Repository navigation
Expand file tree
/
Copy pathStrongAI.java
More file actions
1647 lines (1573 loc) · 46.9 KB
/
Copy pathStrongAI.java
File metadata and controls
1647 lines (1573 loc) · 46.9 KB
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
import java.util.ArrayList;
import java.util.Arrays;
import java.util.HashMap;
import java.util.HashSet;
import java.util.List;
import java.util.Random;
/**
* Legal player. It never reads a face-down card or an unseen stock card on the
* real deal. When several moves are possible it samples deals consistent with
* what it has seen and counts only proved wins (search shows a forced win after
* that candidate in that sample). Losing or unproved samples cannot outvote a
* move that proved a win elsewhere. Fully visible positions are searched
* directly. {@link #chooseOneSample} is the older one-sample 52.9% path.
*/
final class StrongAI {
private static final int SAMPLES = 20;
private static final int SWITCH_MARGIN = 3;
private static final boolean USE_EXACT = false;
private static final int ORACLE_STEPS = 180;
private static final int EXACT_BUDGET = 120000;
/** When false, {@link #choose} uses the original one-sample Legal path. */
static boolean USE_VOTING = true;
/** Shuffled hidden layouts voted on when hidden cards remain. Tuned on a small seed set. */
static int VOTE_SAMPLES = 8;
/** Independent per-move, per-sample node budget for proved-win searches. */
static int VOTE_NODE_LIMIT = 6000;
/**
* V2 Legal options. Default off so {@code legal} / {@code vote:8:6000} still
* reproduce the 568/1000 player until a v2 config beats that full-set result.
*/
static boolean USE_V2 = false;
/** 568/1000 first pass plus extras that do not consume the instance rng. */
static boolean USE_PLUS = false;
/** Keep layouts that still match every card revealed so far; never peek at remaining hidden cards. */
static boolean USE_PARTICLES = false;
/** If no sample proves a win, score candidates by rollout progress across samples instead of one shuffle. */
static boolean USE_PROGRESS_FALLBACK = false;
/** Add extra samples only when the proved-win vote is tied/close or empty. */
static boolean USE_ADAPTIVE = false;
static int EXTRA_SAMPLES = 8;
/** When hidden cards are at most this many, enumerate every layout. */
static int ENUM_HIDDEN = 4;
/** If no prove after the first pass, retry the same layouts at this budget (0 = skip). */
static int SECOND_PASS_LIMIT = 0;
/** Node budget when the position is fully visible. */
static int VISIBLE_NODE_LIMIT = WideBound.LEGAL_NODE_LIMIT;
private final Random rng;
private final long agentSeed;
private int switches;
/** Posterior samples of the current hidden slots; identities never read from the true deal. */
private final ArrayList<int[]> particles = new ArrayList<int[]>();
StrongAI(long seed) {
agentSeed = seed;
rng = new Random(seed);
}
static final class Outcome {
final boolean won;
final int moves;
Outcome(boolean won, int moves) {
this.won = won;
this.moves = moves;
}
}
Outcome play(Klondike g) {
HashSet<Long> seen = new HashSet<Long>();
int moves = 0;
particles.clear();
while (moves < 5000) {
if (g.won()) {
g.finish();
return new Outcome(true, moves);
}
long h = g.hash();
if (!seen.add(h)) {
return new Outcome(false, moves);
}
Klondike before = USE_PARTICLES ? g.copy() : null;
Klondike.Move move = choose(g);
if (move == null) {
return new Outcome(false, moves);
}
g.apply(move);
if (USE_PARTICLES && before != null) {
filterParticles(before, move, g);
}
moves++;
}
return new Outcome(false, moves);
}
Klondike.Move choose(Klondike real) {
if (USE_V2) {
return chooseVotingV2(real);
}
if (USE_VOTING) {
return chooseVoting(real);
}
return chooseOneSample(real);
}
/**
* Original Legal path that won 529/1000 (52.9%): one shuffled sample, then
* {@link WideBound#winningMove} with {@link WideBound#LEGAL_NODE_LIMIT}.
*/
Klondike.Move chooseOneSample(Klondike real) {
if (real.won()) {
return Klondike.Move.auto();
}
if (real.hiddenCount() <= 0) {
Klondike.Move seen = WideBound.winningMove(real);
return seen != null ? seen : SmartAI.heuristic(real);
}
int hidden = real.hiddenCount();
int[] base = new int[hidden];
real.fillHidden(base);
int[] perm = base.clone();
shuffle(perm);
Klondike world = real.copy();
world.assignHidden(perm);
Klondike.Move advised = WideBound.winningMove(world);
return advised != null ? advised : SmartAI.heuristic(real);
}
/**
* Multi-sample Legal path. Hidden cards are completed only from the unseen
* remainder (the Legal player already knows those identities by elimination);
* the true slot assignment is shuffled away and never searched. Each
* candidate is scored by how many samples it proves as a forced win.
*/
Klondike.Move chooseVoting(Klondike real) {
if (real.won()) {
return Klondike.Move.auto();
}
if (real.hiddenCount() <= 0) {
int budget = USE_VOTING && VISIBLE_NODE_LIMIT > 0 ? VISIBLE_NODE_LIMIT : WideBound.LEGAL_NODE_LIMIT;
// Default VISIBLE_NODE_LIMIT is LEGAL_NODE_LIMIT, so legal/vote:8:6000 stay at 25k.
Klondike.Move seen = WideBound.winningMove(real, budget);
return seen != null ? seen : SmartAI.heuristic(real);
}
List<Klondike.Move> opts = WideBound.orderedOptions(real);
if (opts.isEmpty()) {
return SmartAI.heuristic(real);
}
int hidden = real.hiddenCount();
int[] base = new int[hidden];
real.fillHidden(base);
int cap = (USE_V2 || USE_PLUS) ? ENUM_HIDDEN : 4;
int rounds;
boolean enumerate = hidden <= cap;
if (enumerate) {
rounds = factorial(hidden);
} else {
rounds = VOTE_SAMPLES;
}
int[] proved = new int[opts.size()];
int[] unknown = new int[opts.size()];
int[] lost = new int[opts.size()];
int[] progress = new int[opts.size()];
ArrayList<int[]> worlds = new ArrayList<int[]>();
int examined = 0;
for (int r = 0; r < rounds; r++) {
int[] perm = new int[hidden];
if (enumerate) {
permutation(r, base, perm);
} else {
System.arraycopy(base, 0, perm, 0, hidden);
shuffle(perm);
}
worlds.add(perm);
scoreOneWorld(real, opts, perm, VOTE_NODE_LIMIT, proved, unknown, lost, progress);
examined++;
if (!enumerate && clearProvedLeader(proved, examined, rounds)) {
break;
}
}
// Add-ons use only already-drawn layouts or a visible-hash RNG, so they
// do not consume the instance rng. If they leave the decision unchanged,
// chooseOneSample and later moves stay identical to the 568/1000 player.
if (USE_PLUS) {
if (!enumerate && USE_ADAPTIVE && EXTRA_SAMPLES > 0 && needsAdaptive(proved)) {
for (int i = 0; i < EXTRA_SAMPLES; i++) {
int[] perm = makeSample(real, base, VOTE_SAMPLES + i, false);
worlds.add(perm);
scoreOneWorld(real, opts, perm, VOTE_NODE_LIMIT, proved, unknown, lost, progress);
}
}
int bestNow = bestProved(proved);
if (proved[bestNow] == 0 && SECOND_PASS_LIMIT > VOTE_NODE_LIMIT) {
int[] proved2 = new int[opts.size()];
int[] unknown2 = new int[opts.size()];
int[] lost2 = new int[opts.size()];
int[] progress2 = new int[opts.size()];
for (int r = 0; r < worlds.size(); r++) {
scoreOneWorld(real, opts, worlds.get(r), SECOND_PASS_LIMIT, proved2, unknown2, lost2, progress2);
}
if (proved2[bestProved(proved2)] > 0) {
proved = proved2;
unknown = unknown2;
lost = lost2;
progress = progress2;
}
}
}
int best = bestProved(proved);
if (proved[best] > 0) {
return opts.get(best);
}
if (USE_PLUS && USE_PROGRESS_FALLBACK) {
return opts.get(bestProgress(progress, lost, unknown));
}
return chooseOneSample(real);
}
private void scoreOneWorld(Klondike real, List<Klondike.Move> opts, int[] perm, int budget,
int[] proved, int[] unknown, int[] lost, int[] progress) {
Klondike world = real.copy();
world.assignHidden(perm);
WideBound.RootEval ev = WideBound.evaluateRoots(world, budget);
for (int i = 0; i < opts.size(); i++) {
int found = provedIndex(ev, opts.get(i));
int result = found < 0 ? -1 : ev.result[found];
if (result == 1) {
proved[i]++;
} else if (result == 0) {
lost[i]++;
} else {
unknown[i]++;
}
if (USE_PROGRESS_FALLBACK && result != 0) {
Klondike roll = real.copy();
roll.assignHidden(perm.clone());
progress[i] += rollout(roll, opts.get(i));
}
}
}
private static int provedIndex(WideBound.RootEval ev, Klondike.Move want) {
for (int i = 0; i < ev.opts.size(); i++) {
if (sameMove(ev.opts.get(i), want)) {
return i;
}
}
return -1;
}
/** True when one move's proved-win count cannot be caught by the remaining samples. */
private static boolean clearProvedLeader(int[] proved, int examined, int rounds) {
int best = 0;
int second = 0;
for (int i = 0; i < proved.length; i++) {
if (proved[i] > best) {
second = best;
best = proved[i];
} else if (proved[i] > second) {
second = proved[i];
}
}
if (best <= 0) {
return false;
}
int remaining = rounds - examined;
return best > second + remaining;
}
/**
* Stronger Legal path: optional consistent particles, adaptive extra samples
* on close or empty votes, a deeper second pass, and a progress vote when
* nothing is proved. Hidden cards are never read except as an unseen
* multiset that is then shuffled or filtered against newly visible cards.
*/
Klondike.Move chooseVotingV2(Klondike real) {
if (real.won()) {
return Klondike.Move.auto();
}
if (real.hiddenCount() <= 0) {
Klondike.Move seen = WideBound.winningMove(real, VISIBLE_NODE_LIMIT);
return seen != null ? seen : SmartAI.heuristic(real);
}
List<Klondike.Move> opts = WideBound.orderedOptions(real);
if (opts.isEmpty()) {
return SmartAI.heuristic(real);
}
int hidden = real.hiddenCount();
int[] base = new int[hidden];
real.fillHidden(base);
boolean enumerate = hidden <= ENUM_HIDDEN;
int planned = enumerate ? factorial(hidden) : VOTE_SAMPLES;
ArrayList<int[]> worlds = new ArrayList<int[]>();
if (USE_PARTICLES && !enumerate) {
keepMatchingParticles(hidden);
for (int i = 0; i < particles.size() && worlds.size() < planned; i++) {
worlds.add(particles.get(i).clone());
}
}
while (worlds.size() < planned) {
worlds.add(makeSample(real, base, worlds.size(), enumerate));
if (USE_PARTICLES && !enumerate) {
particles.add(worlds.get(worlds.size() - 1).clone());
}
}
int[] proved = new int[opts.size()];
int[] unknown = new int[opts.size()];
int[] lost = new int[opts.size()];
int[] progress = new int[opts.size()];
int examined = scoreWorlds(real, opts, worlds, 0, worlds.size(), VOTE_NODE_LIMIT,
proved, unknown, lost, progress, !enumerate);
if (!enumerate && USE_ADAPTIVE && EXTRA_SAMPLES > 0 && needsAdaptive(proved)) {
int from = worlds.size();
for (int i = 0; i < EXTRA_SAMPLES; i++) {
worlds.add(makeSample(real, base, worlds.size(), false));
if (USE_PARTICLES) {
particles.add(worlds.get(worlds.size() - 1).clone());
}
}
scoreWorlds(real, opts, worlds, from, worlds.size(), VOTE_NODE_LIMIT,
proved, unknown, lost, progress, true);
}
int best = bestProved(proved);
if (proved[best] == 0 && SECOND_PASS_LIMIT > VOTE_NODE_LIMIT) {
int[] proved2 = new int[opts.size()];
int[] unknown2 = new int[opts.size()];
int[] lost2 = new int[opts.size()];
int[] progress2 = new int[opts.size()];
scoreWorlds(real, opts, worlds, 0, worlds.size(), SECOND_PASS_LIMIT,
proved2, unknown2, lost2, progress2, false);
if (proved2[bestProved(proved2)] > 0) {
proved = proved2;
unknown = unknown2;
lost = lost2;
progress = progress2;
best = bestProved(proved);
}
}
if (proved[best] > 0) {
return opts.get(best);
}
if (USE_PROGRESS_FALLBACK && worlds.size() > 0) {
return opts.get(bestProgress(progress, lost, unknown));
}
// Deterministic one-sample 25k search on the first layout, not a new rng shuffle.
Klondike fallback = real.copy();
fallback.assignHidden(worlds.size() > 0 ? worlds.get(0) : makeSample(real, base, 0, enumerate));
Klondike.Move advised = WideBound.winningMove(fallback);
return advised != null ? advised : SmartAI.heuristic(real);
}
/** A hidden layout consistent with the visible cards. Seeded only from visible state. */
private int[] makeSample(Klondike real, int[] base, int index, boolean enumerate) {
int[] perm = new int[base.length];
if (enumerate) {
permutation(index, base, perm);
} else {
System.arraycopy(base, 0, perm, 0, base.length);
shuffleWith(perm, new Random(sampleSeed(real, index)));
}
return perm;
}
private long sampleSeed(Klondike real, int index) {
return agentSeed * 0xD1B54A32D192ED03L
^ real.visibleHash() * 0x9E3779B97F4A7C15L
^ ((index + 1L) * 0x2127599BF432BE03L);
}
private static void shuffleWith(int[] ids, Random source) {
for (int i = ids.length - 1; i > 0; i--) {
int j = source.nextInt(i + 1);
int tmp = ids[i];
ids[i] = ids[j];
ids[j] = tmp;
}
}
private int scoreWorlds(Klondike real, List<Klondike.Move> opts, List<int[]> worlds, int from, int to,
int budget, int[] proved, int[] unknown, int[] lost, int[] progress, boolean earlyStop) {
int examined = from;
for (int r = from; r < to; r++) {
Klondike world = real.copy();
world.assignHidden(worlds.get(r));
WideBound.RootEval ev = WideBound.evaluateRoots(world, budget);
for (int i = 0; i < opts.size(); i++) {
int found = provedIndex(ev, opts.get(i));
int result = found < 0 ? -1 : ev.result[found];
if (result == 1) {
proved[i]++;
} else if (result == 0) {
lost[i]++;
} else {
unknown[i]++;
}
if (USE_PROGRESS_FALLBACK && result != 0) {
Klondike roll = real.copy();
roll.assignHidden(worlds.get(r).clone());
progress[i] += rollout(roll, opts.get(i));
}
}
examined = r + 1;
if (earlyStop && clearProvedLeader(proved, examined, to)) {
break;
}
}
return examined;
}
private static boolean needsAdaptive(int[] proved) {
int best = bestProved(proved);
if (proved[best] == 0) {
return true;
}
int second = 0;
for (int i = 0; i < proved.length; i++) {
if (i != best && proved[i] > second) {
second = proved[i];
}
}
return proved[best] <= second + 1;
}
private static int bestProved(int[] proved) {
int best = 0;
for (int i = 1; i < proved.length; i++) {
if (proved[i] > proved[best]) {
best = i;
}
}
return best;
}
/** Prefer more progress, then fewer proved losses, then more unproved (not-lost) samples. */
private static int bestProgress(int[] progress, int[] lost, int[] unknown) {
int best = 0;
for (int i = 1; i < progress.length; i++) {
if (progress[i] > progress[best]) {
best = i;
} else if (progress[i] == progress[best]) {
if (lost[i] < lost[best] || (lost[i] == lost[best] && unknown[i] > unknown[best])) {
best = i;
}
}
}
return best;
}
private void keepMatchingParticles(int hidden) {
for (int i = particles.size() - 1; i >= 0; i--) {
if (particles.get(i).length != hidden) {
particles.remove(i);
}
}
}
/**
* Drop hypothesized layouts that would have revealed a different visible card
* than the one just seen. Remaining hidden slots are taken from the particle,
* not from the true deal.
*/
private void filterParticles(Klondike before, Klondike.Move move, Klondike after) {
int hiddenBefore = before.hiddenCount();
ArrayList<int[]> kept = new ArrayList<int[]>();
for (int i = 0; i < particles.size(); i++) {
int[] perm = particles.get(i);
if (perm.length != hiddenBefore) {
continue;
}
Klondike world = before.copy();
world.assignHidden(perm);
world.apply(move);
if (!visibleAgree(world, after)) {
continue;
}
int hidden = after.hiddenCount();
int[] next = new int[hidden];
world.fillHidden(next);
kept.add(next);
}
particles.clear();
particles.addAll(kept);
}
/** Compare only cards a Legal player can see. Face-down and unseen stock are ignored. */
private static boolean visibleAgree(Klondike a, Klondike b) {
if (a.wn != b.wn || a.sn != b.sn || a.stockKnown != b.stockKnown) {
return false;
}
for (int s = 0; s < 4; s++) {
if (a.found[s] != b.found[s]) {
return false;
}
}
for (int c = 0; c < 7; c++) {
if (a.pn[c] != b.pn[c] || a.dn[c] != b.dn[c]) {
return false;
}
for (int i = a.dn[c]; i < a.pn[c]; i++) {
if (a.pile[c][i] != b.pile[c][i]) {
return false;
}
}
}
for (int i = 0; i < a.wn; i++) {
if (a.waste[i] != b.waste[i]) {
return false;
}
}
if (a.stockKnown) {
for (int i = 0; i < a.sn; i++) {
if (a.stock[i] != b.stock[i]) {
return false;
}
}
}
return true;
}
private Klondike.Move unusedEnumeration(Klondike real) {
Klondike.Move usual = SmartAI.heuristic(real);
int hidden = real.hiddenCount();
if (usual == null || !real.stockKnown || hidden <= 0 || hidden > 5) {
return usual;
}
List<Klondike.Move> opts = reveals(real);
if (opts.size() <= 1) {
return usual;
}
int worlds = factorial(hidden);
int[] base = new int[hidden];
real.fillHidden(base);
int[] perm = new int[hidden];
int[] wins = new int[opts.size()];
for (int w = 0; w < worlds; w++) {
permutation(w, base, perm);
for (int i = 0; i < opts.size(); i++) {
Klondike world = real.copy();
world.assignHidden(perm);
if (rollout(world, opts.get(i)) >= 100000) {
wins[i]++;
}
}
}
int best = 0;
for (int i = 1; i < opts.size(); i++) {
if (wins[i] > wins[best]) {
best = i;
} else if (wins[i] == wins[best] && sameMove(opts.get(i), usual)) {
best = i;
}
}
return opts.get(best);
}
/** Move recommended by the full search in every random layout, or null. */
private Klondike.Move consensus(Klondike real) {
int hidden = real.hiddenCount();
int[] base = new int[hidden];
real.fillHidden(base);
Klondike.Move agreed = null;
for (int s = 0; s < 3; s++) {
int[] perm = base.clone();
shuffle(perm);
Klondike world = real.copy();
world.assignHidden(perm);
Klondike.Move advised = WideBound.winningMove(world);
if (advised == null) {
return null;
}
if (agreed == null) {
agreed = advised;
} else if (!sameMove(agreed, advised)) {
return null;
}
}
return agreed;
}
private int[] sampleWins(Klondike real, List<Klondike.Move> revs) {
int hidden = real.hiddenCount();
int[] base = new int[hidden];
real.fillHidden(base);
int[] perm = new int[hidden];
int[] wins = new int[revs.size()];
for (int r = 0; r < SAMPLES; r++) {
System.arraycopy(base, 0, perm, 0, hidden);
shuffle(perm);
for (int i = 0; i < revs.size(); i++) {
Klondike world = real.copy();
world.assignHidden(perm);
if (rollout(world, revs.get(i)) >= 100000) {
wins[i]++;
}
}
}
return wins;
}
private static boolean sameMove(Klondike.Move a, Klondike.Move b) {
return a.kind == b.kind && a.from == b.from && a.index == b.index && a.to == b.to;
}
private int winsFor(Klondike real, Klondike.Move move, int rounds) {
int hidden = real.hiddenCount();
int[] base = new int[hidden];
real.fillHidden(base);
int[] perm = new int[hidden];
int wins = 0;
for (int r = 0; r < rounds; r++) {
System.arraycopy(base, 0, perm, 0, hidden);
shuffle(perm);
Klondike world = real.copy();
world.assignHidden(perm);
if (rollout(world, move) >= 100000) {
wins++;
}
}
return wins;
}
private static List<Klondike.Move> candidates(Klondike g, Klondike.Move usual) {
List<Klondike.Move> ms = new ArrayList<Klondike.Move>();
ms.add(usual);
List<Klondike.Move> revs = reveals(g);
for (int i = 0; i < revs.size(); i++) {
addUnique(ms, revs.get(i));
}
if (g.wn > 0) {
int card = g.wasteTop();
for (int to = 0; to < 7; to++) {
if (g.canPlace(card, to)) {
addUnique(ms, new Klondike.Move(Klondike.TO_TAB, Klondike.WASTE, -1, to));
}
}
if (g.canFound(card)) {
addUnique(ms, new Klondike.Move(Klondike.TO_FOUND, Klondike.WASTE, -1, Klondike.suit(card)));
}
}
for (int c = 0; c < 7; c++) {
if (g.pn[c] > 0 && g.safeFound(g.tail(c))) {
addUnique(ms, new Klondike.Move(Klondike.TO_FOUND, c, g.pn[c] - 1, Klondike.suit(g.tail(c))));
}
}
if (g.canDraw()) {
addUnique(ms, Klondike.Move.draw());
}
return ms;
}
private static void addUnique(List<Klondike.Move> ms, Klondike.Move m) {
for (int i = 0; i < ms.size(); i++) {
Klondike.Move o = ms.get(i);
if (o.kind == m.kind && o.from == m.from && o.index == m.index && o.to == m.to) {
return;
}
}
ms.add(m);
}
/**
* Pick a reveal by how often a perfect choice of later reveals wins in a
* random layout of the unseen cards. Ties go to the deepest buried pile.
*/
private Klondike.Move voteReveals(Klondike real, List<Klondike.Move> revs) {
int hidden = real.hiddenCount();
Klondike.Move prior = deepestReveal(real);
if (hidden <= 0) {
return prior != null ? prior : revs.get(0);
}
int[] base = new int[hidden];
real.fillHidden(base);
int[] perm = new int[hidden];
int[] wins = new int[revs.size()];
int priorIndex = 0;
for (int i = 0; i < revs.size(); i++) {
if (prior != null && moveKey(revs.get(i)) == moveKey(prior)) {
priorIndex = i;
}
}
for (int r = 0; r < SAMPLES; r++) {
System.arraycopy(base, 0, perm, 0, hidden);
shuffle(perm);
for (int i = 0; i < revs.size(); i++) {
Klondike world = real.copy();
world.assignHidden(perm);
if (rollout(world, revs.get(i)) >= 100000) {
wins[i]++;
}
}
}
int best = priorIndex;
// Only leave the usual move when samples say it is losing and another
// reveal wins clearly more often.
if (wins[priorIndex] * 2 < SAMPLES) {
for (int i = 0; i < revs.size(); i++) {
if (wins[i] >= wins[best] + SWITCH_MARGIN) {
best = i;
}
}
}
return revs.get(best);
}
/** Same reveal ranking as the fast policy: deepest buried pile, then buried destination. */
private static Klondike.Move deepestReveal(Klondike g) {
List<Klondike.Move> revs = reveals(g);
Klondike.Move best = null;
int bestScore = -1;
for (int i = 0; i < revs.size(); i++) {
Klondike.Move m = revs.get(i);
int score = g.dn[m.from] * 100 + g.dn[m.to];
if (score > bestScore) {
bestScore = score;
best = m;
}
}
return best;
}
/**
* Shortest sequence of already-visible moves, including draws from a stock
* whose order is known, that uncovers a card. Face-down identities are not read.
*/
private static Klondike.Move planReveal(Klondike start, int maxDepth, int cap) {
ArrayList<Klondike> states = new ArrayList<Klondike>();
ArrayList<Klondike.Move> firsts = new ArrayList<Klondike.Move>();
ArrayList<Integer> depths = new ArrayList<Integer>();
HashSet<Long> seen = new HashSet<Long>();
states.add(start);
firsts.add(null);
depths.add(0);
seen.add(start.visibleHash());
int fd = start.faceDown();
int nodes = 0;
for (int cursor = 0; cursor < states.size(); cursor++) {
Klondike g = states.get(cursor);
int depth = depths.get(cursor);
if (depth >= maxDepth) {
continue;
}
List<Klondike.Move> moves = planMoves(g);
for (int i = 0; i < moves.size(); i++) {
if (nodes++ > cap) {
return null;
}
Klondike child = g.copy();
child.apply(moves.get(i));
Klondike.Move first = firsts.get(cursor) == null ? moves.get(i) : firsts.get(cursor);
if (child.faceDown() < fd || child.won()) {
return first;
}
long h = child.visibleHash();
if (!seen.add(h)) {
continue;
}
states.add(child);
firsts.add(first);
depths.add(depth + 1);
}
}
return null;
}
private static List<Klondike.Move> planMoves(Klondike g) {
List<Klondike.Move> ms = new ArrayList<Klondike.Move>();
ms.addAll(setups(g));
if (g.wn > 0) {
int card = g.wasteTop();
for (int to = 0; to < 7; to++) {
if (g.canPlace(card, to)) {
ms.add(new Klondike.Move(Klondike.TO_TAB, Klondike.WASTE, -1, to));
}
}
if (g.canFound(card)) {
ms.add(new Klondike.Move(Klondike.TO_FOUND, Klondike.WASTE, -1, Klondike.suit(card)));
}
}
for (int c = 0; c < 7; c++) {
if (g.pn[c] > 0 && g.canFound(g.tail(c))) {
ms.add(new Klondike.Move(Klondike.TO_FOUND, c, g.pn[c] - 1, Klondike.suit(g.tail(c))));
}
}
Klondike.Move column = makeColumn(g);
if (column != null) {
ms.add(column);
}
if (g.stockKnown && g.canDraw()) {
ms.add(Klondike.Move.draw());
}
return ms;
}
/** Moves the fast policy is actually choosing between. */
private static List<Klondike.Move> branch(Klondike g) {
List<Klondike.Move> revs = reveals(g);
if (!revs.isEmpty()) {
return revs;
}
List<Klondike.Move> parks = new ArrayList<Klondike.Move>();
if (g.wn > 0) {
int card = g.wasteTop();
for (int to = 0; to < 7; to++) {
if (g.canPlace(card, to)) {
parks.add(new Klondike.Move(Klondike.TO_TAB, Klondike.WASTE, -1, to));
}
}
}
if (parks.size() > 1) {
return parks;
}
return parks;
}
/**
* Score each candidate by how the fast legal policy finishes in random
* layouts of the unseen cards. The true layout is not one of the samples.
*/
private Klondike.Move vote(Klondike real, List<Klondike.Move> options) {
int hidden = real.hiddenCount();
if (hidden <= 0) {
return options.get(0);
}
int rounds = real.stockKnown && hidden <= 4 ? factorial(hidden) : SAMPLES;
int[] base = new int[hidden];
real.fillHidden(base);
int[] score = new int[options.size()];
int[] perm = new int[hidden];
for (int r = 0; r < rounds; r++) {
if (rounds == SAMPLES) {
System.arraycopy(base, 0, perm, 0, hidden);
shuffle(perm);
} else {
permutation(r, base, perm);
}
for (int i = 0; i < options.size(); i++) {
Klondike world = real.copy();
world.assignHidden(perm);
score[i] += rollout(world, options.get(i));
}
}
int best = 0;
for (int i = 1; i < options.size(); i++) {
if (score[i] > score[best]) {
best = i;
}
}
return options.get(best);
}
private int rollout(Klondike g, Klondike.Move first) {
g.apply(first);
HashSet<Long> seen = new HashSet<Long>();
for (int i = 0; i < ORACLE_STEPS; i++) {
if (g.won()) {
return 100000 + (ORACLE_STEPS - i);
}
if (!seen.add(g.hash())) {
break;
}
Klondike.Move m = SmartAI.heuristic(g);
if (m == null) {
break;
}
g.apply(m);
}
return (21 - g.faceDown()) * 40 + g.foundationTotal();
}
/** Perfect-information reveal choice, then the fast policy. Ceiling only. */
private static Klondike.Move thoughtful(Klondike g) {
List<Klondike.Move> revs = reveals(g);
if (revs.size() > 1) {
Klondike.Move best = revs.get(0);
int bestScore = Integer.MIN_VALUE;
for (int i = 0; i < revs.size(); i++) {
int score = underLooksUseful(g, revs.get(i)) * 10 + g.dn[revs.get(i).from] * 5 + g.dn[revs.get(i).to];
if (score > bestScore) {
bestScore = score;
best = revs.get(i);
}
}
return best;
}
return SmartAI.heuristic(g);
}
/**
* Advisor that may read every card. Callers pass either a fully visible
* position or a sample, never the raw hidden deal.
*/
static Klondike.Move oracleMove(Klondike g) {
if (g.won()) {
return null;
}
Klondike.Move reveal = bestCascade(g);
if (reveal != null) {
return reveal;
}
Klondike.Move setup = bestSetup(g);
if (setup != null) {
return setup;
}
Klondike.Move safe = firstSafe(g);
if (safe != null) {
return safe;
}
Klondike.Move park = bestPark(g, true);
if (park != null) {
return park;
}
if (g.sn > 0 && nextCardHelps(g)) {
return Klondike.Move.draw();
}
Klondike.Move column = makeColumn(g);
if (column != null) {
return column;
}
if (g.canDraw()) {
return Klondike.Move.draw();
}
Klondike.Move low = lowFoundation(g);
if (low != null) {
return low;
}
return worry(g);
}
private static Klondike.Move bestCascade(Klondike g) {
Klondike.Move best = null;
int bestScore = 0;
List<Klondike.Move> revs = reveals(g);
for (int i = 0; i < revs.size(); i++) {
Klondike copy = g.copy();
copy.apply(revs.get(i));
int score = 100 + cascade(copy);
if (score > bestScore) {
bestScore = score;
best = revs.get(i);
}
}
return best;
}
/** Reveals that follow from this position, using cards the advisor can see. */
private static int cascade(Klondike g) {
int n = 0;
HashSet<Long> seen = new HashSet<Long>();
while (n < 20 && seen.add(g.hash())) {
if (g.won()) {
return 5000 + n * 10;
}
List<Klondike.Move> revs = reveals(g);
if (revs.isEmpty()) {
break;
}
Klondike.Move pick = revs.get(0);
int pickScore = -1;
for (int i = 0; i < revs.size(); i++) {
int underBonus = underLooksUseful(g, revs.get(i));
int score = g.dn[revs.get(i).from] * 10 + underBonus;
if (score > pickScore) {
pickScore = score;
pick = revs.get(i);
}
}
g.apply(pick);
n++;
}
return n * 10 - g.faceDown();
}
/** Bonus for the card that a reveal would turn up. Advisor-only. */
private static int underLooksUseful(Klondike g, Klondike.Move reveal) {
int col = reveal.from;
if (g.dn[col] <= 0) {
return 0;
}
int under = g.pile[col][g.dn[col] - 1];
int bonus = 0;
if (g.canFound(under) || g.safeFound(under)) {
bonus += 25;
}
for (int to = 0; to < 7; to++) {
if (to != col && g.canPlace(under, to)) {
bonus += 15;
break;
}
}
if (Klondike.rank(under) == 12 && g.emptyColumns() > 0) {
bonus += 20;
}
return bonus;