Author: Dann Corbit
Date: 18:47:10 09/14/00
Go up one level in this thread
On September 14, 2000 at 21:08:56, Amir Ban wrote: >On September 14, 2000 at 16:43:24, Dann Corbit wrote: >>On September 14, 2000 at 16:31:12, Uri Blass wrote: >>>On September 14, 2000 at 15:55:55, Dann Corbit wrote: >>>>Here are a set of tough positions to search deeply. Just finding a mate is not >>>>good enough, uless you can *prove* it is the shortest mate. >>> >>>Finding a mate is good enough even if you cannot prove that it is the shortest >>>mate. >> >>It's good enough to win. It's not good enough to find the most beautiful >>solution. In this case, it is a matter of goals. You can simply ignore any >>where you find checkmates if you don't like that part of the challenge. >> >>>Doing mistakes of not finding the shortest mate is going to change nothing in >>>rating points so I do not see the importance of it for normal chess programs >>>that are not mate solvers. >> >>Only a few of these will be a sure mate in 16 plies. Ignore those, if you so >>choose. >> >>>I do not see the point of searching to 16 plies. >> >>Stop at two plies then. Deeper is better. >> >>>It is easy to search faster if you do more pruning. >> >>But more error prone. If you search 20 plies by pruning but lose in 5 moves >>because of something you pruned out, it's not such a great idea. But if you can >>search 20 plies by extensive pruning and always get the same answers as a brute >>force search, then you have found something spectacular. Alpha-Beta (in >>particular) gets the same answer as exhaustive search, and only requres sqrt(n) >>tests provided you order the moves correctly. > >The point was that the nominal ply depth has little or zero meaning, since the >whole pruning and extension context should be weighed in. Alternatively, it is >possible to change the pruning and extension strategy to reach any ply depth you >wish to name. Of course, and it will not be useful if it introduces a statistically significant amount of error. Programs that add the wrong extensions will perform poorly. >For me, it is enough to note that the challenge is irrelevant for Junior, as it >does not even count depth in plies. Semantics. It expands the depth of the search on each new iteration. Call it what you want, it's depth in plies. I do know that you have special extensions for recaptures, etc. but that could be considered a type of pruning. >Ply depth is not more than a convenient fiction. Ply depth can be rigidly and correctly defined. Most programmers do not exhaustively search plies even with Alpha-Beta because of null-move pruning and similar extensions. Yet there is some sort of progress measure which is roughly comparable for all programs. Depth in ply (or whatever you want to call it) is a much better measure than node count. I suppose the best possible measure is "finding the right answer" but that is a lot harder to estimate than depth in plies.
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.