Side to Move Bonus---does it help?
Moderators: hgm, Rebel, chrisw
-
- Posts: 1494
- Joined: Thu Mar 30, 2006 2:08 pm
Side to Move Bonus---does it help?
Has anyone done testing on the Side to Move bonus, and if so, how much does it help (if anything)? It would seem simple enough to test some public source programs with and without the bonus to see if it helps. Mayve Bob can use his amazing cluster to see the effect...it might be small and require a lot of games.
-
- Posts: 12541
- Joined: Wed Mar 08, 2006 8:57 pm
- Location: Redmond, WA USA
Re: Side to Move Bonus---does it help?
During the midgame, a tempo appears to be worth about 9 centipawns.mjlef wrote:Has anyone done testing on the Side to Move bonus, and if so, how much does it help (if anything)? It would seem simple enough to test some public source programs with and without the bonus to see if it helps. Mayve Bob can use his amazing cluster to see the effect...it might be small and require a lot of games.
I guess that the result will be specific to individual programs.
-
- Posts: 20943
- Joined: Mon Feb 27, 2006 7:30 pm
- Location: Birmingham, AL
Re: Side to Move Bonus---does it help?
It is _very_ small. I have one in Crafty. +5 for side on move in middlegame, +8 in the endgame. Worth a couple of Elo. And we spent a good bit of time testing different values to reach those as optimal [for crafty of course, YMMV].mjlef wrote:Has anyone done testing on the Side to Move bonus, and if so, how much does it help (if anything)? It would seem simple enough to test some public source programs with and without the bonus to see if it helps. Mayve Bob can use his amazing cluster to see the effect...it might be small and require a lot of games.
-
- Posts: 201
- Joined: Thu Mar 22, 2007 7:12 pm
- Location: Netherlands
Re: Side to Move Bonus---does it help?
Rotor uses a bonus of 5 cp, this was roughly the optimum when minimizing tree search sizes for the LCT II set of positions.mjlef wrote:Has anyone done testing on the Side to Move bonus, and if so, how much does it help (if anything)? It would seem simple enough to test some public source programs with and without the bonus to see if it helps. Mayve Bob can use his amazing cluster to see the effect...it might be small and require a lot of games.
-
- Posts: 20943
- Joined: Mon Feb 27, 2006 7:30 pm
- Location: Birmingham, AL
Re: Side to Move Bonus---does it help?
Not sure how minimizing search space infers anything at all about an optimal value for any scoring term...Jan Brouwer wrote:Rotor uses a bonus of 5 cp, this was roughly the optimum when minimizing tree search sizes for the LCT II set of positions.mjlef wrote:Has anyone done testing on the Side to Move bonus, and if so, how much does it help (if anything)? It would seem simple enough to test some public source programs with and without the bonus to see if it helps. Mayve Bob can use his amazing cluster to see the effect...it might be small and require a lot of games.
-
- Posts: 719
- Joined: Thu Mar 09, 2006 1:21 am
- Location: Portland Oregon
Re: Side to Move Bonus---does it help?
The side to move bonus values are the only eval values that are the same in Fruit now as they where in Fruit 2.2 and it makes a huge negative difference to take the bonus out. I have always been uncomfortable with a huge side to move bonus but in Fruits case it works. I am sure there is a better why to do tempo I just have not figured it out myself yet. Maybe some type of table full of move bonus scores for each to and from for each piece given the game phase. Also some engines apply the bonus outside of eval so the bonus is not used for pruning decisions.
-
- Posts: 719
- Joined: Thu Mar 09, 2006 1:21 am
- Location: Portland Oregon
Re: Side to Move Bonus---does it help?
Fewer nodes searched the less time taken. One would hope that is best solution sooner.bob wrote:Not sure how minimizing search space infers anything at all about an optimal value for any scoring term...Jan Brouwer wrote:Rotor uses a bonus of 5 cp, this was roughly the optimum when minimizing tree search sizes for the LCT II set of positions.mjlef wrote:Has anyone done testing on the Side to Move bonus, and if so, how much does it help (if anything)? It would seem simple enough to test some public source programs with and without the bonus to see if it helps. Mayve Bob can use his amazing cluster to see the effect...it might be small and require a lot of games.
-
- Posts: 6401
- Joined: Thu Mar 09, 2006 8:30 pm
- Location: Chicago, Illinois, USA
Re: Side to Move Bonus---does it help?
There was a long thread started by Don Dailey some time ago. A given sidetomove bonus wll decrease the tree size because it smooth the jumps in eval from ply to ply. It is possible that this may be main the reason why a small bonus helps, at least a little. If that is the case, optimizing for tree size will be a good approximation for optimum strength..bob wrote:Not sure how minimizing search space infers anything at all about an optimal value for any scoring term...Jan Brouwer wrote:Rotor uses a bonus of 5 cp, this was roughly the optimum when minimizing tree search sizes for the LCT II set of positions.mjlef wrote:Has anyone done testing on the Side to Move bonus, and if so, how much does it help (if anything)? It would seem simple enough to test some public source programs with and without the bonus to see if it helps. Mayve Bob can use his amazing cluster to see the effect...it might be small and require a lot of games.
Miguel
-
- Posts: 12541
- Joined: Wed Mar 08, 2006 8:57 pm
- Location: Redmond, WA USA
Re: Side to Move Bonus---does it help?
Heuristically, we can also see that the calculation has created a value that is in line with that calculated by other approaches.michiguel wrote:There was a long thread started by Don Dailey some time ago. A given sidetomove bonus wll decrease the tree size because it smooth the jumps in eval from ply to ply. It is possible that this may be main the reason why a small bonus helps, at least a little. If that is the case, optimizing for tree size will be a good approximation for optimum strength..bob wrote:Not sure how minimizing search space infers anything at all about an optimal value for any scoring term...Jan Brouwer wrote:Rotor uses a bonus of 5 cp, this was roughly the optimum when minimizing tree search sizes for the LCT II set of positions.mjlef wrote:Has anyone done testing on the Side to Move bonus, and if so, how much does it help (if anything)? It would seem simple enough to test some public source programs with and without the bonus to see if it helps. Mayve Bob can use his amazing cluster to see the effect...it might be small and require a lot of games.
Miguel
-
- Posts: 201
- Joined: Thu Mar 22, 2007 7:12 pm
- Location: Netherlands
Re: Side to Move Bonus---does it help?
It is a bit indirect, smaller search trees imply stronger play.bob wrote:Not sure how minimizing search space infers anything at all about an optimal value for any scoring term...Jan Brouwer wrote:Rotor uses a bonus of 5 cp, this was roughly the optimum when minimizing tree search sizes for the LCT II set of positions.mjlef wrote:Has anyone done testing on the Side to Move bonus, and if so, how much does it help (if anything)? It would seem simple enough to test some public source programs with and without the bonus to see if it helps. Mayve Bob can use his amazing cluster to see the effect...it might be small and require a lot of games.
I see this bonus more as a smoothing factor, to make the score of a PV less dependent from whether the PV length is odd or even.