Author: Dan Andersson
Date: 16:46:22 08/23/01
Go up one level in this thread
>I read a paper by Plaat, Schaefer, et. al. which used the basic negamax >framework to compare AB, negascout, SSS*, DUAL* and MTD(f). The paper indicated >a dramatic (ok, 5-10%) improvement using MTD(f) when the move ordering was >"good" (i.e. within 10-15% of "best" order.) There was a flurry of papers on >this algorithm. Was it ultimately found wanting? No, but the most common search implementations are PVS-Negascout. If you use MTD(f), some of the usual tricks of the trade may be counter productive. And some other ideas that are barely break even in PVS, give large gains in combination with MTD(f). For example: Lazy eval is hard to do right, while ETC is a gain. IMO. YMMV.
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.