For some time I've been wondering if I could devise an automated way of training joseki, with my eyes set on Anki (a pretty cool memorization program). Now I've finally reached presentable results.
Basically, there are many flash cards, each showing two full board positions. The question is where comes the next move. You pick an answer, then you see what the player actually played. The games are chosen in such a way that in both the continuation is locally the same. (In fact there are 6 diagrams for each joseki pattern, but each time you get to see only two chosen at random.)
Obviously, there are many possible caveats to this: both the patterns and the diagrams were compiled by computer, so some patterns are subpatterns of others, some are outdated, some games are not representative. Also it asks for just one following move, not a longer logical sequence, focuses on memorization instead of understanding... As of now, it appears to me that many positions are too ambiguous to learn a single "correct" answer.
Nevertheless, Jojola passed one week of my personal testing and I like it so far, so I wanted to share and ask for a feedback from others. Of course, please rate, hate, bait, ... I can elaborate further on how I've built it if anyone is interested.
The attached file is the Jojola deck with 200 most common patterns. The current full version consists of 12000 patterns though the number might change as I'm still working on it.
For the record, I've extracted the patterns and the positions from GoGoD Spring 2018. I consider this a fair use not harming GoGoD copyright in any way. I've generated the diagrams myself and consider the deck to be in public domain, free for anyone to use and redistribute as they see fit. I'll share the scripts I've used if anyone wants to see them (but I don't have the time to make them easily runnable and modifiable at the moment).
(Jojola is an abbreviation for "joseki joseki Laman")
Example:
| Attachments: |
Jojola.zip [344.48 KiB]
Downloaded 413 times |