Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: MTD(f)

Author: Stuart Cracraft

Date: 18:59:10 07/27/04

Go up one level in this thread


On July 27, 2004 at 15:04:19, Chris Welty wrote:

>On July 26, 2004 at 23:28:04, Stuart Cracraft wrote:
>
>>But in this position (Reinfeld Win-at-Chess, position #1) I get millions of
>>iterations, no convergence, and a never-ending mtdf() loop.
>>
>>[D] 2rr3k/pp3pp1/1nnqbN1p/3pN3/2pP4/2P3Q1/PPB4P/R4RK1 w - Qg6 0 32
>>
>>This is the output deep into the run of the code below.
>>
>>:
>>:
>>mtdf iteration=9026 lowerbound=7038 upperbound= 9999999 g = 7038 beta = 7038
>>mtdf iteration=9027 lowerbound=7039 upperbound= 9999999 g = 7039 beta = 7039
>
>You're only moving a millipawn per iteration. As Tony Werten pointed out,
>fail-soft a/b and widening the search window will help (I'd recommend both).
>
>Once that's working your next problem will be search instabilities, where a
>window at +1000 returns +950 and a window at +950 returns +1000. One way to
>solve this is by only widening the window rather than doing null-window searches
>at each value.
>
>Good luck,
>
>Chris

Thanks -- seems like a whole new world.

Curious what % of the best programs use pvs, negascout, or mtdf...



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.