Arimaa Forum - Print Page


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

        Arimaa >> Site Discussion >> Apparent bug in bot_ArimaaScoreP1
        
(Message started by: Janzert on Jun 13^th, 2012, 4:36pm)

Title: Apparent bug in bot_ArimaaScoreP1
Post by Janzert on Jun 13^th, 2012, 4:36pm

Gurugsk recently posted a comment about Game 234463 (http://arimaa.com/arimaa/gameroom/opengamewin.cgi?client=1&gameid=234463&role=v&side=b). What caught my eye when watching it though is that it seems to expose a bug in ArimaaScoreP1.

This bot is suppose to choose it's move by maximizing the old arimaa scoring method for the move it makes[1]. The full formula is given by omar in this thread (http://arimaa.com/arimaa/forum/cgi/YaBB.cgi?board=talk;action=display;num=1213818953;start=0#3). The important part though is that the score is only determined by how far rabbits are advanced and what pieces are left on the board. Where advanced rabbits and more pieces give a higher score.

The bug is shown in the above game at moves 68, 70, 72, 73, 76 and 79, ArimaaScoreP1 sacrifices a more advanced rabbit to just move a less advanced rabbit laterally. These moves should have resulted in a decrease in the score while just shuffling other pieces would have kept the score the same.

I'm rather surprised that this hasn't been noticed before, have I just forgotten or missed where this was discovered previously?

With so many games played by the ArimaaScore bots at this point; I'm also not sure whether it is better to fix the bot or just leave it as is and note that it doesn't actually "play to spec". It would be rather interesting to go back through old games and see how often buggy moves have been played.

Janzert

1. I'm not sure if what is supposed to happen when choosing between moves with a tied score has been specified. I would guess either a random move is chosen or it is deterministic but arbitrary based on the move generator.

Title: Re: Apparent bug in bot_ArimaaScoreP1
Post by Fritzlein on Jun 15^th, 2012, 9:05am

I vaguely remember concluding years ago that ArimaaScoreP1 didn't follow the score formula exactly, and I believe I commented on it somewhere, but it didn't seem worth making a fuss about. The score formula doesn't have a special status in my mind; it is just one arbitrary beginner evaluation function among many.

Title: Re: Apparent bug in bot_ArimaaScoreP1
Post by Janzert on Jun 16^th, 2012, 6:37pm

on 06/15/12 at 09:05:04, Fritzlein wrote:

The score formula doesn't have a special status in my mind; it is just one arbitrary beginner evaluation function among many.

The only reason ArimaaScoreP1 had special status in my mind was because of its position as ranking anchor. I am a bit fond of the simple, reproducible anchor player. But that is really just a theoretical consideration, in practical terms I can't say that it matters.

Janzert