home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky gnu.chess:504 rec.games.chess:12114
- Newsgroups: gnu.chess,rec.games.chess
- Path: sparky!uunet!stanford.edu!nntp.Stanford.EDU!928s4
- From: 928s4@leland.Stanford.EDU (Benjamin Sloss)
- Subject: Re: GNU Chess ratings estimates...
- Message-ID: <1992Dec29.170710.9011@leland.Stanford.EDU>
- Summary: depends on the hardware
- Sender: ?@leland.Stanford.EDU
- Organization: DSG, Stanford University, CA 94305, USA
- Date: Tue, 29 Dec 92 17:07:10 GMT
- Lines: 38
-
- Tony-Preston@cup.portal.com (ANTHONY FRANCIS PRESTON) writes:
- >I have the Amiga version of the GNU chess program and think it is pretty
- >good. My questions are 1) what are the estimates for rating strength(USCF?)
- >versus the various time controls? 2) how deep will the search for moves
- >go versus the time controls. I am only looking for intellegent estimates!
-
- To estimate the strength of the program, it is necessary to know how fast
- the program runs on your hardware. Perform the following test:
- * With the book turned off, set the time controls to some long interval
- (say move/10 min) and set the search depth to 6 ('depth' <cr> '6' <cr>)
- Turn on 'post' and then type 'go'. Record the time it takes your machine
- to calculate the 6-ply search for White's first move (it should get Nf3).
-
- Now, we need to make two assumptions. The first is fairly well established:
- every doubling of processor speed improves the program's play by roughly 75
- points. The second is a baseline rating for GNUChess, which I will
- estimate as 2300 on an RS/6000.
-
- The RS calculates the search in 27 sec. Thus, you can estimate the rating
- of GNUChess on your Amiga by the formula 2300 - 75 * (log2(t) - log2(27))
- where t is the number of seconds it took your machine to calculate the
- search.
-
- In answer to your second question (depth of search vs time), the answer
- depends greatly on the type of position being searched. Endgames, with their
- reduced piece count (and thus reduced movelist) are searched much deeper than
- complicated middlegames, in a given amount of time. Perhaps a more useful
- question is "my program searched n ply in time t; how long will it take
- to search n+1 ply?". The answer here is about 6t for middlegames, and
- perhaps 3t for endgames. This is again due to the differences in movelist
- size (which averages 40 in the middlegame, much less in the endgame) and
- the effeciency of the alpha-beta pruning algorithm.
-
- --
- Ben Sloss Email: ben@osc.versant.com Work: (415) 329-7500 x172
- Quote: "It's all fun and games until someone gets paid."
- Transportation: '83 CB650SC (straight roads) '93 CBR 900RR (twisty roads)
- '91 XR 250L (dirty roads) '87 Mustang 5.0 (wet roads)
-