Thursday, March 10, 2005

The suicide checkers opening book

Suicidal Cake is computing an opening book for suicide checkers, the book currently has 50'000 positions. That may sound like little compared to the 1.6 million positions for my regular checkers book. Keep in mind though that the regular book has to cover all 174 3-movers, while the suicide book is just for GAYP - it can leave out lots of lines. Many lines in suicide checkers are considered by humans to lose more or less instantly; for example, the following is believed to be a loss for white:

[Black ""]
[White ""]
[Result "*"]
1. 11-15 24-19 2. 15x24 28x19 3. 12-16 19x12 *

(By similar logic probably the starting moves 1. 10-14 and 1. 9-14 are probably considered to be very weak or even outright losses.) The reasoning is that red will play 3-7 at the right moment; the right moment being the one where it wins on the spot with a clearance. White can do nothing about red playing 3-7 at some point, and must be extremely careful on every move in order not to lose straightaway. By this logic, red continues with 7-11, immediately setting up the possibility of playing 3-7. A lot of moves now already lose by force, such as 22-17 which allows an immediate clearance:

[Result "1-0"]
1. 11-15 24-19 2. 15x24 28x19 3. 12-16 19x12 4. 7-11 22-17 5. 10-14 17x10 6. 6x15 23-19 7. 15x24 27x20 8. 3-7 {here we go!} 12x3 9. 9-13 3x10 10. 1-6 10x1 11. 13-17 21x14 12. 5-9 14x5 13. 11-16 20x11 14. 4-8 11x4 15. 2-6 1x10 1-0

This brings me back to the question of intermediate goals in suicide checkers: One of them is the establishment of such a clearance set-up; Suicidal Cake recognizes this specific pattern of red men on 3,8 vs a white man on 12, and gives it a big bonus. However, if my opening book is right, then this is not the end of the story: The book says that after 7-11, white must continue 27-24, and will draw. However, the path to the draw is very narrow and the last word may not be spoken - the book is not 100% clear on this line.

5 comments:

John said...

Martin, is there a way to set the program to value certain squares given a position?

Martin Fierz said...

Hi John,

well, for me there is: in the evaluation function I can add patterns I want Cake to recognize, and assign specific values for these patterns. For the average user, there is no such possibility - all patterns and values are hard-coded into the program.
"Power users" can change the simple checkers source code; however, simple checkers is not a strong program of course.

Yuri Sorkin said...

> the following is believed to be a loss for white:

> ...
> 1. 11-15 24-19 2. 15x24 28x19 3. 12-16 19x12 *

> (By similar logic probably the starting moves 1. 10-14 and
> 1. 9-14 are probably considered to be very weak or even outright losses.)

That logic is inapplicable. Note that per Russian rules after 1. 11-15 24-19 2. 15x24 28x19 3. 12-16 19x12 4. 7-11 and any next move full clearence follows. Same hold for other beginning moves like 1.9-13 24-19, but NOT for the other color.

Winning by black (white begins in poddavki) after 1. 10-14 or 1. 9-14 in still in years unproven. In 1980's there was a special correspondence tournament held with starting move 1.10-14, and better players won against weaker ones by both colors.

Martin Fierz said...

Hi Yuri,

I thought that just the pattern of a man on 12 vs men on 3,8 - and colors reversed - was thought to be a huge advantage. It probably is too. Whether or not 9-14 or 10-14 are really losses is another matter, but I think it's clear that these moves "look bad" to the human observer. Correct or not?

Yuri Sorkin said...

Absolutely correct, of course.