Author: Dave Gomboc
Date: 16:27:05 04/06/99
Go up one level in this thread
On April 05, 1999 at 18:00:28, pete wrote: >yes; i thought that but simply didn't write it as I don't want to advertise >anything :) > >in fact for fritz the same thing i stated about genius is true on another >hardware and time control level ... > >fritz is optimized for ruling SSDF and winning NUNN matches same time control I >think and if you use it close to SSDF settings it probably is the strongest >prog, meens very rarely fritz will play MUCH better if it has 2 hours to find a >move than 10 minutes ( a thing not true for junior for example :) ) > >pete If you are trying to make your software better, and you go and ask a grandmaster to come up with a balanced set of positions that display a wide variety of styles of positions, and you find that you are not playing some of the positions well, you will put some effort into understanding why that is the case, and correcting your software as much as you are able to handle these types of positions. Saying that Fritz is "optimized for ... winning NUNN matches" is nearing the point where one insults the software developer for trying to improve their software. A few years ago, computers couldn't play a KID Classical worth a damn. No program could. It was the easiest win on the block, you just launched an attack on the kingside while White shuffled pieces around on the queenside. The computer programs would think, gee, this guy has a weak king, I have a big advantage. They never saw the danger until it was too late. Today, ask Fritz to assess a typical KID Classical position, and it will probably report a score somewhere near even. Having a good test suite of positions is an important part of writing strong chess-playing software. Sometimes, the tests are static: here are 400 positions, search for an amount of time, is the new version of the software assessing the position correctly? In the "Nunn test" case, the test is over a whole game, and this is important too, because that is what we are after, the strong playing of full games, not just static positions. Indeed, this is the only way one can test improvements to time management code. When a developer has an idea of how to improve their search or evaluation, the first thing they may do is run the new version on a test set of positions that they have. Based on this quick assessment, they will decided if it is worth the extra compute time to play a series of games with this new version against various software programs (e.g. older versions, or competing products). Improving Fritz's performance from the Nunn positions has improved Fritz's overall standard of play. There is nothing underhanded about this, period. Dave Gomboc
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.