| Life In 19x19 http://www.lifein19x19.com/ |
|
| How to properly compile a Go database - a case study http://www.lifein19x19.com/viewtopic.php?f=10&t=10582 |
Page 1 of 1 |
| Author: | macelee [ Sat Jul 19, 2014 11:13 am ] |
| Post subject: | How to properly compile a Go database - a case study |
Compiling a Go game record database is a hard job! Old games before the Internet age are almost always transcribed from printing materials, such as Go yearbooks and other game collections published in oriental countries. Before the wide adoption of computer-based typesetting systems, Go games in these books had to be prepared manually, which means human errors were inevitable. Good editors of game collections need to be aware of this problem, and spend time to fix such human errors wherever possible. Here is a real example - the game in question was played between Yan An 7-dan and Wang Hui 6-dan during the 1992 Chinese Individual Championship. The game record can be found on page 1488 of a set of Chinese books published in early 1990's - the set widely known as '4000 games'. The printed game records clearly contain errors. Because the players are not exactly famous and the tournament not very popular, it's unlikely that we can find an independent source. So some hard work is required to fix it. Here I will show three versions of this game record: The first version was taken from an early version of GoGoD CD (game ID 1992-09-02d). The highly respected T Mark Hall apparently spot the problem too. He managed to fixed move 44 which was printed one cell to the left in the book. But he had to stop at move 169, and left a note saying "the printed record is defective from (at least) move 168, which probably needs to be transposed with 169 because a ko fight is shown there, but the subsequent moves do not make sense and are omitted here. 235 moves were shown". The second version can be found on 'gobase.org'. The gobase owner also attempted to fix the game, in a different way. As we can see, his fix was not correct either, because in the end there's a big ko at H6 which is not settled. When I compiled this game for Go4Go, I first fixed Yan An's rank as there's clear evidence that he became 7-dan 3 years earlier. Then I spent several hours on the move sequence and finally found a sensible way to fix it. I realised that move 101, on the same row as move 44, was also shifted one cell to the left - this isn't obvious because move 101 as printed on the book doesn't look like an incorrect move. But as you can see once I shifted this stone, the game can progress natually to the end. Solving such problem is like working on jigsaw puzzles, it is hard to do, but once you find the solution you know it is correct. Over the years I have fixed about 100 games like the one above. It was immensely satisfying to be able to solve such puzzles. I hope you agree with me that when I say that Go4Go collection is of the best quality there is a very good reason to make that claim. Unfortunately, there are always people there not respecting these hard works. Recently I found a big 'collection' of games that apparatently contain most games I manually fixed. A while ago, another collection appeared online and claimed to be in 'public domain' apparently contains GoGoD's version of the game above. Ask yourself this question: what is the probability for an independent author to use exactly the following result field as GoGoD does and also use GoGoD's signature way for Chinese Komi: "RE[B+R (moves beyond 169 not given)]" "KM[2.75]" It takes a lot of more work than downloading a large quantity of games online and processing them with computer programs to create a Go game collection. When a new game collection is out there from nowhere, there's almost certainly something fishy in it. Please think twice before giving your support to those new 'collections'. |
|
| Author: | Bonobo [ Sat Jul 19, 2014 11:33 am ] |
| Post subject: | Re: How to properly compile a Go database - a case study |
Forgive me, please, you are a MUCH stronger player than I am, but I don’t understand this: macelee wrote: [..] What I don’t understand is this: why should the unsettled Ko be an indication for incorrect transcription when “B+R”? As said, I am a weak player (~13-15k) … does what you say mean that W would in any case play the Ko fight to the end before resigning?The second version [..] As we can see, his fix was not correct either, because in the end there's a big ko at H6 which is not settled. [..] TIA, Tom |
|
| Author: | macelee [ Sat Jul 19, 2014 11:56 am ] |
| Post subject: | Re: How to properly compile a Go database - a case study |
Bonobo wrote: Forgive me, please, you are a MUCH stronger player than I am, but I don’t understand this: macelee wrote: [..] What I don’t understand is this: why should the unsettled Ko be an indication for incorrect transcription when “B+R”? As said, I am a weak player (~13-15k) … does what you say mean that W would in any case play the Ko fight to the end before resigning?The second version [..] As we can see, his fix was not correct either, because in the end there's a big ko at H6 which is not settled. [..] TIA, Tom Hi Tom, the ko at H6 is about the life/death of 7 stones, having a value of at least 19 points. This is significantly larger than the yose moves they played (for example move 235 only has a value of 2 points). So clearly the ko should not be there. |
|
| Page 1 of 1 | All times are UTC - 8 hours [ DST ] |
| Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group http://www.phpbb.com/ |
|