Handicap and Rating difference
Mitchell E. Timin
in rating means N handicap stones should yield a fair game. This
holds for both dan and kyu scales, and also 1 dan vs 1 kyu should
be equivalent to N = 1.
But what about in the professional World? I have heard, or read,
that professional ratings are more finely divided, so that it
takes a rating difference of 3 to be worth one handicap stone. Hence
a 9 dan professional gives no more than three stones to a shodan,
and two stones to a 3 dan.
Is this true? Do any of our computer rating systems work this way?
Mitchell Timin
Stewart Strait
: As is well known, within the amateur GO world, a difference of N
: in rating means N handicap stones should yield a fair game. This
: holds for both dan and kyu scales, and also 1 dan vs 1 kyu should
: be equivalent to N = 1.
This is not "well known" but rather is an important point of controversy.
Some rating systems such as the American GO Association accept your
point of view. Others, such as the Netherlands GO Association (at
least as things were about two years ago), take the view that N
handicap stones should give a difference of N-0.5 in rating, because
White moves first in handicap games and the difference between
drawing lots for the first move and giving it to White is regarded
by "N-0.5" type rating systems as like half a handicap stone.
I run an isolated local rating system for the San Diego GO Club,
using an "N-0.5" type algorithm originated by one of the other players
about 7 or 8 years ago. Since the algorithm has needed some
modifications, I try to follow this controversy.
DR. ROY SCHMIDT
: Mitchell E. Timin (ti...@netcom.com) wrote:
: : As is well known, within the amateur GO world, a difference of N
: : in rating means N handicap stones should yield a fair game. This
: : holds for both dan and kyu scales, and also 1 dan vs 1 kyu should
: : be equivalent to N = 1.
: This is not "well known" but rather is an important point of controversy.
: Some rating systems such as the American GO Association accept your
: point of view. Others, such as the Netherlands GO Association (at
: least as things were about two years ago), take the view that N
: handicap stones should give a difference of N-0.5 in rating, because
: White moves first in handicap games and the difference between
: drawing lots for the first move and giving it to White is regarded
: by "N-0.5" type rating systems as like half a handicap stone.
??? Black moves first in all handicap games. Under the Japanese rules and
traditional Chinese rules, for handicaps of two or more moves, the
placement of the moves is specified. But just because Black's first moves
are restricted doesn't mean White moves first!
Anyway, the definition of the rank system is such that the ranks have
precisely the meaning stated by Mitchell. If the ranks are out of balance,
rather than change the handicap system, change the ranks to agree with
reality.
Roy
--
Roy Schmidt sch...@uxmail.ust.hk
Information & Systems Management Dept, School of Business and Management
The University of Science and Technology
Clearwater Bay, Sai Kung, HONG KONG
Stewart Strait
: Stewart Strait (str...@crash.cts.com) wrote:
: : Mitchell E. Timin (ti...@netcom.com) wrote:
: : : As is well known, within the amateur GO world, a difference of N
: : : in rating means N handicap stones should yield a fair game. This
: : : holds for both dan and kyu scales, and also 1 dan vs 1 kyu should
: : : be equivalent to N = 1.
: : This is not "well known" but rather is an important point of controversy.
: : Some rating systems such as the American GO Association accept your
: : point of view.
I was somewhat inaccurate here. For N>1 the AGA system says that a
difference of N ranks mean that N handicap stones are fair. For N=1
the AGA says that the fair handicap is 5.5 points reverse komi
(White gives Black komi, the reverse of the procedure for equal
players).
: : Others, such as the Netherlands GO Association (at
: : least as things were about two years ago), take the view that N
: : handicap stones should give a difference of N-0.5 in rating, because
: : White moves first in handicap games and the difference between
: : drawing lots for the first move and giving it to White is regarded
: : by "N-0.5" type rating systems as like half a handicap stone.
Depending on your choice of definitions, you may prefer to say
"White has the first unrestricted move in handicap games" and
discuss "drawing lots for the first unrestricted move" after the N
handicap stones are placed.
: Anyway, the definition of the rank system is such that the ranks have
: precisely the meaning stated by Mitchell. If the ranks are out of balance,
: rather than change the handicap system, change the ranks to agree with
: reality.
Changing the ranks to agree with reality is only possible if the
handicap system also agrees with reality in certain ways. If the
rank difference assigned to one handicap is not correctly related to
the rank difference assigned to the other handicaps, there will be
an inconsistancy that cannot be removed by changing the ranks.
While the most visible difference between some rating systems is
that some have wider strength differences between adjacent ranks
than others, this difference of scale is not, as far as I know, the
subject of the controversy.
If a player A gets a fair game giving K stones to B, and B gets a
fair game giving L stones to C, then the controversy is about how
many stones A should give C to get a fair game. Any of the systems
of handicaps and ratings or ranks we are discussing imply an answer
to this question. The traditional system says A should give K+L
stones to C. The "N-0.5" systems say K+L-0.5 stones.
An extreme example: consider a handicap system where a difference in
rank of 2 means 2 stone handicaps, but a rank difference of 4 means
9 stone handicaps. If we assign the ranks so the 2 stone games are
fair, we know Black will win almost all the 9 stone games. If we
assign the ranks so the 9 stone games are fair (a much larger
strength difference between adjacent ranks), then we know White will
win almost all the 2 stone games. This means that the extreme
example handicap system is inconsistant with the usual results of
human go, regardless of the ranks we assign.
Another example: Suppose A gives B 4 stones, and B gives C 3 stones,
and these are fair games. The traditional handicap system says A
should give C 7 stones, while the "N-0.5" system says 6.5 stones
(perhaps drawing lots for 6 vs. 7 stones). If either system gives
fair handicaps, then the other system will have half-stone
inconsistancies that cannot be removed by reranking the players.
The issue is which system is most consistant with reality. Both
sides have large amounts of data to support their positions, but as
far as I know, no one has a sample that looks large enough and free
enough of bias to be really persuasive. It takes a lot of handicap
go games to look for half-stone discrepancies in triples of ranks!
Dave Ring
>The issue is which system is most consistant with reality. Both
>sides have large amounts of data to support their positions, but as
>far as I know, no one has a sample that looks large enough and free
>enough of bias to be really persuasive. It takes a lot of handicap
>go games to look for half-stone discrepancies in triples of ranks!
This seems somewhat of a straw man. Are there really people who understand
the argument for "N-0.5" (in your terminology) and still disagree with it?
--
Dave Ring
dwr...@tam2000.tamu.edu
Stewart Strait
: Are there really people who understand
: the argument for "N-0.5" (in your terminology) and still disagree with it?
I think so. The American GO Association chose a modification of the
traditional system a few years ago.
Handicap Rank Diff.
Traditional AGA N-0.5
0 (komi) 0 0 0
1 (no komi) 1 0.5 0.5
reverse komi not used? 1 1 if used
2 2 2 1.5
N, N>2 N N N-0.5
I think all of these systems are in current use by people who
are familiar with arguments such as mine. It's nice to think
that you get a strength difference with the desired additivity
property just by counting the number of moves gained by
the handicap carefully, and I happen to believe it's the best
simple system, but I think several organizations have a lot
of potentially good data that they are interpreting badly.
(At least badly on a rather difficult fine point.)
BTW, if any of the AGA rating officials are reading this, my
information on your system is about two years out of date.
If the formulas have changed, you may want to email them to
me.
Ray Tomes
I have done extensive analysis of thousands of games played in
New Zealand, where we use the n-0.5 system, and these support
the validity of that system.
In our system each player has a rating that is a multiple of
0.05 stone. The movements depend on the result and the
relative strengths of the two players.
-1.0 to -1.95 is 2 dan
0.0 to -0.95 is 1 dan
0.05 to 1.0 is 1 kyu
1.05 to 2.0 is 2 kyu and so on.
If two players ratings are different by 1.05 to 2.0 stones then
the correct handicap for them to play on is 2 stones. In other
words the difference is rounded up to the next whole number if
not already a whole number. Difference of less than .5 is an even
game.
The system works well and there are no anomalies that I can find
in the results which are caused by bias in the system.
Both players move by .15 except that white giving 4-6 stones
moves by .10 and 7-9 stones by .05 and 10+ stones not at all.