-
Notifications
You must be signed in to change notification settings - Fork 0
/
playoff.go
593 lines (528 loc) · 14 KB
/
playoff.go
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
package main
import (
"flag"
"fmt"
"log"
"math"
"math/rand"
"os"
"time"
)
// MAXIMIZER, MINIMIZER, UNSET
// are used to denote which player, and also
// as indexes into arrays for too-clever output and
// win/loss indicators.
const (
MAXIMIZER = 1 // Computer plays MAXIMIZER
MINIMIZER = -1 // Computer has human play MINIMIZER
UNSET = 0
WIN = 10000
LOSS = -10000
)
// Board - internal representation of a traditional Kalah board
type Board struct {
maxpits [7]int
minpits [7]int
player int // which player made the move resulting in this configuration
}
type chooserFunction func(bd Board, print bool) (bestpit int, bestvalue int)
type player struct {
name string
bd Board
moveFn chooserFunction
}
// MCTS holds values that func chooseMonteCarlo() needs, but
// aren't passed in as arguments.
type MCTS struct {
iterations int
uctk float64
}
type AlphaBeta struct {
maxPly int
}
var winningStonesCount int
func main() {
player1Type := flag.String("1", "M", "first player type")
player2Type := flag.String("2", "A", "second player type")
maxDepthPtr := flag.Int("d", 6, "maximum lookahead depth, moves for each side")
stoneCountPtr := flag.Int("n", 4, "number of stones per pit")
iterationPtr := flag.Int("i", 200000, "Number of iterations for MCTS")
uctkPtr := flag.Float64("U", 1.414, "UCTK factor, MCTS only")
flag.Parse()
winningStonesCount = 6 * *stoneCountPtr
maximizer := constructPlayer(*player1Type, *stoneCountPtr, *maxDepthPtr, *iterationPtr, *uctkPtr)
minimizer := constructPlayer(*player2Type, *stoneCountPtr, *maxDepthPtr, *iterationPtr, *uctkPtr)
rand.Seed(time.Now().UTC().UnixNano())
// func main's copy of the board.
var bd Board
for i := 0; i < 6; i++ {
bd.maxpits[i] = *stoneCountPtr
bd.minpits[i] = *stoneCountPtr
}
player := MAXIMIZER
GAMELOOP:
for {
fmt.Printf("%v\n> ", bd)
_, err := fmt.Scanf("\n")
if err != nil {
log.Print(err)
}
var pit, value int
var minNxt, maxNxt int
switch player {
case MAXIMIZER:
pit, value = maximizer.moveFn(maximizer.bd, false)
fmt.Printf("%s chooses %d (%d)\n", maximizer.name, pit, value)
maxNxt, _ = makeMove(&(maximizer.bd), pit, MAXIMIZER)
minNxt, _ = makeMove(&(minimizer.bd), pit, MINIMIZER)
if maxNxt != (0 - minNxt) {
fmt.Printf("maximizer says %d goes next\n", maxNxt)
fmt.Printf("minimizer says %d goes next\n", 0-minNxt)
}
case MINIMIZER:
pit, value = minimizer.moveFn(minimizer.bd, false)
fmt.Printf("%s chooses %d (%d)\n", minimizer.name, pit, value)
minNxt, _ = makeMove(&(minimizer.bd), pit, MAXIMIZER)
maxNxt, _ = makeMove(&(maximizer.bd), pit, MINIMIZER)
if maxNxt != (0 - minNxt) {
fmt.Printf("maximizer says %d goes next\n", maxNxt)
fmt.Printf("minimizer says %d goes next\n", 0-minNxt)
}
}
player, _ = makeMove(&bd, pit, player)
gameEnd, winner := checkEnd(&bd)
compare3(&bd, &(maximizer.bd), &(minimizer.bd))
if player != maxNxt || player != (0-minNxt) {
fmt.Printf("referee says %d goes next\n", player)
fmt.Printf("maximizer says %d goes next\n", maxNxt)
fmt.Printf("minimizer says %d goes next\n", 0-minNxt)
}
if gameEnd {
w := "player 1"
if winner == MINIMIZER {
w = "player 2"
}
fmt.Printf("Game over, %s won\n", w)
break GAMELOOP
}
}
fmt.Printf("Final:\n%v\n", bd)
}
func (p Board) String() string {
var top, mid, bot string
top = fmt.Sprintf(" %2d %2d %2d %2d %2d %2d\n",
p.maxpits[5],
p.maxpits[4],
p.maxpits[3],
p.maxpits[2],
p.maxpits[1],
p.maxpits[0])
bot = fmt.Sprintf(" %2d %2d %2d %2d %2d %2d",
p.minpits[0],
p.minpits[1],
p.minpits[2],
p.minpits[3],
p.minpits[4],
p.minpits[5])
mid = fmt.Sprintf("%2d %2d\n", p.maxpits[6], p.minpits[6])
return top + mid + bot
}
func (ab *AlphaBeta) chooseMove(bd Board, print bool) (bestpit int, bestvalue int) {
return chooseAlphaBeta(bd, ab.maxPly, print)
}
func chooseAlphaBeta(bd Board, maxPly int, print bool) (bestpit int, bestvalue int) {
bestvalue = 2 * LOSS // -infinity
bestpit = 0
var bd2 Board
for pit, stones := range bd.maxpits[0:6] {
if stones > 0 {
copy(bd2.maxpits[:], bd.maxpits[:])
copy(bd2.minpits[:], bd.minpits[:])
bd2.player = bd.player
makeMove(&bd2, pit, MAXIMIZER)
var value int
if end, winner := checkEnd(&bd2); end {
switch winner {
case MAXIMIZER:
value = WIN
case MINIMIZER:
value = LOSS
default: // end of game, but no winner
value = 0
}
} else {
value = alphaBeta(&bd2, 1, MINIMIZER, 2*LOSS, 2*WIN, maxPly)
}
if value > bestvalue {
bestvalue = value
bestpit = pit
}
// makeMove() does a lot to bd2, just dump it.
}
}
return bestpit, bestvalue
}
// alphaBeta does alpha-beta minimaxing. Computer is maximizer, human is minimizer.
// Pass current game board (bd *Board) by reference to avoid having the compiler
// create struct-copying code for each call to alphaBeta.
func alphaBeta(bd *Board, ply, player, alpha, beta int, maxPly int) (value int) {
if ply > maxPly {
// static value function: difference between pots less ply depth,
// so that all things equal, choose the shortest path to a win,
// plus some empirical amount of the seeds in computer's pits.
return (bd.maxpits[6] - bd.minpits[6]) - ply +
(bd.maxpits[0]+bd.maxpits[1]+bd.maxpits[2]+bd.maxpits[3]+bd.maxpits[4]+2*bd.maxpits[5])/3
}
// checkEnd() should get the case where someone already has
// more than half the stones in their pot, so alphaBeta()
// only has to do depth check
switch player {
case MAXIMIZER:
var bd2 Board
for pit, stones := range bd.maxpits[0:6] {
if stones != UNSET {
copy(bd2.maxpits[:], bd.maxpits[:])
copy(bd2.minpits[:], bd.minpits[:])
bd2.player = bd.player
nextplayer, plydelta := makeMove(&bd2, pit, player)
if end, winner := checkEnd(&bd2); end {
switch winner {
case MAXIMIZER:
value = WIN - ply
case MINIMIZER:
value = LOSS + ply
default:
value = 0
}
} else {
value = alphaBeta(&bd2, ply+plydelta, nextplayer, alpha, beta, maxPly)
}
if value > alpha {
alpha = value
}
if beta <= alpha {
return value
}
}
}
case MINIMIZER:
var bd2 Board
for pit, stones := range bd.minpits[0:6] {
if stones != 0 {
copy(bd2.maxpits[:], bd.maxpits[:])
copy(bd2.minpits[:], bd.minpits[:])
bd2.player = bd.player
nextplayer, plydelta := makeMove(&bd2, pit, player)
if end, winner := checkEnd(&bd2); end {
switch winner {
case MAXIMIZER:
value = WIN - ply
case MINIMIZER:
value = LOSS + ply
default:
value = 0
}
} else {
value = alphaBeta(&bd2, ply+plydelta, nextplayer, alpha, beta, maxPly)
}
if value < beta {
beta = value
}
if beta <= alpha {
return value
}
}
}
}
return value
}
func makeMove(bd *Board, pit int, player int) (nextplayer int, plydelta int) {
var sides [2]*[7]int
if pit > 5 {
fmt.Printf("problem player %d move %d, pit > 6: %s\n", player, pit, bd)
}
nextplayer = -player
plydelta = 1
switch player {
case MAXIMIZER:
sides[0] = &(bd.maxpits)
sides[1] = &(bd.minpits)
case MINIMIZER:
sides[0] = &(bd.minpits)
sides[1] = &(bd.maxpits)
}
S := 0 // side of player is always 0
hand := sides[S][pit]
sides[S][pit] = UNSET
if hand == 0 {
panic(fmt.Errorf("problem player %d move %d, empty pit:\n%s\n", player, pit, bd))
}
bonusmove := false
for i := pit + 1; hand > 0; {
// last stone, on player's side, last pit is empty,
// and pit across has stones.
if hand == 1 && S == 0 && i < 6 && sides[S][i] == 0 && sides[S^1][5-i] > 0 {
sides[S][6] += sides[S^1][5-i] + 1
sides[S^1][5-i] = 0
sides[S][i]-- // so no special cases just below
}
if !(S == 1 && i == 6) {
sides[S][i]++
hand--
}
if i == 6 {
i = 0
S ^= 1 // flip to other side of board
if hand == 0 {
bonusmove = true
}
} else {
i++
}
}
bd.player = player
if bonusmove {
nextplayer = player
plydelta = 0
}
return nextplayer, plydelta
}
// checkEnd figures out if the current game board, passed by reference
// to avoid compiler-generated struct copying, represents a win/loss/tie
// and for which player.
func checkEnd(bd *Board) (end bool, winner int) {
if bd.maxpits[6] > winningStonesCount {
return true, MAXIMIZER
}
if bd.minpits[6] > winningStonesCount {
return true, MINIMIZER
}
winner = UNSET
maxsidesum := 0
minsidesum := 0
for i := 0; i < 6; i++ {
maxsidesum += bd.maxpits[i]
minsidesum += bd.minpits[i]
}
if minsidesum == 0 || maxsidesum == 0 {
end = true
for i := 0; i < 6; i++ {
bd.maxpits[i] = UNSET
bd.minpits[i] = UNSET
}
bd.maxpits[6] += maxsidesum
bd.minpits[6] += minsidesum
}
if end {
winner = bd.maxpits[6] - bd.minpits[6]
// Ties can happen, winner == 0 in that case, which == UNSET
switch {
case winner > 0:
winner = MAXIMIZER
case winner < 0:
winner = MINIMIZER
}
}
return end, winner
}
type gameState struct {
board Board
}
type Node struct {
move int
player int
childNodes []*Node
untriedMoves []int
parent *Node
visits int
wins float64
}
// chooseMonteCarlo - based on current board, return the best pit
// for MAXIMIZER to pick up and drop down the board.
func (p *MCTS) chooseMonteCarlo(bd Board, print bool) (bestpit int, value int) {
root := &Node{
player: MINIMIZER, // opponent made last move
untriedMoves: make([]int, 0, 6),
}
// by definition the next player is MAXIMIZER.
// Fill in MAXIMIMIZER's untried moves
for i := 0; i < 6; i++ {
if bd.maxpits[i] != 0 {
root.untriedMoves = append(root.untriedMoves, i)
}
}
state := &Board{}
for iter := 0; iter < p.iterations; iter++ {
// reset game state tracker
for i := 0; i < 7; i++ {
state.maxpits[i] = bd.maxpits[i]
state.minpits[i] = bd.minpits[i]
}
state.player = root.player
nextPlayer := -root.player
node := root
// Selection
for len(node.untriedMoves) == 0 && len(node.childNodes) > 0 {
node = node.selectBestChild(p.uctk)
// Filling state from a game tree, so use node.move, node.player,
// ignoring nextPlayer for now.
nextPlayer, _ = makeMove(state, node.move, node.player)
}
gameEnd, winner := checkEnd(state)
// Expansion
if !gameEnd && len(node.untriedMoves) > 0 {
mv := node.randomUntried()
nextPlayer, _ = makeMove(state, mv, nextPlayer)
node = node.addChild(mv, nextPlayer, state)
gameEnd, winner = checkEnd(state)
}
// Simulation
if !gameEnd {
for !gameEnd {
mv := state.randomMove(nextPlayer)
nextPlayer, _ = makeMove(state, mv, nextPlayer)
gameEnd, winner = checkEnd(state)
}
}
// Back propagation
for node != nil {
node.visits++
if winner == node.player {
node.wins++
} else if winner == 0 {
node.wins += 0.5
}
node = node.parent
}
}
// Select child move with the largest number of visits
bestChild := root.childNodes[0]
mostVisits := bestChild.visits
for _, c := range root.childNodes[1:] {
if c.visits > mostVisits {
bestChild = c
mostVisits = bestChild.visits
}
}
return bestChild.move, int(bestChild.wins / float64(bestChild.visits) * 100.)
}
func (bd *Board) randomMove(player int) int {
if player == MAXIMIZER {
for {
i := rand.Intn(6)
if bd.maxpits[i] != 0 {
return i
}
}
}
for {
i := rand.Intn(6)
if bd.minpits[i] != 0 {
return i
}
}
fmt.Printf("Board.randomMove(%d) shouldn't get here\n", player)
return 0
}
func (n *Node) randomUntried() int {
ln := len(n.untriedMoves)
randIdx := rand.Intn(ln)
ln--
mv := n.untriedMoves[randIdx]
n.untriedMoves[randIdx] = n.untriedMoves[ln]
n.untriedMoves = n.untriedMoves[:ln]
return mv
}
func (n *Node) addChild(mv int, nextPlayer int, state *Board) *Node {
if mv > 5 {
fmt.Printf("addChild, move %d illegal\n", mv)
fmt.Printf("parent node: %d/%d, untried moves %v\n",
n.move, n.player, n.untriedMoves)
fmt.Printf("next player %d, state.player %d\n%s\n",
nextPlayer, state.player, state)
panic("bad child move")
}
newChild := &Node{
move: mv,
player: state.player,
parent: n,
untriedMoves: remainingMoves(state, nextPlayer),
}
n.childNodes = append(n.childNodes, newChild)
return newChild
}
func (n *Node) selectBestChild(uctk float64) *Node {
bestScore := n.childNodes[0].ucb1(uctk)
bestChild := n.childNodes[0]
for _, c := range n.childNodes[1:] {
score := c.ucb1(uctk)
if score > bestScore {
bestScore = score
bestChild = c
}
}
return bestChild
}
func remainingMoves(bd *Board, player int) []int {
mvs := make([]int, 0, 6)
if player == MAXIMIZER {
for i := 0; i < 6; i++ {
if bd.maxpits[i] != 0 {
mvs = append(mvs, i)
}
}
return mvs
}
for i := 0; i < 6; i++ {
if bd.minpits[i] != 0 {
mvs = append(mvs, i)
}
}
return mvs
}
func (n *Node) ucb1(uctk float64) float64 {
v := float64(n.visits)
return n.wins/v +
uctk*math.Sqrt(math.Log(float64(n.parent.visits+1))/v)
}
/*
type chooserFunction func(bd Board, print bool) (bestpit int, bestvalue int)
type player struct {
bd Board
moveFn chooserFunction
}
*/
func constructPlayer(typ string, stonesPerPit int, maxDepth int, mctsIterations int, uctk float64) *player {
var p player
for i := 0; i < 6; i++ {
p.bd.maxpits[i] = stonesPerPit
p.bd.minpits[i] = stonesPerPit
}
switch typ {
case "M": // MCTS+UCB1
mcts := &MCTS{iterations: mctsIterations, uctk: uctk}
p.moveFn = mcts.chooseMonteCarlo
p.name = "MCTS"
case "A": // Alpha-beta minimaxing
// func alphaBeta(bd *Board, ply, player, alpha, beta int, maxPly int) (value int) {
ab := &AlphaBeta{maxPly: 2 * maxDepth}
p.moveFn = ab.chooseMove
p.name = "A/B"
default:
fmt.Fprintf(os.Stderr, "Unknown player type %q\n", typ)
}
return &p
}
func compare3(ref, max, min *Board) {
for i := 0; i < 7; i++ {
if ref.maxpits[i] != max.maxpits[i] ||
ref.maxpits[i] != min.minpits[i] ||
ref.minpits[i] != max.minpits[i] ||
ref.minpits[i] != min.maxpits[i] {
fmt.Printf("Boards disagree:\n")
fmt.Printf("referee:\n%v\n", ref)
fmt.Printf("maximizer:\n%v\n", max)
fmt.Printf("minimizer:\n%v\n", min)
}
}
}