Author: James Swafford
Date: 05:40:39 01/25/06
Go up one level in this thread
On January 25, 2006 at 04:19:17, Ross Boyd wrote: >On January 25, 2006 at 01:34:54, Joseph Ciarrochi wrote: > >>We learn early on that if we have a winning advantage, we want to eliminate any >>chance of counterplay by our openent. We usually would like to simplify the >>position. >> >>I've noticed that rybka and other engines often won't simplify the position , if >>there is the slightest loss of points (say -.1). So lets say black and white >>still have q's on the board, but black has a clearly winning advantage (+6). if >>black gets the chance, it should trade those queens (assume this really is a >>good idea). >> >> >>Are there any engines that will do this?, i.e., take a slight decrease in >>advantage to simplify the position and garuntee the win? > >Hiarcs springs to mind.... it will occasionally throw away material to simplify >into an easy win. You will see this kind of behaviour in most engines that use >egtbs... ie. they will throw wood to get a tablebase win/mate. > I've observed this behaviour (even in my own stuff), and it's not because of any programming-for-simplification. It's just a side effect of the search seeing a win from TB's with N men on the board, and having to throw one away to reach that position since it doesn't have N+1 men TB's. Not quite the same as gracefully simplifiying down... -- James >In my engine I don't do anything to simplify (apart from the above)... although, >I have an idea in the pipeline which I call "waypoints" which is a type of late >middle-game recognizer code that will actively force the search towards easily >won endgames. It will be used mainly for hyper-aggressive endgame pruning. I >expect its play to appear stark raving mad to the trained eye - even though it >should still retain the win. > >On the flip side, when its losing/defending it will doggedly try to keep the >game complicated, uncertain and in uncharted waters. > >Well, that's the idea in theory... > >Ross
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.