Author: Bruce Moreland
Date: 15:44:30 07/26/01
Go up one level in this thread
On July 26, 2001 at 17:37:02, Kevin Stafford wrote: >The argument isn't really about node-rate in this case, but I can see the >confusion as that is the topic of the thread. I was coming back to an earlier >argument made by Otello that positional knowledge is effectively useless in >chess, because chess is all tactics. My point was that even the fastest engines >today (measured however you wish, depth, NPS, etc) have a good amount of >positional knowledge, and that positional considerations are hardly made >obsolete by gaining the ply or two over the competition you'll get by using a >quick (but less-intelligent) eval function. I hope that makes some sense. Positional knowledge complements search by increasing accuracy in cases where there is not enough depth to resolve a tactical issue, and by identifying cases where search is apt to be of no use. A less intelligent eval may be better, and may not. The dumbest eval is "return 0". That will miss everything that is not caught by search (mates). No amount of search will create a program that plays *well* using such an eval function, even though the program will search deeply because everything cuts off. If you add a material term, you've created a much smarter eval function, by the program will make terrible positional mistakes. If you add static piece-square tables to this you have something that will beat an IM at 5 0 blitz a lot of the time. I think that you need to compromise when adding terms that are more esoteric and more expensive. Adding a lot of them and squeezing the search out sounds like a bad plan. I think that all you'd do is see that you are losing after a 7 ply search rather than the 10 that your opponent needs. bruce
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.