this->noDuplicateRange(range1,initial_board_long);
range2 = this->noDuplicateRange(range2,initial_board_long);
this->range1 = range1;
this->range2 = range2;
this->player_number = 2;
this->ranges = vector<vector<PrivateCards>>(this->player_number);
this->ranges[0] = range1;
this->ranges[1] = range2;
this->compairer = compairer;
this->deck = deck;
this->use_isomorphism = use_isomorphism;
this->use_halffloats = use_halffloats;
this->rrm = RiverRangeManager(compairer);
this->iteration_number = iteration_number;
vector<vector<PrivateCards>> private_cards(this->player_number);
private_cards[0] = range1;
private_cards[1] = range2;
pcm = PrivateCardsManager(private_cards,this->player_number,Card::boardInts2long(this->initial_board));
this->debug = debug;
this->print_interval = print_interval;
this->monteCarolAlg = monteCarolAlg;
this->accuracy = accuracy;
if(num_threads == -1){
num_threads = omp_get_num_procs();
}
qDebug().noquote() << QString::fromStdString(tfm::format(QObject::tr("Using %s threads").toStdString().c_str(),num_threads));
this->num_threads = num_threads;
this->distributing_task = false;
omp_set_num_threads(this->num_threads);
setTrainable(this->tree->getRoot());
this->root_round = this->tree->getRoot()->getRound();
if(this->root_round == GameTreeNode::GameRound::PREFLOP){
this->split_round = GameTreeNode::GameRound::FLOP;
}else if(this->root_round == GameTreeNode::GameRound::FLOP){
this->split_round = GameTreeNode::GameRound::TURN;
}else if(this->root_round == GameTreeNode::GameRound::TURN){
this->split_round = GameTreeNode::GameRound::RIVER;
}else{
// do not use multithread in river, really not necessary
this->split_round = GameTreeNode::GameRound::PREFLOP;
}
}
const vector<PrivateCards> &PCfrSolver::playerHands(int player) {
if(player == 0){
return range1;
}else if (player == 1){
return range2;
}else{
throw runtime_error("player not found");
}
}
vector<vector<float>> PCfrSolver::getReachProbs() {
vector<vector<float>> retval(this->player_number);
for(int player = 0;player < this->player_number;player ++){
vector<PrivateCards> player_cards = this->playerHands(player);
vector<float> reach_prob(player_cards.size());
for(int i = 0;i < player_cards.size();i ++){
reach_prob[i] = player_cards[i].weight;
}
retval[player] = reach_prob;
}
return retval;
}
vector<PrivateCards>
PCfrSolver::noDuplicateRange(const vector<PrivateCards> &private_range, uint64_t board_long) {
vector<PrivateCards> range_array;
unordered_map<int,bool> rangekv;
for(PrivateCards one_range:private_range){
if(rangekv.find(one_range.hashCode()) != rangekv.end())
throw runtime_error(tfm::format("duplicated key %s",one_range.toString()));
rangekv[one_range.hashCode()] = true;
uint64_t hand_long = Card::boardInts2long(one_range.get_hands());
if(!Card::boardsHasIntercept(hand_long,board_long)){
range_array.push_back(one_range);
}
}
return range_array;
}
void PCfrSolver::setTrainable(shared_ptr<GameTreeNode> root) {
if(root->getType() == GameTreeNode::ACTION){
shared_ptr<ActionNode> action_node = std::dynamic_pointer_cast<ActionNode>(root);
int player = action_node->getPlayer();
if(this->trainer == "cfr_plus"){
//vector<PrivateCards> player_privates = this->ranges[player];
//action_node->setTrainable(make_shared<CfrPlusTrainable>(action_node,player_privates));
throw runtime_error(tfm::format("trainer %s not supported",this->trainer));
}else if(this->trainer == "discounted_cfr"){
vector<PrivateCards>* player_privates = &this->ranges[player];
//action_node->setTrainable(make_shared<DiscountedCfrTrainable>(action_node,player_privates));
int num;
GameTreeNode::GameRound gr = this->tree->getRoot()->getRound();
int root_round = GameTreeNode::gameRound2int(gr);
int current_round = GameTreeNode::gameRound2int(root->getRound());
int gap = current_round - root_round;
if(gap == 2) {
num = this->deck.getCards().size() * this->deck.getCards().size() +
this->deck.getCards().size() + 1;
}else if(gap == 1) {
num = this->deck.getCards().size() + 1;
}else if(gap == 0) {
num = 1;
}else{
throw runtime_error("gap not understand");
}
action_node->setTrainable(vector<shared_ptr<Trainable>>(num),player_privates);
}else{
throw runtime_error(tfm::format("trainer %s not found",this->trainer));
}
vector<shared_ptr<GameTreeNode>> childrens = action_node->getChildrens();
for(shared_ptr<GameTreeNode> one_child:childrens) setTrainable(one_child);
}else if(root->getType() == GameTreeNode::CHANCE) {
shared_ptr<ChanceNode> chance_node = std::dynamic_pointer_cast<ChanceNode>(root);
shared_ptr<GameTreeNode> children = chance_node->getChildren();
setTrainable(children);
}
else if(root->getType() == GameTreeNode::TERMINAL){
}else if(root->getType() == GameTreeNode::SHOWDOWN){
}
}
vector<int> PCfrSolver::getAllAbstractionDeal(int deal){
vector<int> all_deal;
int card_num = this->deck.getCards().size();
if(deal == 0){
all_deal.push_back(deal);
} else if (deal > 0 && deal <= card_num){
int origin_deal = int((deal - 1) / 4) * 4;
for(int i = 0;i < 4;i ++){
int one_card = origin_deal + i + 1;
Card *first_card = const_cast<Card *>(&(this->deck.getCards()[origin_deal + i]));
uint64_t first_long = Card::boardInt2long(
first_card->getCardInt());
if (Card::boardsHasIntercept(first_long, this->initial_board_long))continue;
all_deal.push_back(one_card);
}
} else{
//cout << "______________________" << endl;
int c_deal = deal - (1 + card_num);
int first_deal = int((c_deal / card_num) / 4) * 4;
int second_deal = int((c_deal % card_num) / 4) * 4;
for(int i = 0;i < 4;i ++) {
for(int j = 0;j < 4;j ++) {
if(first_deal == second_deal && i == j) continue;
Card *first_card = const_cast<Card *>(&(this->deck.getCards()[first_deal + i]));
uint64_t first_long = Card::boardInt2long(
first_card->getCardInt());
if (Card::boardsHasIntercept(first_long, this->initial_board_long))continue;
Card *second_card = const_cast<Card *>(&(this->deck.getCards()[second_deal + j]));
uint64_t second_long = Card::boardInt2long(
second_card->getCardInt());
if (Card::boardsHasIntercept(second_long, this->initial_board_long))continue;
int one_card = card_num * (first_deal + i) + (second_deal + j) + 1 + card_num;
//cout << ";" << this->deck.getCards()[first_deal + i].toString() << "," << this->deck.getCards()[second_deal + j].toString();
all_deal.push_back(one_card);
}
}
//cout << endl;
}
return all_deal;
}
vector<float> PCfrSolver::cfr(int player, shared_ptr<GameTreeNode> node, const vector<float> &reach_probs, int iter,
uint64_t current_board,int deal) {
switch(node->getType()) {
case GameTreeNode::ACTION: {
shared_ptr<ActionNode> action_node = std::dynamic_pointer_cast<ActionNode>(node);
return actionUtility(player, action_node, reach_probs, iter, current_board,deal);
}case GameTreeNode::SHOWDOWN: {
shared_ptr<ShowdownNode> showdown_node = std::dynamic_pointer_cast<ShowdownNode>(node);
return showdownUtility(player, showdown_node, reach_probs, iter, current_board,deal);
}case GameTreeNode::TERMINAL: {
shared_ptr<TerminalNode> terminal_node = std::dynamic_pointer_cast<TerminalNode>(node);
return terminalUtility(player, terminal_node, reach_probs, iter, current_board,deal);
}case GameTreeNode::CHANCE: {
shared_ptr<ChanceNode> chance_node = std::dynamic_pointer_cast<ChanceNode>(node);
return chanceUtility(player, chance_node, reach_probs, iter, current_board,deal);
}default:
throw runtime_error("node type unknown");
}
}
vector<float>
PCfrSolver::chanceUtility(int player, shared_ptr<ChanceNode> node, const vector<float> &reach_probs, int iter,
uint64_t current_board,int deal) {
vector<Card>& cards = this->deck.getCards();
//float[] cardWeights = getCardsWeights(player,reach_probs[1 - player],current_board);
int card_num = node->getCards().size();
if(card_num % 4 != 0) throw runtime_error("card num cannot round 4");
// 可能的发牌情况,2代表每个人的holecard是两张
int possible_deals = node->getCards().size() - Card::long2board(current_board).size() - 2;
int oppo = 1 - player;
//vector<float> chance_utility(reach_probs[player].size());
vector<float> chance_utility = vector<float>(this->ranges[player].size());
fill(chance_utility.begin(),chance_utility.end(),0);
int random_deal = 0;
if(this->monteCarolAlg==MonteCarolAlg::PUBLIC) {
if (this->round_deal[GameTreeNode::gameRound2int(node->getRound())] == -1) {
random_deal = random(1, possible_deals + 1 + 2);
this->round_deal[GameTreeNode::gameRound2int(node->getRound())] = random_deal;
} else {
random_deal = this->round_deal[GameTreeNode::gameRound2int(node->getRound())];
}
}
//vector<vector<vector<float>>> arr_new_reach_probs = vector<vector<vector<float>>>(node->getCards().size());
vector<vector<float>> results(node->getCards().size());
//fill(results.begin(),results.end(),nullptr);
vector<float> multiplier;
if(iter <= this->warmup){
multiplier = vector<float>(card_num);
fill(multiplier.begin(), multiplier.end(), 0);
for (int card_base = 0; card_base < card_num / 4; card_base++) {
int cardr = std::rand() % 4;
int card_target = card_base * 4 + cardr;
int multiplier_num = 0;
for (int i = 0; i < 4; i++) {
int i_card = card_base * 4 + i;
if (i == cardr) {
Card *one_card = const_cast<Card *>(&(node->getCards()[i_card]));
uint64_t card_long = Card::boardInt2long(
one_card->getCardInt());
if (!Card::boardsHasIntercept(card_long, current_board)) {
multiplier_num += 1;
}
} else {
Card *one_card = const_cast<Card *>(&(node->getCards()[i_card]));
uint64_t card_long = Card::boardInt2long(
one_card->getCardInt());
if (!Card::boardsHasIntercept(card_long, current_board)) {
multiplier_num += 1;
}
}
}
multiplier[card_target] = multiplier_num;
}
}
vector<int> valid_cards;
valid_cards.reserve(node->getCards().size());
for(int card = 0;card < node->getCards().size();card ++) {
shared_ptr<GameTreeNode> one_child = node->getChildren();
Card *one_card = const_cast<Card *>(&(node->getCards()[card]));
uint64_t card_long = Card::boardInt2long(one_card->getCardInt());//Card::boardCards2long(new Card[]{one_card});
if (Card::boardsHasIntercept(card_long, current_board)) continue;
if (iter <= this->warmup && multiplier[card] == 0) continue;
if (this->color_iso_offset[deal][one_card->getCardInt() % 4] < 0) continue;
valid_cards.push_back(card);
}
#pragma omp parallel for schedule(static)
for(int valid_ind = 0;valid_ind < valid_cards.size();valid_ind++) {
int card = valid_cards[valid_ind];
shared_ptr<GameTreeNode> one_child = node->getChildren();
Card *one_card = const_cast<Card *>(&(node->getCards()[card]));
uint64_t card_long = Card::boardInt2long(one_card->getCardInt());//Card::boardCards2long(new Card[]{one_card});
uint64_t new_board_long = current_board | card_long;
if (this->monteCarolAlg == MonteCarolAlg::PUBLIC) {
throw runtime_error("parallel solver don't support public monte carol");
}
//cout << "Card deal:" << one_card->toString() << endl;
vector<PrivateCards> &playerPrivateCard = (this->ranges[player]);
vector<PrivateCards> &oppoPrivateCards = (this->ranges[1 - player]);
vector<float> new_reach_probs = vector<float>(oppoPrivateCards.size());
#ifdef DEBUG
if (playerPrivateCard.size() != this->ranges[player].size()) throw runtime_error("length not match");
if (oppoPrivateCards.size() != this->ranges[1 - player].size()) throw runtime_error("length not match");
#endif
int player_hand_len = this->ranges[oppo].size();
for (int player_hand = 0; player_hand < player_hand_len; player_hand++) {
PrivateCards &one_private = this->ranges[oppo][player_hand];
uint64_t privateBoardLong = one_private.toBoardLong();
if (Card::boardsHasIntercept(card_long, privateBoardLong)) {
new_reach_probs[player_hand] = 0;
continue;
}
new_reach_probs[player_hand] = reach_probs[player_hand] / possible_deals;
}
#ifdef DEBUG
if (Card::boardsHasIntercept(current_board, card_long))
throw runtime_error("board has intercept with dealt card");
#endif
int new_deal;
if(deal == 0){
new_deal = card + 1;
} else if (deal > 0 && deal <= card_num){
int origin_deal = deal - 1;
#ifdef DEBUG
if(origin_deal == card) throw runtime_error("deal should not be equal");
#endif
new_deal = card_num * origin_deal + card;
new_deal += (1 + card_num);
} else{
throw runtime_error(tfm::format("deal out of range : %s ",deal));
}
if(this->distributing_task && node->getRound() == this->split_round) {
results[one_card->getNumberInDeckInt()] = vector<float>(this->ranges[player].size());
//TaskParams taskParams = TaskParams();
}else {
vector<float> child_utility = this->cfr(player, one_child, new_reach_probs, iter, new_board_long, new_deal);
results[one_card->getNumberInDeckInt()] = child_utility;
}
}
for(int card = 0;card < node->getCards().size();card ++) {
Card *one_card = const_cast<Card *>(&(node->getCards()[card]));
vector<float> child_utility;
int offset = this->color_iso_offset[deal][one_card->getCardInt() % 4];
if(offset < 0) {
int rank1 = one_card->getCardInt() % 4;
int rank2 = rank1 + offset;
#ifdef DEBUG
if(rank2 < 0) throw runtime_error("rank error");
#endif
child_utility = results[one_card->getNumberInDeckInt() + offset];
exchange_color(child_utility,this->pcm.getPreflopCards(player),rank1,rank2);
}else{
child_utility = results[one_card->getNumberInDeckInt()];
}
if(child_utility.empty())
continue;
#ifdef DEBUG
if(child_utility.size() != chance_utility.size()) throw runtime_error("length not match");
#endif
if(iter > this->warmup) {
for (int i = 0; i < child_utility.size(); i++)
chance_utility[i] += child_utility[i];
}else{
for (int i = 0; i < child_utility.size(); i++)
chance_utility[i] += child_utility[i] * multiplier[card];
}
}
#ifdef DEBUG
if(this->monteCarolAlg == MonteCarolAlg::PUBLIC) {
throw runtime_error("not possible");
}
if(chance_utility.size() != this->ranges[player].size()) {
throw runtime_error("size problems");
}
#endif
return chance_utility;
}
vector<float>
PCfrSolver::actionUtility(int player, shared_ptr<ActionNode> node, const vector<float> &reach_probs, int iter,
uint64_t current_board,int deal) {
int oppo = 1 - player;
const vector<PrivateCards>& node_player_private_cards = this->ranges[node->getPlayer()];
vector<float> payoffs = vector<float>(this->ranges[player].size());
fill(payoffs.begin(),payoffs.end(),0);
vector<shared_ptr<GameTreeNode>>& children = node->getChildrens();
vector<GameActions>& actions = node->getActions();
shared_ptr<Trainable> trainable;
/*
if(iter <= this->warmup){
vector<int> deals = this->getAllAbstractionDeal(deal);
trainable = node->getTrainable(deals[0]);
}else{
trainable = node->getTrainable(deal);
}
*/
trainable = node->getTrainable(deal,true,this->use_halffloats);
#ifdef DEBUG
if(trainable == nullptr){
throw runtime_error("null trainable");
}
#endif
const vector<float> current_strategy = trainable->getcurrentStrategy();
#ifdef DEBUG
if (current_strategy.size() != actions.size() * node_player_private_cards.size()) {
node->printHistory();
throw runtime_error(tfm::format(
"length not match %s - %s \n action size %s private_card size %s"
,current_strategy.size()
,actions.size() * node_player_private_cards.size()
,actions.size()
,node_player_private_cards.size()
));
}
#endif
//为了节省计算成本将action regret 存在一位数组而不是二维数组中，两个纬度分别是（该infoset有多少动作,该palyer有多少holecard）
vector<float> regrets(actions.size() * node_player_private_cards.size());
vector<vector<float>> all_action_utility(actions.size());
int node_player = node->getPlayer();
vector<vector<float>> results(actions.size());
for (int action_id = 0; action_id < actions.size(); action_id++) {
if (node_player != player) {
vector<float> new_reach_prob = vector<float>(reach_probs.size());
for (int hand_id = 0; hand_id < new_reach_prob.size(); hand_id++) {
float strategy_prob = current_strategy[hand_id + action_id * node_player_private_cards.size()];
new_reach_prob[hand_id] = reach_probs[hand_id] * strategy_prob;
}
//#pragma omp task shared(results,action_id)
results[action_id] = this->cfr(player, children[action_id], new_reach_prob, iter,
current_board,deal);
}else {
//#pragma omp task shared(results,action_id)
results[action_id] = this->cfr(player, children[action_id], reach_probs, iter,
current_board,deal);
}
}
//#pragma omp taskwait
for (int action_id = 0; action_id < actions.size(); action_id++) {
vector<float> action_utilities = results[action_id];
if(action_utilities.empty()){
continue;
}
all_action_utility[action_id] = action_utilities;
// cfr结果是每手牌的收益，payoffs代表的也是每手牌的收益，他们的长度理应相等
#ifdef DEBUG
if (action_utilities.size() != payoffs.size()) {
cout << ("errmsg") << endl;
cout << (tfm::format("node player %s ", node->getPlayer())) << endl;
node->printHistory();
throw runtime_error(
tfm::format(
"action and payoff length not match %s - %s", action_utilities.size(),
payoffs.size()
)
);
}
#endif
for (int hand_id = 0; hand_id < action_utilities.size(); hand_id++) {
if (player == node->getPlayer()) {
float strategy_prob = current_strategy[hand_id + action_id * node_player_private_cards.size()];
payoffs[hand_id] += strategy_prob * (action_utilities)[hand_id];
} else {
payoffs[hand_id] += (action_utilities)[hand_id];
}
}
}
if (player == node->getPlayer()) {
for (int i = 0; i < node_player_private_cards.size(); i++) {
//boolean regrets_all_negative = true;
for (int action_id = 0; action_id < actions.size(); action_id++) {
// 下面是regret计算的伪代码
// regret[action_id * player_hc: (action_id + 1) * player_hc]
// = all_action_utilitiy[action_id] - payoff[action_id]
regrets[action_id * node_player_private_cards.size() + i] =
(all_action_utility[action_id])[i] - payoffs[i];
}
}
if(!this->distributing_task && !this->collecting_statics) {
if (iter > this->warmup) {
trainable->updateRegrets(regrets, iter + 1, reach_probs);
}/*else if(iter < this->warmup){
vector<int> deals = this->getAllAbstractionDeal(deal);
shared_ptr<Trainable> one_trainable = node->getTrainable(deals[0]);
one_trainable->updateRegrets(regrets, iter + 1, reach_probs[player]);
}*/
else {
// iter == this->warmup
vector<int> deals = this->getAllAbstractionDeal(deal);
shared_ptr<Trainable> standard_trainable = nullptr;
for (int one_deal : deals) {
shared_ptr<Trainable> one_trainable = node->getTrainable(one_deal,true,this->use_halffloats);
if (standard_trainable == nullptr) {
one_trainable->updateRegrets(regrets, iter + 1, reach_probs);
standard_trainable = one_trainable;
} else {
one_trainable->copyStrategy(standard_trainable);
}
}
}
}
if(this->collecting_statics || (iter % this->print_interval == 0)){
float oppo_sum = 0;
vector<float> oppo_card_sum = vector<float> (52);
fill(oppo_card_sum.begin(),oppo_card_sum.end(),0);
const vector<PrivateCards>& oppo_hand = playerHands(oppo);
for(int i = 0;i < oppo_hand.size();i ++){
oppo_card_sum[oppo_hand[i].card1] += reach_probs[i];
oppo_card_sum[oppo_hand[i].card2] += reach_probs[i];
oppo_sum += reach_probs[i];
}
const vector<PrivateCards>& player_hand = playerHands(player);
vector<float> evs(actions.size() * node_player_private_cards.size(),0.0);
for (int action_id = 0; action_id < actions.size(); action_id++) {
for (int hand_id = 0; hand_id < node_player_private_cards.size(); hand_id++) {
float one_ev = (all_action_utility)[action_id][hand_id];//current_strategy; //[hand_id + action_id * node_player_private_cards.size()];
int oppo_same_card_ind = this->pcm.indPlayer2Player(player,oppo,hand_id);
float plus_reach_prob;
const PrivateCards& one_player_hand = player_hand[hand_id];
if(oppo_same_card_ind == -1){
plus_reach_prob = 0;
}else{
plus_reach_prob = reach_probs[oppo_same_card_ind];
}
float rp_sum = (
oppo_sum - oppo_card_sum[one_player_hand.card1]
- oppo_card_sum[one_player_hand.card2]
+ plus_reach_prob);
evs[hand_id + action_id * node_player_private_cards.size()] = one_ev / rp_sum;
}
}
// TODO if evs contains nan what should we do?
trainable->setEv(evs);
}
}
return payoffs;
}
vector<float>
PCfrSolver::showdownUtility(int player, shared_ptr<ShowdownNode> node, const vector<float> &reach_probs,
int iter, uint64_t current_board,int deal) {
// player win时候player的收益，player lose的时候收益明显为-player_payoff
int oppo = 1 - player;
float win_payoff = node->get_payoffs(ShowdownNode::ShowDownResult::NOTTIE,player,player);
float lose_payoff = node->get_payoffs(ShowdownNode::ShowDownResult::NOTTIE,oppo,player);
const vector<PrivateCards>& player_private_cards = this->ranges[player];
const vector<PrivateCards>& oppo_private_cards = this->ranges[oppo];
const vector<RiverCombs>& player_combs = this->rrm.getRiverCombos(player,player_private_cards,current_board);
const vector<RiverCombs>& oppo_combs = this->rrm.getRiverCombos(oppo,oppo_private_cards,current_board);
vector<float> payoffs = vector<float>(player_private_cards.size());
float winsum = 0;
vector<float> card_winsum = vector<float> (52);//node->card_sum;
fill(card_winsum.begin(),card_winsum.end(),0);
int j = 0;
for(int i = 0;i < player_combs.size();i ++){
const RiverCombs& one_player_comb = player_combs[i];
while (j < oppo_combs.size() && one_player_comb.rank < oppo_combs[j].rank){
const RiverCombs& one_oppo_comb = oppo_combs[j];
winsum += reach_probs[one_oppo_comb.reach_prob_index];
card_winsum[one_oppo_comb.private_cards.card1] += reach_probs[one_oppo_comb.reach_prob_index];
card_winsum[one_oppo_comb.private_cards.card2] += reach_probs[one_oppo_comb.reach_prob_index];
j ++;
}
payoffs[one_player_comb.reach_prob_index] = (winsum
- card_winsum[one_player_comb.private_cards.card1]
- card_winsum[one_player_comb.private_cards.card2]
) * win_payoff;
}
// 计算失败时的payoff
float losssum = 0;
vector<float>& card_losssum = card_winsum;
fill(card_losssum.begin(),card_losssum.end(),0);
j = oppo_combs.size() - 1;
for(int i = player_combs.size() - 1;i >= 0;i --){
const RiverCombs& one_player_comb = player_combs[i];
while (j >= 0 && one_player_comb.rank > oppo_combs[j].rank){
const RiverCombs& one_oppo_comb = oppo_combs[j];
losssum += reach_probs[one_oppo_comb.reach_prob_index];
card_losssum[one_oppo_comb.private_cards.card1] += reach_probs[one_oppo_comb.reach_prob_index];
card_losssum[one_oppo_comb.private_cards.card2] += reach_probs[one_oppo_comb.reach_prob_index];
j --;
}
payoffs[one_player_comb.reach_prob_index] += (losssum
- card_losssum[one_player_comb.private_cards.card1]
- card_losssum[one_player_comb.private_cards.card2]
) * lose_payoff;
}
return payoffs;
}
vector<float>
PCfrSolver::terminalUtility(int player, shared_ptr<TerminalNode> node, const vector<float> &reach_prob, int iter,
uint64_t current_board,int deal) {
float player_payoff = node->get_payoffs()[player];
int oppo = 1 - player;
const vector<PrivateCards>& player_hand = playerHands(player);
const vector<PrivateCards>& oppo_hand = playerHands(oppo);
vector<float> payoffs = vector<float>(this->playerHands(player).size());
float oppo_sum = 0;
vector<float> oppo_card_sum = vector<float> (52);
fill(oppo_card_sum.begin(),oppo_card_sum.end(),0);
for(int i = 0;i < oppo_hand.size();i ++){
oppo_card_sum[oppo_hand[i].card1] += reach_prob[i];
oppo_card_sum[oppo_hand[i].card2] += reach_prob[i];
oppo_sum += reach_prob[i];
}
for(int i = 0;i < player_hand.size();i ++){
const PrivateCards& one_player_hand = player_hand[i];
if(Card::boardsHasIntercept(current_board,Card::boardInts2long(one_player_hand.get_hands()))){
continue;
}
int oppo_same_card_ind = this->pcm.indPlayer2Player(player,oppo,i);
float plus_reach_prob;
if(oppo_same_card_ind == -1){
plus_reach_prob = 0;
}else{
plus_reach_prob = reach_prob[oppo_same_card_ind];
}
payoffs[i] = player_payoff * (
oppo_sum - oppo_card_sum[one_player_hand.card1]
- oppo_card_sum[one_player_hand.card2]
+ plus_reach_prob
);
}
return payoffs;
}
void PCfrSolver::findGameSpecificIsomorphisms() {
// hand isomorphisms
vector<Card> board_cards = Card::long2boardCards(this->initial_board_long);
for(int i = 0;i <= 1;i ++){
vector<PrivateCards>& range = i == 0?this->range1:this->range2;
for(int i_range = 0;i_range < range.size();i_range ++) {
PrivateCards one_range = range[i_range];
uint32_t range_hash[4]; // four colors, hash of the isomorphisms range + hand combos
for(int i = 0;i < 4;i ++)range_hash[i] = 0;
for (int color = 0; color < 4; color++) {
for (Card one_card:board_cards) {
if (one_card.getCardInt() % 4 == color) {
range_hash[color] = range_hash[color] | (1 << (one_card.getCardInt() / 4));
}
}
}
for (int color = 0; color < 4; color++) {
for (int one_card_int:{one_range.card1,one_range.card2}) {
if (one_card_int % 4 == color) {
range_hash[color] = range_hash[color] | (1 << (one_card_int / 4 + 16));
}
}
}
// TODO check whethe hash is equal with others
}
}
// chance node isomorphisms
uint16_t color_hash[4];
for(int i = 0;i < 4;i ++)color_hash[i] = 0;
for (Card one_card:board_cards) {
int rankind = one_card.getCardInt() % 4;
int suitind = one_card.getCardInt() / 4;
color_hash[rankind] = color_hash[rankind] | (1 << suitind);
}
for(int i = 0;i < 4;i ++){
this->color_iso_offset[0][i] = 0;
for(int j = 0;j < i;j ++){
if(color_hash[i] == color_hash[j]){
this->color_iso_offset[0][i] = j - i;
continue;
}
}
}
for(int deal = 0;deal < this->deck.getCards().size();deal ++) {
uint16_t color_hash[4];
for(int i = 0;i < 4;i ++)color_hash[i] = 0;
// chance node isomorphisms
for (Card one_card:board_cards) {
int rankind = one_card.getCardInt() % 4;
int suitind = one_card.getCardInt() / 4;
color_hash[rankind] = color_hash[rankind] | (1 << suitind);
}
Card one_card = this->deck.getCards()[deal];
int rankind = one_card.getCardInt() % 4;
int suitind = one_card.getCardInt() / 4;
color_hash[rankind] = color_hash[rankind] | (1 << suitind);
for (int i = 0; i < 4; i++) {
this->color_iso_offset[deal + 1][i] = 0;
for (int j = 0; j < i; j++) {
if (color_hash[i] == color_hash[j]) {
this->color_iso_offset[deal + 1][i] = j - i;
continue;
}
}
}
}
}
void PCfrSolver::purnTree() {
// TODO how to purn the tree, use wramup to start training in memory-save mode, and switch to purn tree directly to both save memory and speedup
}
void PCfrSolver::stop() {
this->nowstop = true;
}
void PCfrSolver::train() {
vector<vector<PrivateCards>> player_privates(this->player_number);
player_privates[0] = pcm.getPreflopCards(0);
player_privates[1] = pcm.getPreflopCards(1);
if(this->use_isomorphism){
this->findGameSpecificIsomorphisms();
}
BestResponse br = BestResponse(player_privates,this->player_number,this->pcm,this->rrm,this->deck,this->debug,this->color_iso_offset,this->split_round,this->num_threads,this->use_halffloats);
br.printExploitability(tree->getRoot(), 0, tree->getRoot()->getPot(), initial_board_long);
vector<vector<float>> reach_probs = this->getReachProbs();
ofstream fileWriter;
if(!this->logfile.empty())fileWriter.open(this->logfile);
uint64_t begintime = timeSinceEpochMillisec();
uint64_t endtime = timeSinceEpochMillisec();
for(int i = 0;i < this->iteration_number;i++){
for(int player_id = 0;player_id < this->player_number;player_id ++) {
this->round_deal = vector<int>{-1,-1,-1,-1};
//#pragma omp parallel
{
//#pragma omp single
{
//this->distributing_task = true;
cfr(player_id, this->tree->getRoot(), reach_probs[1 - player_id], i, this->initial_board_long,0);
//throw runtime_error("returning...");
}
}
}
if( (i % this->print_interval == 0 && i != 0 && i >= this->warmup) || this->nowstop) {
endtime = timeSinceEpochMillisec();
long time_ms = endtime - begintime;
qDebug().noquote() << "-------------------";
float expliotibility = br.printExploitability(tree->getRoot(), i + 1, tree->getRoot()->getPot(), initial_board_long);
qDebug().noquote() << QObject::tr("time used: ") << float(time_ms) / 1000 << QObject::tr(" second.");
if(!this->logfile.empty()){
json jo;
jo["iteration"] = i;
jo["exploitibility"] = expliotibility;
jo["time_ms"] = time_ms;
fileWriter << jo << endl;
}
if(expliotibility <= this->accuracy){
break;
}
if(this->nowstop){
this->nowstop = false;
break;
}
//begintime = timeSinceEpochMillisec();
}
}
qDebug().noquote() << QObject::tr("collecting statics");
this->collecting_statics = true;
for(int player_id = 0;player_id < this->player_number;player_id ++) {
this->round_deal = vector<int>{-1,-1,-1,-1};
//#pragma omp parallel
{
//#pragma omp single
{
//this->distributing_task = true;
cfr(player_id, this->tree->getRoot(), reach_probs[1 - player_id], this->iteration_number, this->initial_board_long,0);
}
}
}
this->collecting_statics = false;
this->statics_collected = true;
qDebug().noquote() << QObject::tr("statics collected");
if(!this->logfile.empty()) {
fileWriter.flush();
fileWriter.close();
}
}
void PCfrSolver::exchangeRange(json& strategy,int rank1,int rank2,shared_ptr<ActionNode> one_node){
if(rank1 == rank2)return;
int player = one_node->getPlayer();
vector<string> range_strs;
vector<vector<float>> strategies;
for(int i = 0;i < this->ranges[player].size();i ++){
string one_range_str = this->ranges[player][i].toString();
if(!strategy.contains(one_range_str)){
for(auto one_key:strategy.items()){
cout << one_key.key() << endl;
}
cout << "strategy: " << strategy << endl;
throw runtime_error(tfm::format("%s not exist in strategy",one_range_str));
}
vector<float> one_strategy = strategy[one_range_str];
range_strs.push_back(one_range_str);
strategies.push_back(one_strategy);
}
exchange_color(strategies,this->ranges[player],rank1,rank2);
for(int i = 0;i < this->ranges[player].size();i ++) {
string one_range_str = this->ranges[player][i].toString();
vector<float> one_strategy = strategies[i];
strategy[one_range_str] = one_strategy;
}
}
void PCfrSolver::reConvertJson(const shared_ptr<GameTreeNode>& node,json& strategy,string key,int depth,int max_depth,vector<string> prefix,int deal,vector<vector<int>> exchange_color_list) {
if(depth >= max_depth) return;
if(node->getType() == GameTreeNode::GameTreeNodeType::ACTION) {
json* retval;
if(key != ""){
strategy[key] = json();
retval = &(strategy[key]);
}else{
retval = &strategy;
}
shared_ptr<ActionNode> one_node = std::dynamic_pointer_cast<ActionNode>(node);
vector<string> actions_str;
for(GameActions one_action:one_node->getActions()) actions_str.push_back(one_action.toString());
(*retval)["actions"] = actions_str;
(*retval)["player"] = one_node->getPlayer();
(*retval)["childrens"] = json();
json& childrens = (*retval)["childrens"];
for(int i = 0;i < one_node->getActions().size();i ++){
GameActions& one_action = one_node->getActions()[i];
shared_ptr<GameTreeNode> one_child = one_node->getChildrens()[i];
vector<string> new_prefix(prefix);
new_prefix.push_back(one_action.toString());
this->reConvertJson(one_child,childrens,one_action.toString(),depth,max_depth,new_prefix,deal,exchange_color_list);
}
if((*retval)["childrens"].empty()){
(*retval).erase("childrens");
}
shared_ptr<Trainable> trainable = one_node->getTrainable(deal,false);
if(trainable != nullptr) {
(*retval)["strategy"] = trainable->dump_strategy(false);
for(vector<int> one_exchange:exchange_color_list){
int rank1 = one_exchange[0];
int rank2 = one_exchange[1];
this->exchangeRange((*retval)["strategy"]["strategy"],rank1,rank2,one_node);
}
}
(*retval)["node_type"] = "action_node";
}else if(node->getType() == GameTreeNode::GameTreeNodeType::SHOWDOWN) {
}else if(node->getType() == GameTreeNode::GameTreeNodeType::TERMINAL) {
}else if(node->getType() == GameTreeNode::GameTreeNodeType::CHANCE) {
json* retval;
if(key != ""){
strategy[key] = json();
retval = &(strategy[key]);
}else{
retval = &strategy;
}
shared_ptr<ChanceNode> chanceNode = std::dynamic_pointer_cast<ChanceNode>(node);
const vector<Card>& cards = chanceNode->getCards();
shared_ptr<GameTreeNode> childerns = chanceNode->getChildren();
vector<string> card_strs;
for(Card card:cards)
card_strs.push_back(card.toString());
json& dealcards = (*retval)["dealcards"];
for(int i = 0;i < cards.size();i ++){
vector<vector<int>> new_exchange_color_list(exchange_color_list);
Card& one_card = const_cast<Card &>(cards[i]);
vector<string> new_prefix(prefix);
new_prefix.push_back("Chance:" + one_card.toString());
int card = i;
int offset = this->color_iso_offset[deal][one_card.getCardInt() % 4];
if(offset < 0) {
for(int x = 0;x < cards.size();x ++){
if(
Card::card2int(cards[x]) ==
(Card::card2int(cards[card]) + offset)
){
card = x;
break;
}
}
if(card == i){
throw runtime_error("isomorphism not found while dump strategy");
}
vector<int> one_exchange{one_card.getCardInt() % 4,one_card.getCardInt() % 4 + offset};
new_exchange_color_list.push_back(one_exchange);
}
int card_num = this->deck.getCards().size();
int new_deal;
if(deal == 0){
new_deal = card + 1;
} else if (deal > 0 && deal <= card_num){
int origin_deal = deal - 1;
#ifdef DEBUG
if(origin_deal == card) throw runtime_error("deal should not be equal");
#endif
new_deal = card_num * origin_deal + card;
new_deal += (1 + card_num);
} else{
throw runtime_error(tfm::format("deal out of range : %s ",deal));
}
if(exchange_color_list.size() > 1){
throw runtime_error("exchange color list shouldn't be exceed size 1 here");
}
string one_card_str = one_card.toString();
if(exchange_color_list.size() == 1) {
int rank1 = exchange_color_list[0][0];
int rank2 = exchange_color_list[0][1];
if(one_card.getCardInt() % 4 == rank1){
one_card_str = Card::intCard2Str(one_card.getCardInt() - rank1 + rank2);
}else if(one_card.getCardInt() % 4 == rank2){
one_card_str = Card::intCard2Str(one_card.getCardInt() - rank2 + rank1);
}
}
this->reConvertJson(childerns,dealcards,one_card_str,depth + 1,max_depth,new_prefix,new_deal,new_exchange_color_list);
}