Author: Tord Romstad
Date: 06:48:37 03/26/04
Go up one level in this thread
On March 26, 2004 at 07:28:53, Vasik Rajlich wrote: >On March 26, 2004 at 03:02:04, Tord Romstad wrote: > >>On March 26, 2004 at 02:39:29, Johan de Koning wrote: >> >>>>I thought so, too, but after watching the first two games of a blitz match >>>>between the normal version of my engine and an identical version with the >>>>null move disabled I am already beginning to doubt it. The non-nullmove >>>>version is leading by 2-0. >>> >>>IIRC it was only 2 days ago Will Singleton posted a funny remark about >>>discarding null moves after 10 games. :-) >> >>Yes, I remember. :-) >> >>I wasn't really close to discarding null moves, of course, but I was surprised >>that the version without null move had any chance at all, and that the >>difference in search depth was not greater. I expect the null move version >>to win by a clear margin when enough games are played, but it will be >>interesting to see how big the difference is. After 8 games, it's 4-4. >> >>>>The difference in search depth isn't as big as I expected. In the middle >>>>game, the null move version typically reaches about 12 plies, while the >>>>version without null move reaches 11 plies. >>> >>>Hmmm... that sounds pretty incredible. >>>We're not talking about 11 ply full-width here, are we? >> >>No, I do lots of pruning besides null-move, and reach rather high search >>depths even without it. Here are the number of nodes I need to search >>n plies (n from 1 to 13) with and without null move from the opening >>position: >> >>Plies Nodes (null move off) Nodes (null move on) >> 1 55 55 >> 2 228 261 >> 3 965 439 >> 4 3,741 2,318 >> 5 13,213 6,638 >> 6 35,056 18,964 >> 7 37,398 36,095 >> 8 77,226 60,240 >> 9 117,274 109,939 >>10 239,355 237,775 >>11 555,661 412,969 >>12 1,942,523 1,084,089 >>13 12,060,312 3,531,279 >> >>I have no good explanation for the bizarrely irregular effective >>branching factor of my non-nullmove search here. I wonder how it is >>possible that hardly any nodes are needed in the 7th iteration. >> >>Tord > >Null move should be much more profitable for an engine with no other selectivity >than for an engine which is already very selective to begin with. Just about >every reduction you can think of is partially handled by null move. > >Junior doesn't even use null move. (According to Amir, this is clearly stated in >the archives.) > >Tord, even if Gothmog with null move eventually beats Gothmog without, It seems that it does. The score is 41-27 in favor of the null move version right now. >it would >be a much more fair experiment if you spent some time tuning the search without >null move, and had that play against the current Gothmog - which is tuned to >search with null move. You might be surprised ... You are right, of course. At the moment I have other tasks with higher priority in my engine, but eventually I hope to find some replacement for or generalization of recursive null move search which works even better. I still don't know how to achieve this, but I strongly believe that it is possible. Making good use of the null move observation (that Zugzwang is rare in chess) is probably essential in order to write a good search, and I also think that allowing one side to make several moves in succession can be a very powerful technique (in order to find threats, plans, trajectories and similar stuff; there are probably big improvements to be found in this area). However, it is hard to believe that conventional null move pruning is the best way to combine these ideas. It should be possible to come up with something better, probably with the help of some more domain-dependent knowledge. This is, of course, too vague to be of any use to anyone. If I ever make any interesting discoveries in this area, I will of course post it here. >Anyway, can we make some conclusions about the King search? :-) I don't own the King (nor even a computer capable of running it), and I am also not very good at figuring out how chess engines work by looking at their search output. On the other hand, you seem to have some skill in this area, and it would be fascinating to read your thoughts. Johan is of course also welcome to write about the King's search. :-) Tord
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.