Author: Daniel Shawul
Date: 02:27:38 09/28/05
Go up one level in this thread
On September 28, 2005 at 03:28:14, Tord Romstad wrote: >On September 28, 2005 at 00:55:01, Daniel Shawul wrote: > >>On September 27, 2005 at 18:20:10, Tord Romstad wrote: >>> >>>Yes. When I get some time for chess programming again (not any time soon, >>>I'm afraid) I want to give parallell search a try. I think it will be easier to >>>get >>>a parallell search working if I start with a very simple and minimalistic >>>search function. MTD is much easier to code than the other popular alpha >>>beta variants, and can be implemented in very few lines of code. It also >>>seems to be somewhat easier to parallellise, because all searches are done >>>with a null window. This means that it is never necessary to adjust the >>>bounds for other processors when one processor finds an improvement >>>to alpha or beta. >> >> If that is the only reason you are switching back, i think you might >>be disappointed. Especially if you use YBW you already get good enough bounds >>with the first move. > >It is not really the only reason. The general idea is that a simple search >should be easier to parallellise than a more complex search, and MTD is >much simpler than PVS. I am not sure if simpler code implies easier to parallelize. For example recursive search is easier to code than non-recursive search, but non-recursive is preferable for parallel search. I didn't implement MTD myself but i guess it is simpler to code because it needs only one bound and researches are avoided with in the search tree. Are there any other reasons? Otherwise PVS simes to fit well the YBW idea. First move searched with full window, the rest with zero window. Researches in both case indicate search overhead. I would love to check out your MTD search when glaurung is release. best Daniel > >And yes, YBW is what I will end up using, I think. > >Tord
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.