Author: Volker Böhm
Date: 15:41:53 05/27/04
Go up one level in this thread
On May 27, 2004 at 11:59:43, Ed Schröder wrote: >On May 26, 2004 at 13:34:23, Robert Hyatt wrote: > >>On May 26, 2004 at 12:30:33, Tord Romstad wrote: >> >>>On May 26, 2004 at 12:13:03, Robert Hyatt wrote: >>> >>>>On May 26, 2004 at 10:53:16, Stefano Gemma wrote: >>>> >>>>>I've used an recursive-iterative deepening. I try to explain in my poor english. >>>>>At any N plyes i've applyed iterative deepening as for the root. Suppose to >>>>>start with 2 plies. The next iterations you should search 4 then 6 then 8 plies >>>>>etc, form the root. But, when you're searching 8 plies depth, and you are at a >>>>>position located at ply 2, why don't use iterative deepening starting from ply >>>>>2, instead to do a full search of the remaining 6 plies? So i've tried to >>>>>consider positions at ply 2 (and 4 and 6...) as they were at the root, and start >>>>>a search to ply 4, then 6, then 8. Sometimes works better, sometimes worse... >>>>> >>>>>I have tried different schemas, in Raffaela. The best seems the schema 2-4. You >>>>>increment the iterative deepening by 2 plies (one chess move by colour) and, for >>>>>any ply, you make an iterative deepening with increment 4. In some position, 2-2 >>>>>was better. >>>>> >>>>>I've abandoned this idea, for now, because i'm working on a new engine and i've >>>>>little time for my hobby, but i think that could be interesting. >>>>> >>>>>Ciao!!! >>>>> >>>>>Stefano Gemma >>>> >>>> >>>>That sounds like an interesting idea that is worth testing. IE at _any_ ply >>>>where you want to do a depth=N search, you iterate and do a depth=1, 2, ..., N >>>>to work your way up to that point. With luck the depth 1 to n-1 searches will >>>>be cheap with hash information, if there is none, move ordering will probably be >>>>broken anyway and this might improve things. >>> >>>This is exactly what I always thought was called "internal iterative >>>deepening", but I have recently learned that others use this term to refer >>>to something different. >>> >>>I use something similar to Stefano's technique (if I understood it correctly) >>>at all internal nodes where the remaining depth is high (currently 5 plies or >>>more) and I expect a fail high. >>> >>>Tord > >>I have used "IID" for years, but in a very restricted way, namely to handle the >>case along the PV where I have no hash move. I've never tried it _everywhere_ >>before, so have no data. But I intend to try to see if it is something that >>could work, or if it is a waste... > >Try to use IID in another way. Actually I found 2 ways that perform better than >the original IID idea. > >Preamble: RD = remaining depth > R = reduction depth > >1) if (RD >= 3) search all moves with R=(RD-2)+(RD/2) then search all moves >again with the full depth. The principle applies only once in an unsorted tree, >thus is not recursive. > >2) if (RD >= 3) search a move with R=(RD-2)+(RD/2) and when finished research >that move again with the full depth. The principle applies only once in an >unsorted tree, thus is not recursive. > >I have tried RD>=2 too but found that normal move ordering is superior. > >Ed Sorry if my question is silly. For example with RD == 10 R gets (10-2) + (10 / 2) == 8 + 5 == 13. What to do with "13"? It cannot be the new remaining depth and it cannot be R - 13 as this is negative. Is there a bug in the formular or do I miss something? Thanks Volker
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.