BikMove v1.1

[AartBik]

I just released version 1.1 of "BikMove" (an 8x8 checkers engine that plugs in Martin Fierz' CheckerBoard GUI). Compared to the previous version I announced earlier in this forum, I have improved the evaluation function and implemented a transposition table. The nodes-per-second rating dropped two-fold as a result, but the win-loss ratio over 1.0 is about a factor of three in a 5 second engine match, so it seems the right trade-off. Download at the usual place: http://www.aartbik.com/MISC/checkers.html

I am planning to post some results against other engines later.

[Ed Gilbert]

Hi Aart,

Nice to see some activity on a checkers engine for checkerboard. I played with it a bit tonight. I guess you have fixed sizes for hashtable and egdb cache? Those menus are grayed out under Engine Options.

If you have any interest in making a 10x10 international draughts version I have a (very) doctored up version of checkerboard for 10x10 that you're welcome to use.

-- Ed

update: I found the Engine About info and see you're using 256mb for egdb cache. BTW there's a minor copy/paste error in the copyright file -- the word "chess" instead of checkers.

[AartBik]

I guess you have fixed sizes for hashtable and egdb cache? Those menus are grayed out under Engine Options.

Yes, fixed sizes for all tables. Currently 240MB for transposition table and 256MB for the egdb cache. Like you found, I report the latter in "About", but in the next version I will further enhance the options mechanism (at least the get part). The "Help" and "Static Evaluation" features work with v1.1 already.

Good catch finding the "chess" mistake in the license by the way!

If you have any interest in making a 10x10 international draughts version I have a (very) doctored up version of checkerboard for 10x10 that you're welcome to use.

That definitely sounds interesting, although it may depend a little bit on the success of BikMove. If it remains the weakest 8x8 engine on the block, I probably have little to contribute to 10x10 As I write this, an engine match with simple checkers is running.

[AartBik]

Here are the results of a first test match between BikMove v1.1 (240MB hash, 256MB database cache, 2-8 piece endgames) and SimpleCheckers v1.14, allocating 1 second per move for the 3-move openings. For good sportsmanship, since SimpleCheckers does not support the endgame database, BikMove will claim any win/loss/draw result when detected (i.e. it does not try to draw in a lost position). A very thin victory given that BikMove on the average thinks about two plies deeper (not counting capture extensions) and queries endgame databases during the search. Clearly I have a lot of work to do before I can even think about playing with the big guys.....

Code: Select all

BikMove v1.1 vs. SimpleCheckers v1.14 : W-L-D: 75-64-149

[AartBik]

I have a lot of work to do before I can even think about playing with the big guys.....

Yes, a lot of work......
Here are the results of another match.

Code: Select all

BikJump v1.1 vs. GUI checkers v1.05 : W-L-D : 6-183-99

[64_bit_checkers_engine]

By the way, I gave Jon Kreuzer, the GUI Checkers author, my 6-piece perfect play databases and the access code to probe them in RAM (over a year ago actually).

He says it plays much better with them.

I see he has not released it yet, I sent him an email inquiring about it.

[TAILLE]

By the way, I gave Jon Kreuzer, the GUI Checkers author, my 6-piece perfect play databases and the access code to probe them in RAM (over a year ago actually).

He says it plays much better with them.

I see he has not released it yet, I sent him an email inquiring about it.

Hi,
What do you mean exactly by "6-piece perfect play databases" ? I guess it is the optimisation of the distance to win but I am not familiar with checkers rules. In international draughts game everyboby knows that this "6-piece perfect play databases" is note quite easy to build due to the existence of the "25 moves" rule. Does it exist such rule in checkers ? If yes contratulation for having managed to build this perfect play database.

[64_bit_checkers_engine]

What do you mean exactly by "6-piece perfect play databases" ? I guess it is the optimisation of the distance to win but I am not familiar with checkers rules.

Hi,

Yes, I solved "Distance to Win" for the 6-piece database (and larger databases) for checkers. This is different from your "Distance to Conversion" for International Draughts. Below is the longest win for Red to Move with 7 pieces in 8x8 checkers.



Red to move will win in 253 plies from the position shown above.

Some of my data for the smaller 6-piece database is shown here:

http://www.liquidnitrogenoverclocking.com/report.txt

It is a large text file, if you click on it, you might have to wait a while for your browser to load it if your internet connection is not fast.

In international draughts game everyboby knows that this "6-piece perfect play databases" is note quite easy to build due to the existence of the "25 moves" rule. Does it exist such rule in checkers ?

The database I built only computes the Distance To Win as if no such "rules for making progress" exist. The rules that are in place to prevent "human exhaustion" should not apply to computers, in my opinion. There are some remarkably complex positions in 8x8 checkers, and I am only interested in finding the longest wins given "unlimited number of moves" and "the perfect way to play" no matter how long it takes.

But I understand why International Draughts has the 25-move rule.