RE: [computer-go] Tactical move generation

["Anders Kierulf" <anders@xxxxxxxxxxx>]

> It's generally good enough to play on the liberties of the
> target group, on liberties of liberties of that group, on
> liberties of adjacent groups with same or fewer liberties,
> and in points next to liberties of enemy groups where playing
> on the liberty itself would lead to immediate capture.
>
> David

It really depends on your goal. David's list of moves will get you an
efficient solution to most capture problems, especially ones that crop up in
computer Go games. If you want to be sure to include the killing/saving
move, you need to do something like Thomas Thomsen's lambda search, but even
then you're artificially limiting the number of liberties during the search,
so there will be real-life capture problems you can't solve. If your goal is
to solve most capture problems most efficiently, you might start with
David's set of moves and add moves for special cases like:
- Approach moves.
- Cross-cut tesuji.
- Moves that force approach moves to be made.

Another issue: When you compare the number of liberties of a block and its
adjacent blocks, do you just count the liberties, or do you include approach
moves and compare the number of moves it would take to capture the block?
The latter is more correct, but may be sufficienctly slower that you end up
solving fewer problems overall within a given time limit. Also, sometimes
you need to play at the liberties of blocks adjacent to blocks adjacent to
the block in question -- how far do you go?

I've found the collection of 80 problems in "Rescue and Capture" by Yang
YiLun (Yutopian) to be a good source of interesting capture problems.

Anders Kierulf
www.smartgo.com


_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/