Arimaa Forum - Print Page


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

        Arimaa >> Bot Development >> Bot_FirstTry
        
(Message started by: 99of9 on Oct 2^nd, 2003, 5:14pm)

Title: Bot_FirstTry
Post by 99of9 on Oct 2^nd, 2003, 5:14pm

In case anyone's been following the failings of Bot_Firsttry...
I just discovered that it was implementing the 99 rabbit system in REVERSE! So no wonder it was trying so hard to kill the first rabbit, and didn't care much after that. I did wonder why it often sacrificed camels for rabbits.

Somehow it was able to beat Shallowblue anyway :-)

So, let's get down to business!

Title: Re: Bot_FirstTry
Post by omar on Oct 5^th, 2003, 10:41pm

I see it's been winning against Arimaazilla a few times.

Also it's moving the rabbits forward more now.
But maybe a little too soon though :-)

Omar

Title: Re: Bot_FirstTry
Post by 99of9 on Oct 6^th, 2003, 8:10am

Yes, that bug fix made a lot of difference in the end!

It's about 50:50 with arimaazilla now.

Actually I haven't changed anything about pushing rabbits forward, I just changed how it values rabbits... earlier it wasn't willing to push them forward *at all* because it would risk losing the 99 point rabbit, now it is all too keen to push them forward, because at worst it will only lose 1 point.

But you're right, I will have to fiddle the evaluation function a great deal later, and not just for the rabbits! However recently I've been working on overall speed, because although it may be equal with arimaazilla, it takes about 20 times as long to play! Some speed improvement already, but I know a few more big chunks left to streamline. Hey, maybe I'll even get an extra ply or two if I get it quick enough :).

Title: Re: Bot_FirstTry
Post by fotland on Oct 6^th, 2003, 9:31am

How fast is it now? Speedy is searching at about 300k nodes per second now

Title: Re: Bot_FirstTry
Post by 99of9 on Oct 6^th, 2003, 12:53pm

Well, first can I say that this is the first alpha/beta I've ever coded, and second, it's running on a 1GHz machine, and third, I'm really a scientist not a computer scientist...

So don't be shocked too much by how slow it goes!

I'll calculate the number of nodes tonight. By the way, how exactly is a node defined? Will "# of boards at which an evaluation is called" do?

Title: Re: Bot_FirstTry
Post by fotland on Oct 6^th, 2003, 2:57pm

I'm running on a 2.4 GHz Pentium.

I'm using the usual computer chess definition of nodes,
which is the number of board positions visited. For me, I just count calls to make_move(). Most of the board positions are not evaluated of course.

There is quite a bit of variation in the speed, depending on the position, since some of the evaluation terms are quite expensive to calculate, and I haven't implemented evaluation caching yet. The range is from about 110K nps to about 450K nps.

Title: Re: Bot_FirstTry
Post by 99of9 on Oct 6^th, 2003, 3:40pm

Hmm... so you include board positions midway through a 4 play move? Is that different to a board position where the bot plays a 2 play move and then stops?

I do evaluations on 18k board positions per second.

So I guess no matter how you count it I still have to get an order of magnitude or so.

99

Title: Re: Bot_FirstTry
Post by fotland on Oct 7^th, 2003, 1:19am

I include positions partway through a 4-step move, since I make moves one step at a time.

Title: Re: Bot_FirstTry
Post by 99of9 on Oct 7^th, 2003, 4:07pm

Now up to 30k moves per second.

Title: Re: Bot_FirstTry
Post by fotland on Oct 7^th, 2003, 10:52pm

The game today against Omar had so many advanced pawns in complex positions that it dropped ot 90K nps, almost the same as yours, since my machine is over twice as fast.