Author: Brian Kostick
Date: 17:51:38 04/07/01
Go up one level in this thread
On April 06, 2001 at 23:59:55, Tina Long wrote: >Hi guys, >I've done all this before with Rebel Century, I worked for weeks, never formed a > conclusion, so I went to Japan for some months to forget it all. >What I say below about CM8k can also be roughly translated to Rebel Century. >Century has the added benefits (complications) of ONite search syle & the like. > > >Following brief discussion of CM8k personalities in >http://www.icdchess.com/forums/1/message.shtml?161873 >I tried a little research. > >I'm interested in CM8k playing at 1 hour and 12 hours per move (from the old >Komputer Korner days) on a P3 1ghz computer. > >I was using my Personality "CM128" default CM8k with 128m hash.(selectivity = 6) > >I then tried the same with selectivity = 12 >I then tried the same with selectivity = 2 > >At 12: Depths reached were 1/11, 2/14 & the like. >At 2: Depths reached were 7/9 10/12 & the like. > >2: for the same positions, looked MUCH deeper at "brute force" and only SLIGHTLY >shallower at selective depth than 12:. > >While most moves chosen were the same, when they differed I considered the 2: >were more sound (but then I'm just a dumb human) and more likely to agree with >Hiarcs732 & Tigger14. > >What is the opinion of CCC people on the strength of the decisions made, in very >long time period games or analysis: > >Is it better to have low selectivity and have a MUCH larger safer brute force >search? >Is it better to have high selectivity to have a SLIGHTLY deeper selective depth >search? >What's the benefit of that 1 or 2 extra ply compared to the risk of not checking >the full width beyond 1 or 2 plys? > >I feel the results will differ depending on the particular positions, & I'd be >inclined to have high selectivity in the Early & Middle Game & lower selectivity >in the Endgame. > >BUT I can quite easily see why just the opposite might be best. > >Your Opinions will be appreciated, > >See Ya >Tina Long Tina, I see Marc has already given some info, which I read with interest. I did a very simple test, it has no statistical backing... but I post for you to browse. I picked a position, random so to speak. I decided to search for Chessmaster's choice for move 26. xxx in The Web vs. Chessmaster ( now running at: http://www.chessbrain.de/cd/ce/ce.html ) I am willing to dedicate limited time, lets say approx 20 min, 733MHz. I believe Chessmaster, with default setings, will converge on Qa4 in this time. I then ran some different Selective Search settings. N/A signifies I didn't wait to find out... Total time to ply depth does not mean I let all the ply move finish, but rather that at least Qa4 at this depth was reached. SS Setting, Time to Qa4, Ply Reading, Total Time to ply depth SS6 3:17 6/12 19:52 SS12 2:55 1/13 19:52 SS9 2:56 4/13 25:35 SS7 3:10 N/A N/A SS8 2:52 N/A N/A In this case I can see that 3 minutes is required to converge on Qa3. If 40/120 time control I might not want to exceed 3 min. and I want a low Selective Search for a wide base. I quess I'd pick SS8. I can see your point on processor power, time avaliable, ect... Now I'm no computer chess coder and I think I learned something. Selective Search setting sets the spread between base and deep lines. i.e. in the numbers given above: SS6 = 6/12 (diff. 6); SS12 = 1/13 (diff. 12);, SS9 = 4/13 (diff. 9), ect... Maybe that was completely obvious to some... $0.02 worth, and it might not be worth that. Regards, BK
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.