You may turn on a log of all games in a rollout. If you do so you observe that the error when using the cli is caused by gnubg not doubling on the first turn as it should. Later doubles for both players are made correctly (not relevant for this position) and anything else you may think of is seemingly correct. Is it possible that the rollout position is initialized so that gnubg skips the doubling decision?
Christian. On 8/21/06, [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote:
Hi all, this bug in in the bug-tracker on gnubg.org, but it looks so huge to me that it's maybe worth checking if anybody is having a look at it. GNU Backgammon Position ID: GwAAYAMAAAAAAA Match ID : cAkFAAAAAAAA +24-23-22-21-20-19------18-17-16-15-14-13-+ O: gnubg OOO | O O | | | 0 points OO | O O | | | OO | | | | OO | | | | OO | | | | | |BAR| |v (Cube: 1) XX | | | | XX | | | | XX | | | | XX | X X | | | Rolled 21 XXX | X X | | | 0 points +-1--2--3--4--5--6-------7--8--9-10-11-12-+ X: MaX A hint says : 1. Cubeful 2-ply 2/off 1/off Eq.: +0.722 86.1% 0.0% 0.0% - 13.9% 0.0% 0.0% 2-ply cubeful prune [world class] 2. Cubeful 2-ply 2/1 2/off Eq.: -1.000 ( -1.722) 14.4% 0.0% 0.0% - 85.6% 0.0% 0.0% 2-ply cubeful prune [world class] Notice that after 2/1 2/off, player 1 will double out player 2 (hence the equity of -1.000). Now selecting the two moves and doing a rollout we get the same result (equities of +0.722 and -1.000) in the GUI version, but a different result in the command-line (CLI) version : it gives equities of +0.722 and -0.713. As pointed out by the person that reported the bug (Philippe Michel), -0.713 is the cubeless equity. It is not a coincidence, since the same behavior appears for the following position : GNU Backgammon Position ID: GwAAYAsAAAAAAA Match ID : cAkFAAAAAAAA +24-23-22-21-20-19------18-17-16-15-14-13-+ O: gnubg OOO | O O | | | 0 points OO | O O | | | OO | | | | OO | | | | OO | | | | | |BAR| |v (Cube: 1) XX | | | | XX | | | | XX | | | | XX | X X | | | Rolled 21 XX | X X X | | | 0 points +-1--2--3--4--5--6-------7--8--9-10-11-12-+ X: MaX Here, no matter the move, player 1 is gonna double out player 2 next turn. While GUI version correctly shows equities equal to -1.000 after a rollout, CLI version shows cubeless equities ... (MaX) hint 1. Cubeful 2-ply 2/off 1/off Eq.: -1.000 12.0% 0.0% 0.0% - 88.0% 0.0% 0.0% 2-ply cubeful prune [world class] 2. Cubeful 2-ply 3/off Eq.: -1.000 ( +0.000) 12.0% 0.0% 0.0% - 88.0% 0.0% 0.0% 2-ply cubeful prune [world class] 3. Cubeful 2-ply 2/1 2/off Eq.: -1.000 ( +0.000) 9.6% 0.0% 0.0% - 90.4% 0.0% 0.0% 2-ply cubeful prune [world class] 4. Cubeful 2-ply 3/1 2/1 Eq.: -1.000 ( +0.000) 0.0% 0.0% 0.0% - 100.0% 0.0% 0.0% 2-ply cubeful prune [world class] (MaX) rollout =1 =2 =3 =4 1. Rollout 2/off 1/off Eq.: -0.761 12.0% 0.0% 0.0% - 88.0% 0.0% 0.0% CL -0.761 CF -0.761 [ 0.0% 0.0% 0.0% - 0.0% 0.0% 0.0% CL 0.000 CF 0.000] Truncated cubeful rollout (depth 11) with var.redn. 648 games, Mersenne Twister dice gen. with seed 1036242101 and quasi-ran dom dice Stop when best play is enough JSDs ahead: limit 1.96 (min. 144 games) Play: world class 2-ply cubeful prune [world class] keep the first 0 0-ply moves and up to 8 more moves within equity 0.16 Skip pruning for 1-ply moves. Cube: 2-ply cubeful prune [world class] 2. Rollout 3/off Eq.: -0.761 ( +0.000) 12.0% 0.0% 0.0% - 88.0% 0.0% 0.0% CL -0.761 CF -0.761 [ 0.0% 0.0% 0.0% - 0.0% 0.0% 0.0% CL 0.000 CF 0.000] Truncated cubeful rollout (depth 11) with var.redn. 648 games, Mersenne Twister dice gen. with seed 1036242101 and quasi-ran dom dice Stop when best play is enough JSDs ahead: limit 1.96 (min. 144 games) Play: world class 2-ply cubeful prune [world class] keep the first 0 0-ply moves and up to 8 more moves within equity 0.16 Skip pruning for 1-ply moves. Cube: 2-ply cubeful prune [world class] 3. Rollout 2/1 2/off Eq.: -0.809 ( -0.048) 9.6% 0.0% 0.0% - 90.4% 0.0% 0.0% CL -0.809 CF -0.809 [ 0.0% 0.0% 0.0% - 0.0% 0.0% 0.0% CL 0.000 CF 0.000] Truncated cubeful rollout (depth 11) with var.redn. 144 games, Mersenne Twister dice gen. with seed 1036242101 and quasi-ran dom dice Stop when best play is enough JSDs ahead: limit 1.96 (min. 144 games) Play: world class 2-ply cubeful prune [world class] keep the first 0 0-ply moves and up to 8 more moves within equity 0.16 Skip pruning for 1-ply moves. Cube: 2-ply cubeful prune [world class] 4. Rollout 3/1 2/1 Eq.: -1.000 ( -0.239) 0.0% 0.0% 0.0% - 100.0% 0.0% 0.0% CL -1.000 CF -1.000 [ 0.0% 0.0% 0.0% - 0.0% 0.0% 0.0% CL 0.000 CF 0.000] Truncated cubeful rollout (depth 11) with var.redn. 144 games, Mersenne Twister dice gen. with seed 1036242101 and quasi-ran dom dice Stop when best play is enough JSDs ahead: limit 1.96 (min. 144 games) Play: world class 2-ply cubeful prune [world class] keep the first 0 0-ply moves and up to 8 more moves within equity 0.16 Skip pruning for 1-ply moves. Cube: 2-ply cubeful prune [world class] Pretty scary to me ... MaX _______________________________________________ Bug-gnubg mailing list Bug-gnubg@gnu.org http://lists.gnu.org/mailman/listinfo/bug-gnubg
_______________________________________________ Bug-gnubg mailing list Bug-gnubg@gnu.org http://lists.gnu.org/mailman/listinfo/bug-gnubg