Arimaa Forum - Print Page


    
      
        Arimaa Forum
        (http://arimaa.com/arimaa/forum/cgi/YaBB.cgi)
      

        Arimaa >> Game Analysis >> Game 148397 - akbaggins vs Arimaazilla
        
(Message started by: akbaggins on Jul 7^th, 2010, 8:46am)

Title: Game 148397 - akbaggins vs Arimaazilla
Post by akbaggins on Jul 7^th, 2010, 8:46am

Well, I hope the link's right:

http://arimaa.com/arimaa/gameroom/comments.cgi?gid=148397

Ok, so it was time for me to play the dreaded Arimaazilla, and I got to a pretty bad start as was expected. Then all of a sudden, in the 15th move Arimaazilla loses after repeating the position 3 times!

Did Arimaazilla just short circuit or what?

Title: Re: Game 148397 - akbaggins vs Arimaazilla
Post by Fritzlein on Jul 7^th, 2010, 9:21am

Arimaazilla dates back to early 2003. It was the first competent bot ever written. At that point in the Arimaa Challenge, developers had a lot of low-hanging fruit, such that avoiding repetition losses was not the easiest way to improve a bot's playing strength.

Don Dailey stopped developing his bot Occam (on which Arimaazilla is based) when he realized he wasn't going to win the Challenge. Thus he never got around to repetition avoidance. Reading about his successful chess engine Komodo makes me wonder how far he would have gotten with with his Arimaa bot had he stuck with it.

Title: Re: Game 148397 - akbaggins vs Arimaazilla
Post by akbaggins on Jul 7^th, 2010, 10:10am

Well then I won't be satisfied with this result and try to beat it again later.

btw, just ordered Beginning Arimaa for my birthday :)

Title: Re: Game 148397 - akbaggins vs Arimaazilla
Post by Fritzlein on Jul 7^th, 2010, 10:24am

Happy Birthday! I hope it's a gift that keeps on giving...

Title: Re: Game 148397 - akbaggins vs Arimaazilla
Post by rbarreira on Jul 7^th, 2010, 10:25am

I thought this was actually due to a change in the rules before I read Fritzlein's reply.

In my bot it was a quite high priority to implement the repetition rule, as a lot of the first few games ended in repetition. But I guess it wouldn't be so much an issue on a more evolved bot.

PS: Now that I think about it, the repetition rule might not have been so much of a priority for the first developed bot, as repetitions are much more rare when playing against humans, who don't tend to undo a bot's moves nearly as much as other bots do.

Title: Re: Game 148397 - akbaggins vs Arimaazilla
Post by akbaggins on Jul 7^th, 2010, 9:08pm

@Fritzlein,

Thank you, I'm sure it will help me lots.

@rbarreira

Yes, I thought maybe the bot had been developed before the repetition rule. Maybe I play like a bot myself!

So today I challenged Arimaazilla for a second time.

Around move 10 we got into a tussle over two cats. I was very aware of the repetition rule, and after doing the same thing twice I was ready to change tactics to avoid losing. But then I hear the game over tune--Arimaazilla has repeated 3 times again!!

http://arimaa.com/arimaa/gameroom/comments.cgi?gid=148471

This bot and I have an interesting chemistry :P

Title: Re: Game 148397 - akbaggins vs Arimaazilla
Post by akbaggins on Jul 8^th, 2010, 12:51am

So I played Arimaazilla a 3rd time, and finally managed to get
my rabbit to goal! The bot had captured 7 of my pieces and I had taken nothing from it, but Arimaazilla lacked that killer instinct in the end and I managed to sneak in a rabbit.

http://arimaa.com/arimaa/gameroom/comments.cgi?gid=148485

What's funny about this is that halfway through I was in real danger of repeating a position 3 times, and I probably would have if I hadn't had the rules in mind from prior games.

Title: Re: Game 148397 - akbaggins vs Arimaazilla
Post by rbarreira on Jul 8^th, 2010, 2:56am

Apparently even Bomb suffers from this problem, as my bot found today:

http://arimaa.com/arimaa/gameroom/comments.cgi?gid=148487

Title: Re: Game 148397 - akbaggins vs Arimaazilla
Post by Fritzlein on Jul 8^th, 2010, 3:25am

on 07/08/10 at 02:56:53, rbarreira wrote:

Apparently even Bomb suffers from this problem

Yes, but to a lesser degree. Bomb won't undo the opponent's previous move, but it will fall for a larger cycle. The "no undos" safeguard takes care of most repetitions, but not all. Again, Fotland always thought he had more important ways to improve Bomb than to close that loophole, so he never got around to it.

Title: Re: Game 148397 - akbaggins vs Arimaazilla
Post by akbaggins on Jul 8^th, 2010, 3:30am

You can call that the dog and rabbit dance!!

BTW, Briareus is definitely a scary name for a bot. I remember the creature from the Divine Comedy..

Title: Re: Game 148397 - akbaggins vs Arimaazilla
Post by rbarreira on Jul 8^th, 2010, 3:31am

on 07/08/10 at 03:25:31, Fritzlein wrote:

I'm just surprised that this never caused a loss in the computer championship, at which point it would probably become a priority to fix it... given how easy it is and all.

Title: Re: Game 148397 - akbaggins vs Arimaazilla
Post by rbarreira on Jul 8^th, 2010, 3:36am

on 07/08/10 at 03:30:24, akbaggins wrote:

You can call that the dog and rabbit dance!!

BTW, Briareus is definitely a scary name for a bot. I remember the creature from the Divine Comedy..

I guess a dog and rabbit dance could be better than a dog and pony show :P

As for the name, so far that's the only scary part ;)