Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: recursive parallel search

Author: Daniel Shawul

Date: 21:00:31 12/26/04

Go up one level in this thread


On December 26, 2004 at 22:38:48, Anthony Cozzie wrote:

>On December 26, 2004 at 00:49:37, Daniel Shawul wrote:
>
>>Hi
>> i read in the DTS paper that the main advantage of non recursive
>>search is that splitting is possible anywhere in the tree.
>>My quesion isn't it possible to split at shallower depth on the current line
>>of search (from 0 to current ply P), by just undoing the moves made.
>>If after searching few moves,IE when i get a reliable bound, i immediately undo
>>the move that led to the current position with one of my threads, and then start
>>parallel search at the previous ply. Or i can undo two moves back  and start
>>par. search at P-2.
>>
>>I think with non-recursive search the only thing possible is to split at points
>>on the current line of search.
>>
>>This is like a modified PVS which releases idle processors after a good
>>bound is established to search at previous plies.
>>what do you think?
>>
>>daniel
>
>IMO, if you are going to do the work of putting everything in stack frames which
>you manage yourself, you have done 90% of the work of getting an iterative
>search.  If you get that far, I'll send you my iterative macros and you can
>convert to an iterative search in another 2 hours of work or so.  I should warn
>you, however, that by putting everything into shared stack frames, things become
>_even more_ difficult to debug :)
>
>anthony

  I think i am going to try my luck with the recursive one first.
Iterative search looks very prone to bugs,and hard to try search ideas excepting
the DTS issue. Besides I don't even know
how to make a non-recursive AB search. Pseudocode?
daniel




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.