The rating system was adopted by Czech Go Association at the beginning
of 1998. Originally it was designed to serve only the needs of Czech go
community. Later on we decided to enlarge the tournament database by
including other European tournaments and made it comparable with
the former EGF database. The system has been used for computing
the official EGF ratings since November 1998.
The included events should meet following conditions:
- The event took place not earlier than January 1, 1996.
- The tournament is recognized as
EGF classified event.
- Both handicap and even games are accepted.
Fast games (less than 30 minutes) are not counted because their results are
less consistent. We allow for handicap games since their inclusion helps
to keep the correspondence between ratings and grades. At the same time
it also enables to follow the commonly accepted requirement on the number
of handicap stones to be used in a game between opponents of different
strength if both players should have the same chance of winning.
We would be glad if you could send us any tournament table meeting
the specified conditions and not included so far. If you do so, please,
try to follow our
instructions for sending
the tournament results.
The rating list includes all "European" players who participated
at tournaments that are in the database. An average 1 dan should have
Go Rating about 2100 and the difference between grades is set to 100
(6 dan = 2600, 1 kyu = 2000, 20 kyu = 100 etc.). These values are
also used to initialize the rating of a new player.
The scale of professional grades is set provisionally to 1p =
7d = 2700, 2p = 2730, ... , 9p = 2940.
If player's rating drops below 100, it is reset to GoR=100 which
is fixed as the bottom value.
Example 1: The ratings of top European amateur players Guo Juan
and Lee Hyuk fall in the region of 3p-4p.
Example 2: The player with GoR = 2050 can be regarded as either
a weak 1d or a strong 1k.
Depending on players' tournament results and on the various ranking
systems used in different countries, the correspondence between grades
and GoR may not work quite well especially for lower kyu grades.
However, it gives a relatively good measure of player's strength
provided that he/she has participated in at least 3-5 tournaments.
If a player has not participated at any considered tournament for
some time (this period is set to 2 years for dan players, 12 months
for 1-10 kyu, and 6 months for 11-20 kyu; the actual year/month is
not counted), he/she drops out from the current rating list.
However, the player's rating is kept in the database of players and
it is used once he/she appears again at any tournament in the future.
System description
The rating system is derived from ELO rating system used by
International Chess Federation (FIDE). It is based on the idea that
one can define a probability of winning a game (so called winning
expectancy SE) depending on the difference of opponents
ratings D=RB-RA.
For the player with lower rating (let us call him "player A")
the quantity is given by
(1)
SE(A) =
1 / [eD/a + 1] - ε/2
The winning expectancy of his higher (or evenly) ranked opponent
("player B") is obtained from the equation
(2)
SE(A) + SE(B) = 1 - ε
If ε=0, Eq.(2) simply states that the sum of
both winning expectancies should be normalized to one. However, such setting
suffers from long term deflation as the new improving players take points
from already established players. This is countered by various instruments
like an existence of rating bottom, winning expectancy setting, rating resets
in some specific cases and finally by introduction of a small parameter
ε > 0.
At the moment, we use ε=0.016,
a value fitted to balance rating variations in dan region. Although such
small value has negligible effect on variation of player's rating at one
tournament, the parameter ε allows to tune the
long term system behaviour in a desired way.
A typical behaviour of SE is shown
in Table I where the quantity was calculated with the parameters fixed at the values:
a=115, ε=0. This setting gives about 30% probability
for beating a 1 grade stronger opponent. Since stronger players play more
consistently than the weaker ones, the probability of beating a 1 grade
weaker opponent tends to rise with player's grade. This fact is reflected
in our system by an appropriate dependence of parameter a on the
rating value of player A. The complete setting is shown in Table II where
the corresponding probabilities of beating a 1 grade stronger opponent
are given as well. As one can see, 20 kyu is expected to win about 40%
games with one grade stronger opponent while the top amateur players
should win only 20% of their games with 100 rating points stronger
opponents.
Table I: Winning expectancies SE for some
selected rating differences D
calculated with a=115, ε=0.
D |
20 |
40 |
60 |
80 |
100 |
120 |
140 |
160 |
180 |
200 |
300 |
400 |
SE(D) |
0.457 | 0.414 | 0.372 | 0.333 |
0.295 | 0.260 | 0.228 | 0.199 |
0.173 | 0.149 | 0.069 | 0.030 |
In a single even game the rating of a player changes by
(3)
Rnew - Rold = con *
[ SA - SE(D)]
where SA is the achieved result
(SA = 1, 0 or 0.5 in case of jigo)
and the factor con characterizes the magnitude of the change. In our
system the parameter con is a decreasing function of player's rating
specified in Table II.
Table II: The dependence of parameters con and a on
the rating. For convenience the winning expectancies (in percents) for
beating 100 points stronger opponent are shown as well. We use linear
extrapolation between the points given in the table and con=10,
and a=70 for GoR > 2700.
GoR |
con |
a |
SE(100) |
|
GoR |
con |
a |
SE(100) |
100 |
116 |
200 |
37.8 |
|
1500 |
47 |
130 |
31.7 |
200 |
110 |
195 |
37.5 |
|
1600 |
43 |
125 |
31.0 |
300 |
105 |
190 |
37.1 |
|
1700 |
39 |
120 |
30.3 |
400 |
100 |
185 |
36.8 |
|
1800 |
35 |
115 |
29.5 |
500 |
95 |
180 |
36.5 |
|
1900 |
31 |
110 |
28.7 |
600 |
90 |
175 |
36.1 |
|
2000 |
27 |
105 |
27.8 |
700 |
85 |
170 |
35.7 |
|
2100 |
24 |
100 |
26.9 |
800 |
80 |
165 |
35.3 |
|
2200 |
21 |
95 |
25.9 |
900 |
75 |
160 |
34.9 |
|
2300 |
18 |
90 |
24.8 |
1000 |
70 |
155 |
34.4 |
|
2400 |
15 |
85 |
23.6 |
1100 |
65 |
150 |
33.9 |
|
2500 |
13 |
80 |
22.3 |
1200 |
60 |
145 |
33.4 |
|
2600 |
11 |
75 |
20.9 |
1300 |
55 |
140 |
32.9 |
|
2700 |
10 |
70 |
19.3 |
1400 |
51 |
135 |
32.3 |
|
|
|
|
|
All the following examples are computed with ε set at zero:
Example 3: Both opponents have the same rating
RA=RB=2400. This gives
D=0 and SE=0.5
for both players. If player A wins, his new rating will be
Rnew(A) = 2400 + 15 (1-0.5) = 2407.5
At the same time, the rating of player B drops by 7.5,
i.e. Rnew(B)=2392.5
Example 4: RA=320,
RB=400 and player A wins:
a=189, SE(80)=0.396
Rnew(A) = 320 + 104 (1-0.396) = 383
Rnew(B) = 400 + 100 (0-0.604) = 340
The system also allows to include handicap games assuming that the rating
difference D is reduced by 100(H-0.5),
where H is the number of given handicaps. Note, that it can happen
that the winning expectancy of a weaker player is larger than
SE of the stronger player (i.e. the weaker player is
expected to win the game) if the number of given handicaps (reduced by 0.5)
is larger than the absolute value of (RA-RB)/100.
Example 5: RA=1850,
RB=2400, player A takes 5 handicaps and wins:
D=100,
a=90, SE(100)=0.248
Rnew(A) = 1850 + 33 (1-0.248) = 1875
Rnew(B) = 2400 + 15 (0-0.752) = 2389
Remarks
The post-tournament ratings are calculated assuming that every player
enters the games with all opponents at the same initial (pre-tournament)
rating. It means that the ratings are not reevaluated after each game
(round) and the "new" ratings are computed from the "old" ones adding
all contributions from the games the player completed at a given tournament.
In other words, we assume that the ratings of players do not change in the
process of one tournament.
If a rank professed by the player had improved significantly (at least
by 2 grades for amateur players or by 1 professional grade) with respect
to the highest previously professed rank, the rating of the player is reset.
This measure helps to deal with fast improving players and with players
who participate at included tournaments only occasionally. To avoid
undesirable oscillations in the bottom part of the rating list the drop
of player's rating at one tournament is restricted to 100 points.
The work on the program that calculates the ratings is still in
process. In the future, the program should also be used for some
statistical purposes and for the maintenance of the database
of go players. We welcome any comments and suggestions.
Acknowledgement:
The rating system has been tested, with permission, on the database of
professional games provided with the
GoGoD Encyclopaedia CD
produced by J. Fairbairn and T.M. Hall.
We thank R. Kok and Ch. Gerlach for providing us with their databases of
tournament results and A. Engels for maintaining a web page which helped
us immensely to collect the tournament data. We would also like to thank
J.-L.Gailly for advising us on the correct treatment of handicap games.
Finally, we acknowledge a help of many people who have contributed with
their comments and suggestions, and especially of those who have kept us
informed by sending new tournament results.
Notes on submission of tournament results
The results must be submitted through the
online parser which is present in the backend section of this website.
In order to be allowed to submit data, you will be requested to register a user account.
If you encounter any problems, report them to
egd@europeangodatabase.eu
When the results are sent we need to know:
- the name of the event
- the place (city) where it was hold
- the exact date of the event
- tournament class (or the time limit including
byoyomi)
- handicap strategies (if any handicap games occured)
Tournament classes
EGF recognizes four tournament categories:
- class A:
well organized tournament, no handicaps in the top group, recognized by EGF member
time limit requirements: adjusted time minimum 75 minutes, basic time minimum 60 minutes;
(Fischer time: basic time 45 mins, adj. time for 120 moves: 75 mins - see remarks)
weight for inclusion to EGF ratings: 1.00
N.B.: As of November 5th 2014, according to a decision taken by the EGF board, the constraint about handicap games is removed.
- class B:
well organized tournament recognized by EGF member
time limit requirements: adjusted time minimum 50 minutes, basic time minimum 40 minutes;
(Fischer time: basic time 30 mins, adj. time for 120 moves: 50 mins - see remarks)
weight for inclusion to EGF ratings: 0.75
- class C:
casual or club tournament recognized by EGF member
time limit requirements: adjusted time minimum 30 minutes, basic time minimum 25 minutes;
(Fischer time: basic time 20 mins, adj. time for 120 moves: 30 mins - see remarks)
weight for inclusion to EGF ratings: 0.50
- class D:
Tournaments played on Internet, accepted since 28th July 2015
time limit requirements: adjusted time minimum 50 minutes, basic time minimum 40 minutes;
(Fischer time: basic time 30 mins, adj. time for 120 moves: 50 mins - see remarks)
weight for inclusion to EGF ratings: 0.25
Only tournaments held by national organisations, member of the EGF, are accepted for inclusion into this class.
The organisers must ensure that all the games have been played fairly, and all the players
are members of EGF members.
Remarks:
-
Adjusted time (TA) is calculated as
TA = basic time + time equivalent to 45 (60) moves in standard
(Canadian) byoyomi.
Sudden death - implying adjusted time = basic time - is acceptable,
provided all other criteria are met.
-
As of October 2010 the Rating Commission has approved the inclusion of tournaments played with the Fischer timing system. These are the rules for classify such tournaments:
TA = basic time (BT) + bonus calculated for 120 moves
- Class A: minimum BT 45 minutes, minimum TA 75 minutes (e.g.: 45 minutes + 15'' per move)
- Class B and D: minimum BT 30 minutes, minimum TA 50 minutes (e.g.: 30 minutes + 10'' per move)
- Class C: minimum BT 20 minutes, minimum TA 30 minutes (e.g.: 20 minutes + 5'' per move)
The EGF Ratings Committee reserves the right to adjust
the class of a tournament if considered necessary. The class of
the event should be known before it is held. The organizer may
demote the class in tournament announcement if the lower class
suits better to his intentions (e.g. when new tournament system
or rules are tested). The events that do not meet the specifications
mentioned above are not recognized as classified EGF tournaments.
Tournament table format
The rating program processes the tournament tables written in the
form of ASCII text files. The data record for each player is to be
given in one line. Empty lines and everything following the semicolon
sign are ignored. The results are presented in a string of entries
separated by blanks. The number of rounds (included results) must
be the same for each player and each result entry is given in the form:
<opponent's place><symbol>[/<colour>[<h>]]
where
<symbol> is + (a win), - (a loss) or = (jigo)
<colour> means the colour of stones ('b/w' = black/white)
<h> specifies the number of given handicaps
Examples: 35=, 23+/w (equivalent to 23+/w0), 11-/b5
The handicap specification can be omitted if the handicaps are computed
from the difference of opponents' grades (minus a given reduction).
If the player did not participate in a given round, the
result is still to be specified in one of the following ways:
0+ a free win
0- a free round
0= a free round (0.5 added to the score and MM)
Examples of some tournament tables:
t970920b.h2
| -
| handicap games, handicap = difference
of grades - 2 (defined by the file extension; h9 extension used for
even games)
|
e961122.h9
| -
| even games, example of a
team tournament
|
t990220f.h9
| -
| even and handicap games specified
|
In the
Admin backend you will find extensive documentation, with further specifications for formatting the results table.
References to other rating systems: