RE: [computer-go] Pattern matching - example play

[Vincent Diepeveen <diep@xxxxxxxxx>]

At 13:45 4-12-2004 -0800, Ray Tayek wrote:
>At 11:20 AM 12/4/04, you wrote:
>>... the key to a strong go program is a strong evaluation, not better 
>>search.  The
>>search tricks are irrelevant without good evaluation.  ...
>
>there is probably a lot of agreement about this among go players.
>
>>... Likewise, to play good go, you need o understand things like "aji", ...

>yes, and things like timing and asking moves etc.
>
>i suspect the chess people could do very useful work here. i wonder if 
>these concepts (like agi etc.) could be abstracted or found in some other
game?

In chess aji translates to hung. In general very strong players manage to
delay to collect hung pieces/pawns later than weak players.

>another thing that may be confusing (at least to me , not being a chess 
>player) is the rating systems. iirc, 2200 is a master of some sort in chess.
>
>but in go people start at say 25 kyu and if we suppose that a 9 dan pro is 
>sorta like  a 10 dan amateur, we get a 35 stone handicap difference. if the 
>average go player is 10 kyu worldwide  (i'm guessing), what is the average 
>chess player's rating? (i.am guessing 1200 because of playing on yahoo).

the chess rating system is a more fine grained way to distinguish persons.
it's not comparable to the kyu/dan system.

the average chess rating is dependant upon where you start introducing
players.

Basically you can set all players to 600 rated or 1600 rated, important is
only the relative difference.

Another factor is that there is different rating systems. The American
rating system has more deviation. Inflation is a bad word. It's more rudely
calculated because of different K factors and quick games get counted too.

In europe only slow games get counted and usually only bad players do not
have a rating.

That gradually changes now, meaning that the average world wide rating goes
down bigtime now.

Youth players, who will wipe half USA, who had a few years lesson get
introduced around 1100 - 1200 here.

Youth players in USA i see introduced at server around 600-800.

Most importantly is to take into account what those numbers mean.

I'm currently rated around 2300. That means if i play in a tournament where
i cannot pick out my opponents but the tournament organisation is doing
that and he has also a 2300 rating, i have a 50% chance to win. 

Now let's assume non-international players using k=30.

Rating 1800. Plays opponent from 1800. That means 50% chance.
With k=30 that means in case of a win the rating change over that game
would be for the winner:

score : 1 point minus expected number of points 0.5 (50% from 1) = +0.5 score
30 * 0.5 = +15 rating points based upon 1 game.

Above 2000 rating, the K factor nowadays is 15.
Above 2400 rating (international master level so usually professional
players) the K factor is 10.

There are problems with the rating system.

In theory a player rated around 750 points higher than his opponent gives
the opponent rounded off 0% chance to win from him.

With 300 rating points difference it's already roughly 2-5% or so.

For weak amateur players, this is very true. It happens very seldom but it
happens.

However in my entire life i never lost from someone rated more than 350
points lower. That 350 points loss was 7 years ago at a very loaded weekend
tournament where i (skipping all kind of excuses) played a youth player who
soon after was rated 100 points lower.

Losses against players 100 points lower or higher happen regurarly.

100 points difference means roughly a 70% chance for the higher rated to win.

And a major problem is the differences in K factors for the
(semi-)professional players versus the amateurs.

Note that the lowest international rating possible nowadays is 1800.

That used to be 2000 a while ago and 2200 a year or 10 ago.

So the differences between the professional players get bigger too.

A year or 10 ago grandmasters still in the strength of their life (20-40
years old) were not seen regurarly in the 2400 rating regions.

You see many there now.

So you clearly see that to get to the top only gets harder and harder
because the GM level already starts now what in past was international
master level.

'peeking' is what happens in chess nowadays too.

Basically in go there is perhaps 30 gradations. In chess there is 2300+.

>From 500 to 2800+

Yet it is not comparable at all. A go board is 19x19 and setting up a
single position which i had in a game takes me half a day. A chess board is
8x8 and setting up a position takes 20 seconds, including the beginning
position.

If we would play chess on a 19x19 board (giving 18x18 squares) then
probably it would be popular in just a few countries too.

>so spotting someone a pawn in chess may be like giving some a few handicap 
>stones?
>
>my understanding is that many of the go programs are said to be 3-4 kyu, 
>but i (a 1-dan) can give manyfaces and smart go 7 stones and beat them 
>easily using normal (handicap) play, so they seem to me to be about 8kyu or 
>so (but i have no experience with the latest programs from the competitions)
>
>maybe some ideas like this can put things in perspective?
>
>thanks
>
>---
>ray tayek <http://tayek.com/>, co-chair <http://www.ocjug.org/>, actively 
>seeking telecommuting work.
>
>_______________________________________________
>computer-go mailing list
>computer-go@xxxxxxxxxxxxxxxxx
>http://www.computer-go.org/mailman/listinfo/computer-go/
>
>
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/