TalkChess.com :: View topic - LMR in micro-Max

TalkChess.com
Hosted by Your Move Chess & Games

FAQ

Memberlist

Usergroups

Profile

LMR in micro-Max

TalkChess.com Forum Index -> Computer Chess Club: Programming and Technical Discussions

View previous topic :: View next topic

Author

Message

Uri Blass

Joined: 08 Mar 2006
Posts: 5956
Location: Tel-Aviv Israel

Post subject: Re: LMR in micro-Max Posted: Sun Apr 08, 2007 5:10 pm

hgm wrote:

I tried exempting Pawn moves from reduction (next to captures), and it did not make much difference scorewise (still 56% over 120 games). Only the 10% speed advantage disappeared. (It is much easier to do this in uMax, though, as Pawn double moves cannot be tried first even if they are hash moves. They are then tried in second place, which might lead to reduction of the hash move if you do not ad a lot of code. If Pawn moves are never reduced, this problem disappears automatically.)

To my surprise, at 10 times longer time-control, (5 min per game) the result stays about the same: 56.2% (47+ 41= 32-), Pawn moves + captures exempted, Pawn pushing version. I would have expected the advantage to grow with search depth, as LMR is a branching-ratio changing pruning. Apparently there is no extra advantage in thinking further ahead around the PV than in the exhaustive full-width search part beyond a certain number of moves. So I am increasing the minimum remaining depth at which LMR is applied from d=4 to d=5, which decreases the difference n depth between deepest and worst branches.

As far as I know there is no proof that LMR helps more at longer time control.

I know that zappa1.1 and the baron that do not use LMR are 2 of the program that perform better at long time control.

Another point is that you did not play enough games.

I played 100 games match between 2 version of movei with fixed number of nodes and in one match(100000 nodes per move) the result was 60-40 for version A and in another match(200000 nodes per move with the same positions) the result was a win for version B 54-46.

version B also won a match with 400000 nodes per move 53.5-46.5 but lost by a bigger margin with 1,000,000 nodes per move(I do not remember the result and the match was in a different computer but it was better than 55-45 for version A).

Uri

Display posts from previous:

Subject	Author	Date/Time
LMR in micro-Max	H.G.Muller	Sat Apr 07, 2007 6:44 pm
Re: LMR in micro-Max	Zlaire	Sat Apr 07, 2007 7:31 pm
Re: LMR in micro-Max	H.G.Muller	Sat Apr 07, 2007 7:54 pm
Re: LMR in micro-Max	H.G.Muller	Sun Apr 08, 2007 4:37 pm
Re: LMR in micro-Max	Uri Blass	Sun Apr 08, 2007 5:10 pm
Re: LMR in micro-Max	H.G.Muller	Sun Apr 08, 2007 5:44 pm
Re: LMR in micro-Max	H.G.Muller	Mon Apr 09, 2007 7:47 am
Re: LMR in micro-Max	Robert Hyatt	Mon Apr 09, 2007 8:38 pm
Re: LMR in micro-Max	Dann Corbit	Mon Apr 09, 2007 7:28 pm
Re: LMR in micro-Max	H.G.Muller	Mon Apr 09, 2007 8:17 pm
Re: LMR in micro-Max	Peter Fendrich	Tue Apr 10, 2007 6:05 pm
Re: LMR in micro-Max	William H. Rogers	Tue Apr 10, 2007 8:27 pm
Re: LMR in micro-Max	Peter Fendrich	Tue Apr 10, 2007 8:43 pm
Re: LMR in micro-Max	William H. Rogers	Tue Apr 10, 2007 9:26 pm
Re: LMR in micro-Max	H.G.Muller	Thu Apr 12, 2007 10:44 am
Re: LMR in micro-Max	Uri Blass	Thu Apr 12, 2007 11:26 am
Re: LMR in micro-Max	H.G.Muller	Thu Apr 12, 2007 11:37 am
Re: LMR in micro-Max	Michael Sherwin	Tue Apr 10, 2007 3:38 pm

TalkChess.com Forum Index -> Computer Chess Club: Programming and Technical Discussions

You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum