Author: Mike S.
Date: 04:29:14 06/09/03
Go up one level in this thread
On June 09, 2003 at 06:34:16, Sune Fischer wrote: >(...) >I think the engine should always be tested (when officially trying to measure >the programs strength) using the default values, or whatever values suggested by >the author (should be the one and same probably). >(...) In general, this is true if you think of Fritz, Shredder etc. where you can change several settings too, but nobody seems to have found settings which are significantly better (or they didn't become widely known), so the defaults are considered the best. - Almost everybody uses the defaults with them (I guess), so it's obvious that a rating list is most informative when these engines run with defaults there, too. But The King is an exception, as everybody knows in the computer chess community. For example, it has been explained that the default selectivity was set lower, to provide best performance for CM customers who do not (yet) have such fast computers like the average computer chess enthusiast has. It was believed that high selectivity may harm the performance on slower computers. My interpretation of the king safety values (which are usually increased to with strong custom personalities) is, that these values don't have to be higher against typical human opposition, but are required to be higher against top computer competition. Keep in mind, Chessmaster is a program designed mainly to be used by humans which learn, train and play against the program, more than other programs (they my serve these purposes too and are equipped for it more or less, but are more directed at engine competitions at the same time IMO). >If the author gives the user the option of changing parameters, then that is >fine and good fun, but if the users are able to improve on the default values, >then that means the programmer has not done his job well enough and as such the >engine doesn't deserve to play with these improved settings in getting an >official score. IMO it's not about what an *engine* deserves, but what the reader of such a list deserves, to be informed about. It wouldn't make much sense IMO to display the performance of an engine, that virtually nobody uses set up that way. Who in the computer chess community runs King's default personality? That has Sel. 9 now (CM8K was 6 only), but also 4 MB (!) hash only IIRC, for example... I'd even recommend, in addition to selectivity 12, to adjust the king safety to 150/150, as a good compromise between settings improvement and "easy documentation" of the setting. (I think the effect of the usual small changes of the other values will be very small compared to these, although probably visible in terms of a few Elos, when very many games are played.) I appreciate when a list has playing conditions most similar to the average practise. Then, it can provide the best information value. >(...) >Well that means if one engine is improved by the users, then all the other >engines should have the same chance, or the conditions are simply not equal. I'm sure, if alternative settings of other engine become widely known and generally accepted as being better (before the test is run), they will be used. It seems to me this has been done already with Tiger 15, which was tested with MoreGambit=-1. AFAIK this isn't the default in the CB. GUI, but considered to be better by the Tiger 15 experts (?). Regards, M.Scheidl
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.