Arimaa Forum - Print Page


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

        Arimaa >> Bot Development >> Marwin and Sharp differences?
        
(Message started by: Swynndla on May 10^th, 2011, 8:22pm)

Title: Marwin and Sharp differences?
Post by Swynndla on May 10^th, 2011, 8:22pm

So Sharp won this years Computer Championship, but Marwin won the qualifying playoffs (just) to have the right to play the Arimaa Challenge. This suggests that Sharp is stronger when playing against other bots, while Marwin is stronger against humans.

Why is this? It must be because Sharp does something differently that works well against other bots currently, but that it isn't so effective against humans. Perhaps it could be that Sharp deals with the horizon effect well, and so gets one up against other bots, but for humans, we deal with the horizon effect naturally really well, and so it's not much of an advantage, whereas Marwin has something else (I can't think what) that is more effective against humans?

I guess it just shows that beating bots is a slightly different problem-space than beating humans.

Title: Re: Marwin and Sharp differences?
Post by JimmSlimm on May 10^th, 2011, 10:04pm

interesting question, maybe marwin puts more short-term pressure on the opponent, and for humans that could lead to focusing more on defending than trying to create threats themselves?

Title: Re: Marwin and Sharp differences?
Post by ginrunner on May 10^th, 2011, 10:07pm

if i had to guess (am am more than likely completely wrong) I would say that Marwin probably does a better job at avoiding the fuzzy zone where there are multiple complex outcomes and has a more pointed route in mind. I seem to recall that Sharp uses moves made by players 1800+ in its pruning system. I am sure it understands that these moves are very good but i bet it would be difficult to code in a variable for "this is a good and maybe the best move but some fancy footwork can beat it." Hence when computers don't use that fancy footwork they are at a disadvantage. Sharp makes better moves but Marwin makes moves that are more direct in nature?

Title: Re: Marwin and Sharp differences?
Post by Fritzlein on May 10^th, 2011, 11:53pm

I'm not going to work too hard to explain something that could easily be false, i.e. that might appear to be true only due to chance. Given how little evidence we have, it is quite possible that marwin is better against computers and sharp is better against humans! Both the screening and the computer championship were far too close to prove anything.

I expected sharp to be weaker against humans because of the way it gives up a camel hostage at the drop of a hat. Humans are generally better than bots at exploiting a camel hostage once they have one, so I chalked up sharp's victories in the computer championship to the inability of sharp's silicon opponents to punish a wayward camel. I had to change my mind, though, after seeing the screening and after playing against sharp myself. Sharp has shown the ability to launch a devastating swarm after having given up the hostage, so now I believe sharp is plenty strong against humans too.

Title: Re: Marwin and Sharp differences?
Post by Swynndla on May 25^th, 2011, 6:58pm

After reading lightvector's thesis, I have a suspicion that the types of moves that Sharp does it closer to what humans would do (and expect, even if humans only expect them subconsciously), but for Marwin, which is about the same strength as Sharp, I'm guessing works differently (ie guessing it doesn't have the expert move prediction), and so the good moves Marwin does aren't quite as close as what humans would do (and expect) and so humans find it more difficult to play against Marwin, even if Sharp is slightly stronger when playing against other bots.

That is if Sharp really is stronger when playing bots, and Marwin stronger when playing humans (the results don't show any significant distinction as Fritz pointed out).

Title: Re: Marwin and Sharp differences?
Post by Hippo on May 26^th, 2011, 12:29pm

Sharp didn't lose in the screening games ... problem was most of the unfinished pairs given chance to gain points to sharp ... . "Unfortunately" Sharp lost in finished pairs.

This really does not allow us to say what plays better against humans and what agianst bots.

Sharp had advantage at the start as it was much newer (therefore surprising) than marwin. It was winning at the screening opennig and lost at the end. So may be humans learned quickly how to play against it.