int search_uct(int color, int node_n)
{
NODE *pN = &node[node_n];
CHILD *c = NULL;
int select, z, err, win, current_depth;
for (;;) {
select = select_best_ucb(node_n, color);
c = &pN->child[select];
z = c->z;
err = put_stone(z, color, FILL_EYE_ERR);
if ( err == 0 ) break;
c->z = ILLEGAL_Z; // select other move
}
current_depth = depth;
path[depth++] = c->z;
// playout in first time. <= 10 can reduce node.
if ( c->games <= 0 || depth == D_MAX || (c->z == 0 && depth>=2 && path[depth-2]==0) ) {
win = -playout(flip_color(color));
} else {
if ( c->next == NODE_EMPTY ) c->next = create_node(c->z);
win = -search_uct(flip_color(color), c->next);
}
update_rave(pN, color, current_depth, win);
// update winrate
c->rate = (c->rate * c->games + win) / (c->games + 1);
c->games++;
pN->child_games_sum++;
return win;
}
int main()
{
int color = 1;
int z;
srand( (unsigned)time( NULL ) );
for (;;) {
z = play_one_move(color);
printf("moves=%4d, color=%d, z=%d\n",moves, color, get81(z));
print_board();
record[moves] = z;
moves++;
if ( moves == MAX_MOVES ) {
printf("max moves!\n");
break;
}
if ( z == 0 && moves >= 2 && record[moves-2] == 0 ) {
printf("two pass\n");
break;
}
color = flip_color(color);
}
return 0;
}
int playout(int turn_color)
{
int color = turn_color;
int previous_z = 0;
int loop;
int loop_max = B_SIZE*B_SIZE + 200; // for triple ko
all_playouts++;
for (loop=0; loop<loop_max; loop++) {
// all empty points are candidates.
int empty[BOARD_MAX][2]; // [0]...z, [1]...probability
int empty_num = 0;
int prob_sum = 0;
int x,y,z,err,pr;
for (y=0;y<B_SIZE;y++) for (x=0;x<B_SIZE;x++) {
int z = get_z(x+1,y+1);
if ( board[z] != 0 ) continue;
empty[empty_num][0] = z;
pr = 100;
// pr = get_prob(z, previous_z, color);
empty[empty_num][1] = pr;
prob_sum += pr;
empty_num++;
}
for (;;) {
int i = 0;
if ( empty_num == 0 ) {
z = 0;
} else {
int r = rand() % prob_sum;
int sum = 0;
for (i=0; i<empty_num; i++) {
sum += empty[i][1]; // 0,1,2 [0]=1, [1]=1, [2]=1
if ( sum > r ) break;
}
if ( i==empty_num ) { prt("Err! prob_sum=%d,sum=%d,r=%d,r=%d\n",prob_sum,sum,r,i); exit(0); }
z = empty[i][0];
}
err = put_stone(z, color, FILL_EYE_ERR);
if ( err == 0 ) break; // pass is ok.
prob_sum -= empty[i][1];
empty[i][0] = empty[empty_num-1][0]; // err, delete
empty[i][1] = empty[empty_num-1][1];
empty_num--;
}
if ( flag_test_playout ) record[moves++] = z;
if ( depth < D_MAX ) path[depth++] = z;
if ( z == 0 && previous_z == 0 ) break; // continuous pass
previous_z = z;
// prt("loop=%d,z=%s,c=%d,empty_num=%d,ko_z=%d\n",loop,get_char_z(z),color,empty_num,ko_z);
color = flip_color(color);
}
return count_score(turn_color);
}
void hexbright::print_number(long number) {
// reverse number (so it prints from left to right)
BOOL negative = false;
if(number<0) {
number = 0-number;
negative = true;
}
_color = GLED;
_number=1; // to guarantee printing when dealing with trailing zeros (100 can't be stored as 001, use 1001)
do {
_number = _number * 10 + (number%10);
number = number/10;
_color = flip_color(_color);
} while(number>0);
if(negative) {
set_led(flip_color(_color), 500);
print_wait_time = 600/update_delay;
}
}
// put stone. success returns 0. in playout, fill_eye_err = 1.
int put_stone(int tz, int color, int fill_eye_err)
{
int around[4][3];
int un_col = flip_color(color);
int space = 0;
int wall = 0;
int mycol_safe = 0;
int capture_sum = 0;
int ko_maybe = 0;
int liberty, stone;
int i;
if (tz == 0) { ko_z = 0; return 0; } // pass
// count 4 neighbor's liberty and stones.
for (i = 0; i<4; i++) {
int z, c, liberty, stone;
around[i][0] = around[i][1] = around[i][2] = 0;
z = tz + dir4[i];
c = board[z]; // color
if (c == 0) space++;
if (c == 3) wall++;
if (c == 0 || c == 3) continue;
count_liberty(z, &liberty, &stone);
around[i][0] = liberty;
around[i][1] = stone;
around[i][2] = c;
if (c == un_col && liberty == 1) { capture_sum += stone; ko_maybe = z; }
if (c == color && liberty >= 2) mycol_safe++;
}
if (capture_sum == 0 && space == 0 && mycol_safe == 0) return 1; // suicide
if (tz == ko_z) return 2; // ko
if (wall + mycol_safe == 4 && fill_eye_err) return 3; // eye
if (board[tz] != 0) return 4;
for (i = 0; i<4; i++) {
int lib = around[i][0];
int c = around[i][2];
if (c == un_col && lib == 1 && board[tz + dir4[i]]) {
take_stone(tz + dir4[i], un_col);
}
}
board[tz] = color;
count_liberty(tz, &liberty, &stone);
if (capture_sum == 1 && stone == 1 && liberty == 1) ko_z = ko_maybe;
else ko_z = 0;
return 0;
}
int primitive_monte_calro(int color)
{
int try_num = 30; // number of playout
int best_z = 0;
double best_value;
double win_rate;
int x,y,err,i,win_sum,win;
int ko_z_copy;
int board_copy[BOARD_MAX]; // keep current board
ko_z_copy = ko_z;
memcpy(board_copy, board, sizeof(board));
best_value = -100;
// try all empty point
for (y=0;y<B_SIZE;y++) for (x=0;x<B_SIZE;x++) {
int z = get_z(x+1,y+1);
if ( board[z] != 0 ) continue;
err = put_stone(z, color, FILL_EYE_ERR);
if ( err != 0 ) continue;
win_sum = 0;
for (i=0;i<try_num;i++) {
int board_copy2[BOARD_MAX];
int ko_z_copy2 = ko_z;
memcpy(board_copy2, board, sizeof(board));
win = -playout(flip_color(color));
win_sum += win;
ko_z = ko_z_copy2;
memcpy(board, board_copy2, sizeof(board));
}
win_rate = (double)win_sum / try_num;
// print_board();
// prt("z=%d,win=%5.3f\n",get81(z),win_rate);
if ( win_rate > best_value ) {
best_value = win_rate;
best_z = z;
// prt("best_z=%d,color=%d,v=%5.3f,try_num=%d\n",get81(best_z),color,best_value,try_num);
}
ko_z = ko_z_copy;
memcpy(board, board_copy, sizeof(board)); // resume board
}
return best_z;
}
void hexbright::update_number() {
if(_number>0) { // we have something to do...
#ifdef DEBUG
if(DEBUG==DEBUG_NUMBER) {
static int last_printed = 0;
if(last_printed != _number) {
last_printed = _number;
Serial.print("number remaining (read from right to left): ");
Serial.println(_number);
}
}
#endif
if(!print_wait_time) {
if(_number==1) { // minimum delay between printing numbers
print_wait_time = 2500/update_delay;
_number = 0;
return;
} else {
print_wait_time = 300/update_delay;
}
if(_number/10*10==_number) {
#if (DEBUG==DEBUG_NUMBER)
Serial.println("zero");
#endif
// print_wait_time = 500/update_delay;
set_led(_color, 400);
} else {
set_led(_color, 120);
_number--;
}
if(_number && !(_number%10)) { // next digit?
print_wait_time = 600/update_delay;
_color = flip_color(_color);
_number = _number/10;
}
}
}
if(print_wait_time) {
print_wait_time--;
}
}
void selfplay()
{
int color = 1;
int z,search;
for (;;) {
if ( color == 1 ) {
search = SEARCH_UCT; //SEARCH_PRIMITIVE;
} else {
search = SEARCH_UCT;
}
z = get_computer_move(color, search);
add_moves(z, color);
if ( z == 0 && moves > 1 && record[moves-2] == 0 ) break;
if ( moves > 300 ) break; // too long
color = flip_color(color);
}
print_sgf();
}
static struct npr_rbtree_node *
insert(struct npr_rbtree *t, struct npr_rbtree_node *n,
npr_rbtree_key_t key,
npr_rbtree_value_t v,
int *insert_new)
{
if (n == NULL) {
n = alloc_node(t);
n->left = NULL;
n->right = NULL;
n->v = v;
n->key = key;
n->color = RED;
*insert_new = 1;
return n;
}
if (is_red(n->left) && is_red(n->right)) {
flip_color(n);
}
if (key == n->key) {
n->v = v;
} else if (key < n->key) {
n->left = insert(t, n->left, key, v, insert_new);
} else {
n->right = insert(t, n->right, key, v, insert_new);
}
if (is_red(n->right)) {
n = rotate_left(n);
}
if (is_red(n->left) && is_red(n->left->left)) {
n = rotate_right(n);
}
return n;
}
void update_rave(NODE *pN, int color, int current_depth, double win)
{
int played_color[BOARD_MAX];
int i,z;
int c = color;
memset(played_color, 0, sizeof(played_color));
for (i=current_depth; i<depth; i++) {
z = path[i];
if ( played_color[z] == 0 ) played_color[z] = c;
c = flip_color(c);
}
played_color[0] = 0; // ignore pass
for (i=0; i<pN->child_num; i++) {
CHILD *c = &pN->child[i];
if ( c->z == ILLEGAL_Z ) continue;
if ( played_color[c->z] != color ) continue;
c->rave_rate = (c->rave_games * c->rave_rate + win) / (c->rave_games + 1);
c->rave_games++;
pN->child_rave_games_sum++;
}
}
int main()
{
int color = 1; // 現在の手番の色。黒が1で白が2
int tesuu = 0; // 手数
hama[0] = 0;
hama[1] = 0;
bool pray_pass = false;
srand( (unsigned)time( NULL ) );
char str[256];
char playstr[256];
// stdoutをバッファリングしないように。GTPで通信に失敗するので。
setbuf(stdout, NULL);
setbuf(stderr, NULL); // stderrに書くとGoGuiに表示される。
//loop:
// 盤面初期化
{int i,x,y; for (i=0;i<BOARD_MAX;i++) board[i] = 3; for (y=0;y<B_SIZE;y++) for (x=0;x<B_SIZE;x++) board[get_z(x,y)] = 0; }
// static int score_sum;
// static int loop_count;
for (;;) {
if ( fgets(str, 256, stdin)==NULL ) break; // 標準入力から読む
//fprintf(stderr, playstr);
//strcpy(playstr,str);
if ( strstr(str,"boardsize") ) {
send_gtp("= \n\n"); // "boardsize 19" 19路盤
}
else if ( strstr(str,"clear_board") ) {
for (int i=0;i<BOARD_MAX;i++) board[i] = 3;
for (int y=0;y<B_SIZE;y++)
for (int x=0;x<B_SIZE;x++)
board[get_z(x,y)] = 0;
tesuu = 0;
color = 1;
hama[0] = 0;
hama[1] = 0;
pray_pass = false;
node_num = 0;
send_gtp("= \n\n");
}
else if ( strstr(str,"name") ) send_gtp("= ayasam\n\n");
else if ( strstr(str,"version") ) send_gtp("= 0.0.1\n\n");
else if ( strstr(str,"genmove w") ) {
if (pray_pass) {
send_gtp("= pass\n\n");
} else {
color = 2;
clock_t bt = clock();
all_playouts = 0; // playout回数を初期化
#if 0 // 0 でUCT探索
int z = select_best_move(color); // 原始モンテカルロ
#else
int z = select_best_uct(color); // UCT
#endif
int err = move(z,color); // 打ってみる
if ( err != 0 ) {
show_board();
fprintf( stderr, "Err!\n");
char mvstr[8];
change_z_str(mvstr, z);
fprintf(stderr, "%s\n", mvstr);
exit(0);
}
kifu[tesuu] = z;
tesuu++;
print_board();
char gtstr[12] = "= ";
char mvstr[8];
change_z_str(mvstr, z);
strcat(gtstr, mvstr);
strcat(gtstr, "\n\n");
send_gtp(gtstr);
fprintf(stderr,"play_xy = %s,手数=%d,色=%d,all_playouts=%d\n",mvstr,tesuu,color,all_playouts);
double t = (double)(clock()+1 - bt) / CLOCKS_PER_SEC;
fprintf(stderr,"%.1f 秒, %.0f playout/秒\n",t,all_playouts/t);
}
}
else if ( strstr(str,"genmove b") ) {
if (pray_pass) {
send_gtp("= pass\n\n");
} else {
color = 1;
clock_t bt = clock();
all_playouts = 0; // playout回数を初期化
#if 0 // 0 でUCT探索
int z = select_best_move(color); // 原始モンテカルロ
#else
int z = select_best_uct(color); // UCT
#endif
int err = move(z,color); // 打ってみる
if ( err != 0 ) {
show_board();
fprintf( stderr, "Err!\n");
char mvstr[8];
change_z_str(mvstr, z);
fprintf(stderr, "%s\n", mvstr);
exit(0);
}
kifu[tesuu] = z;
tesuu++;
print_board();
char gtstr[12] = "= ";
char mvstr[8];
change_z_str(mvstr, z);
strcat(gtstr, mvstr);
strcat(gtstr, "\n\n");
send_gtp(gtstr);
fprintf(stderr,"play_xy = %s,手数=%d,色=%d,all_playouts=%d\n",mvstr,tesuu,color,all_playouts);
double t = (double)(clock()+1 - bt) / CLOCKS_PER_SEC;
fprintf(stderr,"%.1f 秒, %.0f playout/秒\n",t,all_playouts/t);
}
}
// 相手の手を受信 "play W D17" のように来る
else if ( strstr(str,"play") ) {
char *tp;
tp = strtok(str, " ");
tp = strtok(NULL, " ");
if (strstr(tp, "W") || strstr(tp, "w")) {
color = 2;
} else {
color = 1;
}
if (tp) {
tp = strtok(NULL, " ");
int z = change_str_z(tp);
int err = play_move(z,color); // 打ってみる
if ( (err != 0) && (err != 3) ) {
show_board();
fprintf( stderr, "Err!\n");
char mvstr[8];
change_z_str(mvstr, z);
fprintf(stderr, "%s\n", mvstr);
strcat(playstr,"Error!:");
if (err == 1) {
strcat(playstr,"1");
} else if (err == 2) {
strcat(playstr,"2");
} else if (err == 3) {
strcat(playstr,"3");
} else if (err == 4) {
strcat(playstr,"4");
}
strcat(playstr," ");
strcat(playstr,tp);
strcat(playstr," -> ");
strcat(playstr,mvstr);
strcat(playstr,"\n");
exit(0);
}
kifu[tesuu] = z;
tesuu++;
if (z == 0) {
pray_pass = true;
} else {
pray_pass = false;
}
print_board();
}
send_gtp("= \n\n");
}
else if ( strstr(str, "showboard")) {
show_board();
send_gtp("= \n\n");
}
else if ( strstr(str, "quit")) {
send_gtp("= \n\n");
break;
}
else if ( strstr(str,"test")) {
for (int n=0;;n++) {
for (int i=0;i<BOARD_MAX;i++) board[i] = 3;
for (int y=0;y<B_SIZE;y++)
for (int x=0;x<B_SIZE;x++)
board[get_z(x,y)] = 0;
tesuu = 0;
color = 1;
hama[0] = 0;
hama[1] = 0;
pray_pass = false;
while (true) {
if (pray_pass) {
break;
} else {
clock_t bt = clock();
all_playouts = 0; // playout回数を初期化
#if 0 // 0 でUCT探索
int z = select_best_move(color); // 原始モンテカルロ
#else
int z = select_best_uct(color); // UCT
#endif
int err = move(z,color); // 打ってみる
if ( err != 0 ) {
show_board();
fprintf( stderr, "Err!\n");
char mvstr[8];
change_z_str(mvstr, z);
fprintf(stderr, "%s\n", mvstr);
exit(0);
}
//kifu[tesuu] = z;
tesuu++;
//print_board();
//char gtstr[12] = "= ";
char mvstr[8];
change_z_str(mvstr, z);
//strcat(gtstr, mvstr);
//strcat(gtstr, "\n\n");
//send_gtp(gtstr);
fprintf(stderr,"play_xy = %s,手数=%d,色=%d,all_playouts=%d\n",mvstr,tesuu,color,all_playouts);
double t = (double)(clock()+1 - bt) / CLOCKS_PER_SEC;
fprintf(stderr,"%.1f 秒, %.0f playout/秒\n",t,all_playouts/t);
color = flip_color(color);
if (z == 0) {
pray_pass = true;
}
}
}
print_board();
printf("%d end\n", n);
}
}
else {
send_gtp("= \n\n"); // それ以外のコマンドにはOKを返す
}
show_board();
}
/*
// 自己対戦のテスト用
int score = count_score(1);
score_sum += score;
loop_count++;
FILE *fp = fopen("out.txt","a+");
fprintf(fp,"Last Score=%d,%d,%d,tesuu=%d\n",score,score_sum,loop_count,tesuu);
fclose(fp);
printf("Last Score=%d,%d,%d,tesuu=%d\n",score,score_sum,loop_count,tesuu); color = 1; tesuu = 0; goto loop;
*/
// SGFで棋譜を吐き出す
/*
printf("(;GM[1]SZ[%d]\r\n",B_SIZE);
for (int i=0; i<tesuu; i++) {
int z = kifu[i];
int y = z / WIDTH;
int x = z - y*WIDTH;
if ( z == 0 ) x = y = 20;
char *sStone[2] = { "B", "W" };
printf(";%s[%c%c]",sStone[i&1], x+'a'-1, y+'a'-1 );
if ( ((i+1)%10)==0 ) printf("\r\n");
}
printf("\r\n)\r\n");
*/
return 0;
}
