Author: Dann Corbit
Date: 10:57:27 12/29/99
Go up one level in this thread
On December 29, 1999 at 09:10:50, Michael Cummings wrote: >On December 29, 1999 at 07:02:45, Graham Laight wrote: > >>On December 29, 1999 at 06:36:45, Michael Cummings wrote: >> >>>The SSDF rankings of the last two lists mean very little. Few programs have been >>>tested on the faster hardware. There are many points of view on which is the >>>best. But this will always be a point of view on paper, it will be very hard to >>>ever pick which is the best. No games will ever be enough in my view >> >>Excuse me, but following this kind of logic, you'll never be able to select >>anything in your life! >> >>Every competition is flawed. >> >>Gary Kasparov might not be the best human player - for all sorts of reasons. >> >>But if you finally want to have a champion, you have to have some sort of >>competition, and you have to accept that the rules are going to favour some >>competitors over others. >> >>-g > >It is very simple. you need to play many games, but the time you would spend >playing these games needed, another version of the program comes out and they >dump the older version for the newer one to test. So you will get results from >the data you have obtained. > >And yes competitions are flawed. France won the soccer world cup to say they are >the best, but they are not ranked number one. I can play 1000 CM6K games of >various personalities and then play them again and get totally different >results. (and I have !!!) > >So how many games does it take ? 10, 100, 500, 1000, 10000 ???? > >I have played top programs against each other and had one win many games in a >row, only to have the program win the other half of the games in a row. > >Usually the cream will rise to the top. But chess programs require much more >games than the SSDF play in my opinion. But again this is a selective argument. > >How many people are willing to say that Deep Blue is better than Kasparov 100% >of the time if they play a series of five game matches ???? > >As you say, there are rules and things are ranked following these rules. I am >not debating that, I am debating the validity of the results and the ranking of >the results. Why do you suspect the results are not valid? Do you have any mathematical basis for the claim, or is it simply that you don't like them. They certainly seem mathematically sound to me (at least for the stated terms of the experiments) >Logic shows you have competition to attain a winner, but logic also shows that >the winner is not always the best. > >How many times have someone or some team won something only to know that yes >they did great on the given day, but they are not really the best. Certainly that is true. You can never be 100% certain. But what better analysis tool for computer verses computer play do we have than the SSDF results? Further, they provide standard deviations in their table so you can know (at least) probability that one program may be stronger or weaker than another under the stated conditions of the tests.
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.