Author: Pekka Karjalainen
Date: 12:51:28 02/13/02
Go up one level in this thread
On February 13, 2002 at 14:12:07, Uri Blass wrote: >On February 13, 2002 at 11:41:20, Edward Seid wrote: > >>Just to tempt the would-be Go programmer, here's a short list of online >>resources. >> >>Computer Go FAQ >>http://www.usgo.org/computer/faq.html > > >I understand nothing > >They say that programs have improved by at most one stone per year. >Can someone translate it to elo. http://igoweb.org/~pahle///go-faq/ says this, begin quote: 1.9 Does Go have an equivalent to ELO rating? Beginners often compare Chess and Go and ask whether the rating systems in both games are related. The Elo system in Chess attributes a rating figure somewhere between 800 (beginners) and 2700 (top grandmasters) to every player. This rating changes depending on the performance of the player in tournaments. The basic idea of Elo is that a player should have a winning expectancy of 69% (or, more precisely, an expected score of .69) against another player whose Elo rating is 100 points less. In the recent history of Go, there have been various attempts of translating this system to our game. The most popular model is the European Rating List, maintained by Ales CIEPLY at http://www.european-go.org/rating/gor.html The basic observation in Go is that the winning expectancy against a player one grade weaker roughly equals 69%, as well. Therefore, it seems natural to translate Go grades to Elo ratings by using steps of 100 for every grade. Ales does so by defining 1 dan = 2100 points. Consequently, 1 kyu = 2000 points, 2 dan = 2200 points, etc. A player with 2050 points is called a weak shodan, whereas a player with 2150 goes through as a strong shodan. However, there are two major problems with this system. Firstly, the winning expectancy of 69% per grade is no constant in Go. It seems to be lower in the weak kyu range and higher in the strong dan range. (The reason for this seeming paradox is that strong dans play more consistently and less erratically than weak kyus.) Therefore, Elo's formulae have to be adjusted in complicated ways to fit the game of Go with the scale mentioned above. Secondly, the traditional grade system 'kyu - dan' is much more popular among Go players than any attempt of an Elo system. Most amateur Go players simply ignore their Elo ratings and rate themselves according to the grade system. This low acceptance of Elo ratings in Go gives them only small significance at the moment. This might change sometime in the future, but probably not all too soon. There is a rating system similar to Elo among Chinese professionals. But as with European amateurs, it does not seem to work very well. :end quote > >There should be no problem to calculate elo in the same way as chess based on >games. > >How much do programs improve per year and what part of that improvement is >thanks for hardware? Very little. > >What is the rating difference in elo and not in stones between the best go >players in the world and the best program? Best player Lee Changho is above 3000 EGF. Best program is hardly 1000. > >Uri Pekka Karjalainen
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.