Posts:
260
Registered:
Dec 16, 2001
From:
Maryland
Age:
53 Home page
about my opening book and program
Posted:
Jul 16, 2000, 12:12 AM
Andre asked me a couple of questions, I'll answer them here in case anyone else is interested.
There is not really another opening book like mine available. There is one book on Pente called Pente Strategy by Tom Braunlich but it shows only a few openings. There are Pente newsletters from the US Pente Assoc from the 80s. You can see some of the newsletters from the link on my page. There actually is another source which does include about 100 openings - Pente Compendium by Tommy Maltell but it is in Swedish. I guess you could get some ideas by looking at the diagrams. Tommy Maltell is the president of RIF and his email is on the RIF page (and you can link there from mine). You can look at my opening book by downloading the Word document on my page (I have a PDF format for Adobe acrobat also) or by using the opening book editor that comes with my program. You may want to edit either one to make your own custom opening book.
How my program works: basically there are two parts - the search tree and the evaluation function. The evaluation function scores a move by looking at the patterns it makes (and blocks) in each of the four possible lines thru it. It has a database of 914 scored linear patterns that it matches. A pattern that matches an open three scores about 256 and a closed four is about 1024. In addition, it keeps track of lines by each possible move in a situation and then it can tell if a move made threatens a double three or a four-three. It checks for captures made, blocked, and threatened. It scores a capture about the same as making an open three. It also considers the value of pieces that are captured - so capturing a keystone scores as much as making a four.
The search tree is a basic alpha-beta algorithm -- that means it looks at the possible moves from a certain position, then looks at the possible responses from each one of those and so forth down the tree. It looks up to 10 moves ahead that way (or more with VCT). It doesn't really consider every possible move at every position or else the tree would be way too big - it only further considers about the 20 best moves from each position. The current version does not store any of the tree in memory but the new windows version I am working on does. It will use a hash table to store each position scored and will detect repeated positions, improving the speed. Let me know if you want any more detail or any references on game programming.
Mark
This message was originally posted at Yahoo's pente group.
