search

SanchoGGP / ggp-base

The General Game Playing Base Package
8 stars 4 forks source link

Shortest win not taken with mate in 1 at SpeedChess #169

Open SteveDraper opened 10 years ago

SteveDraper commented 10 years ago

http://www.ggp.org/view/tiltyard/matches/cffa1a39465b7a18f994370423fa3b3c4df64d7c/ move 40 it chose a continuation with mate in 2+ whereas it had a mate in 1. It's possible that heuristic selection means it completed the deeper path before the shallower one (and thus never spotted the shorter completion) but i seems worthy of investigation

SteveDraper commented 10 years ago

See alo mov 33 in http://www.ggp.org/view/tiltyard/matches/de1b13bee8bc14633208fff1b9bf01ffce55321b/ which is mate in 1 but is not taken

arr28 commented 9 years ago

I thought for a minute that we weren't taking an immediately forced win, but in speedChess, you need to actually take the King, so mate in 1 is actually 3 moves to the end of the game (mate, futile move, capture).

Looking at the graphs, in the first game, the tree was complete a couple of moves earlier, so we'd stopped searching. As you say, we may have completed the deeper path before the shallower one. In the second game, however, we hadn't stopped searching yet. We only had an expected score of ~75. Something must have gone wrong for us not to see a win that's just three moves away.

SteveDraper commented 9 years ago

3 moves is actually a LOT of tree nodes in SpeedChess. If queens are on the board it often has a branching factor of 40-50, so if one of the moves is also anti-heuristic it can easily not be searched.

On Thu, Apr 16, 2015 at 1:38 PM, Andrew Rose notifications@github.com wrote:

I thought for a minute that we weren't taking an immediately forced win, but in speedChess, you need to actually take the King, so mate in 1 is actually 3 moves to the end of the game (mate, futile move, capture).

Looking at the graphs, in the first game, the tree was complete a couple of moves earlier, so we'd stopped searching. As you say, we may have completed the deeper path before the shallower one. In the second game, however, we hadn't stopped searching yet. We only had an expected score of ~75. Something must have gone wrong for us not to see a win that's just three moves away.

— Reply to this email directly or view it on GitHub https://github.com/SanchoGGP/ggp-base/issues/169#issuecomment-93810862.