Author: John Merlino
Date: 10:29:13 02/06/03
Go up one level in this thread
On February 06, 2003 at 07:52:34, Harpo wrote: >After a 15 year hiatus, I'm planning to return to tournament play. In >preparation, I've been playing 30/75 SD60 games against a handful of CM9K's >personalities. I recall reading somewhere that these were guaged with USCF-rated >players, but I would appreciate any opinions from OTB players as to how accurate >these ratings "feel." I'm only curious about the personalities rated from >2100-2500. > >Thanks in advance. The following is copied from the "Gameplay FAQ" on the Chessmaster website: ------------------ Q. How are the ratings for the computer personalities calculated? A. The rating for the Chessmaster personality is based on several factors: the Swedish Rating List (SSDF) for the engine in Chessmaster 6000 on a Pentium 90 processor, an estimation of the strength of the NEW version of The King, based on the SSDF's rating of the OLD version, and several thousand computer vs. computer and human vs. computer games. The SSDF rating was converted from the Swedish scale to the U.S. scale and an adjustment was made for later improvements to the chess engine. Then a series of computer vs. human and computer vs. computer tournaments were played (a total of over 15,000 games, about 3,000 of which were human vs. computer) at a time control of Fischer 5/10 (five minutes, plus 10 seconds for each move). The results of these games were fed into a ratings calculator. Each of the personalities has a base rating as measured on a Pentium II-450 (the minimum spec processor for Chessmaster 9000). When you run Chessmaster 9000 on your system, the program performs a CPU benchmark measurement and adjusts the ratings accordingly. That formula is [Base Rating] + 50 * ln([Your CPU Speed] / [Base CPU Speed]) / ln(2). A reasonable rule of thumb to use is that each doubling of CPU speed results in a rating increase of 50 points. Q. How accurate are the ratings for the personalities? A. The ratings are mathematically precise, but they are mostly based on computer vs. computer games. While the human vs. computer games added much more "real-world" accuracy to the ratings in Chessmaster 9000, many of the Chessmaster personalities have quirks in their playing styles that a human will be able to exploit but a computer opponent will not be able to "see". Discovering and then taking advantage of these weaknesses is part of the fun of Chessmaster, but you might find yourself able to defeat certain personalities that you don't think you should be able to based on their ratings. ------------------ So, as you can see, there are two reasons why the ratings might not be particularly accurate, at least not in your case: 1) The time control for the calibration was a brisk Fischer 5/10. 2) Only about 15% of the games used in the calculations were done with USCF-rated humans. The rest were done with comp vs. comp games so that all of the 120+ "human" personalities in the game could get a rating. The upshot is that the higher the ratings get, the more accurate they are, and vice versa. If you are mostly interested in 2100-2500 personalities, then you are likely to get reasonable games from most of them. However, several of them do have bizarre quirks that will be easy for a human to exploit once they are discovered. What you should do is look at the settings (or, at least, the "Playing Style" description) for any/all of these personalities and verify that they don't have anything way out of whack (e.g. bizarre piece values, very high/low contempt for draw settings, etc.). One example is Shakespeare, who has a very high (base) rating of 2507. However, he LOVES pawns, and values them at 1.5 (i.e. half the value of minor pieces). He prefers to keep a swarm of pawns for the endgame. However, this might make him likely to trade a minor piece for two pawns. Another example is Buck, base rated at 2238. He has the absolute minimum value (zero) for "Control of Center", meaning that he couldn't care less about it. This is probably very easy for a Master+ player to exploit. Other players, however, are rather accurate (most likely within 50 points of their displayed rating, depending on your hardware). Some examples of these would be: Josh Age 12 (2100), Vanessa (2111), Jade (a ruthless defender at 2178), and Lou (who is interesting because she does not use an opening book -- 2367). jm
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.