Disclaimers

This is a unilateral proposal by David Bofinger, not accepted by the AGA. Results are preliminary for a variety of reasons, discussed in Issues below.

Files

The software runs under Microsoft Windows only. Source code (in Delphi) is available.

Within a main directory should be contained:

• A directory "Open", containing:
  • Files with filenames of the form "*.aga": these are output from Neville Smythe's AGATHA program and describe AGA open tournaments.
• A directory "Program", containing:
  • The application "Rater.exe".
  • The Delphi source code, if provided.
• A directory "Input", containing:
  • A text file "Aliases.txt", edit this to tell the program that two different names refer to the same player.
  • A text file "Years.txt". Edit this to change the discounting rate; and the earliest year in which a player can have played their most recent game, and still appear in the active list for output purposes.
• A directory "Output", into which the program places output files, including:
  • Lists of information about players, active or otherwise, in various orders, in HTML and ASCII format.
  • An attendance record, mainly intended for helping people edit the aliases file (see below).

Alias File Structure

Each line in the alias file consists of several columns, separated by tab characters. The first column is the alias, a name by which this player was known in a tournament. The next three columns are the player's family name, personal Western name and personal East Asian (Chinese, Japanese or Korean) name, respectively. The last column has information about the player, its first character describes whether the player is active or not.

For instance, suppose the alias file contains the following lines. (Activity column has been deleted for brevity.)

The meanings of these lines are as follows:

• The first and second lines say that "Bill Chang" and "Chang Tsui" are the same person, and give his full name.
• The third and fourth lines say that "Terry Chang" and "Chang T" are the same person, and give his Western name. His Chinese name isn't recorded and the author would probably be grateful to be told what it is.
• The ! at the start of the fifth line causes it to be ignored by the program. The line is there as a reminder to the author that "Chang T" is not the same person as "Chang Tsui".
• The sixth line describes a player known only as "Chang B". The author would probably be pleased to know more about this player.
• The seventh line says that "Chang B" is probably not the same person as "Bill Chang", though the author isn't sure. Again, the program is effectively going to ignore this line because it starts with !?.

Please note this was only intended as an example: none of these lines appears in the real alias file though there are many lines like them.

Procedure

Most rating schemes assign a strength to each player and adjust the strengths from time to time, either after each game or at fixed intervals. This one does not. Rather, the entire rating scheme is recalculated as often as needed.

This is numerically much more intensive and may be an issue for go associations much larger than Australia's. Of course, improving computer technology will quickly chew away at this problem. The Australian scheme has about 200 players and this scheme is probably not suited to schemes more than, say, ten times larger. However, if anyone from a larger scheme is interested in using this approach I'm happy to work on squeezing some more performance out of the code.

Controls

The input files are automatically read in, and the strengths calculated, when the program is started. The delay for this is a couple of seconds on my PC, which was mid-range in 2002.

The report buttons display a list of player estimated strengths, sorted in various ways. The "Calc" column is the effective dan rating in hundredths of a dan, so a rating of +500 indicates a player who would be average strength amongst 5-dans.

The "Min" and "Max" columns either side of some columns are estimates of -1 and +1 standard deviation respectively. There's very notionally a probability of 83% that the quantity lies somewhere in this range. An exaception to this is the rank, which really is the minimum and the maximum ranks ever claimed.

The slider allows examination of the win-loss record of any particular player.

Credits

System developed and software written by David Bofinger. Thanks to Neville Smythe for providing tournament records. Thanks to Devon Bailey and Robert Vadas for assistance with the aliases file.

I welcome feedback at David.Bofinger@dsto.defenceSpamProofing.gov.au (delete the spamproofing).

This page is hosted by GeoCities, in return for carrying their advertising they will give you a free home page much like mine. Everything on this site varies without notice, especially after I get feedback.