Author: Peter Kappler
Date: 00:18:49 01/21/00
Go up one level in this thread
On January 21, 2000 at 02:30:17, Dann Corbit wrote: >On January 21, 2000 at 02:24:15, Peter Kappler wrote: > >>On January 21, 2000 at 01:16:56, Will Singleton wrote: >> >>>On January 20, 2000 at 22:25:42, Peter Kappler wrote: >>> >>>>>I think that more speed is not the answer, because at 15+ plies, you're already >>>>faced with with seriously diminishing returns per extra ply. Any tactics that >>>>exist in the position (or that a human would ever have a prayer of finding) have >>>>long since been discovered, so the computer's positional evaluation starts to >>>>become the limiting factor. And computer chess still has a long, long way to go >>>>in that area. As you've stated yourself in other threads, the GMs are getting >>>>better at exploiting these weaknesses. >>>> >>>>--Peter >>> >>>I'd have to disagree here. I don't see the diminishing returns as depth >>>increases. There are certainly tactics present at any depth, and one can >>>imagine a positional advantage at 18 ply that couldn't be seen at 17. >>> >> >>Imagine a slow chess program that can only search 5 plies in the middlegame. >>Even simple combinations will be beyond its search horizon. Now, take the same >>code, and put it on a much faster machine that can search 8 plies. The 8 ply >>machine will destroy the 5 ply machine. It won't even be close. The 8 ply >>searcher would probably win every game. At shallow depths, tactics are >>everything. >> >>Now lets imagine the same scenario, but with 15 plies vs 18 plies. The 18 ply >>machine will be stronger, but I think the games would be very closely contested, >>with lots of draws, and a few wins for the 18 ply machine. The concept I'm >>trying to illustrate is something that others have described as a "tactical >>sufficiency threshhold". It's a point (depth) beyond which forced tactical >>sequences are extremely unlikely. >> >>In short, if you compare an n-ply searcher to an n+3 ply searcher, the >>difference in strength will certainly diminish as n grows. > >This statement is false, according to experimentation. Both Robert Hyatt and >Ernst Heinz have shown experimental evidence that there is *NO* drop-off in >value for additional plies. See the "Crafty Goes Deep" and "DarkThought Goes >Deep" for experimental proof of the opposite of what you just said. > I have read that article, too, and all I can say is that you and I interpreted it very differently. Do you really think that the difference between 5 and 8 plies is the same as the difference between 15 and 18 plies? Or 27 vs 30 plies? Or 57 vs 60 plies? Think about it... >>>So, in the arena of comp vs human, speed is really the only answer. Or, more to >>>the point, in the absence of improved algorithms, speed would still do the job. >> >>Speed always helps. Beyond a certain depth, I just think knowledge helps much >>more. >What is knowlege in this case except understanding of the future? knowledge == better static evaluation == less dependency on search > That is >*exactly* what additional plies of search give. I agree. It's a question of efficiency, I guess. I'd rather have 18 plies with a smart eval instead of 20 plies with a dumb one. Remember that we started this as a discussion of how to improve Deep Blue. > Given enough depth, a tactical >program would be thinking strategically. I agree with this, too. --Peter
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.