Arimaa Forum - Print Page


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

        Arimaa >> Bot Development >> hardware independent constant strength of a bot
        
(Message started by: leo on Dec 7^th, 2007, 11:34pm)

Title: hardware independent constant strength of a bot
Post by leo on Dec 7^th, 2007, 11:34pm

The Arimaa Computer Championship asks to

Quote:

provide an option to specify a hardware independent play strength such as the number of nodes to evaluate before stopping, or the number of plys to search, etc. When this option is used the program should not look at the time control. This option will not be used in the championship or challenge match, but will be used after these events when the program is made available for others to play against in the Arimaa gameroom. The purpose of this is so that the program will play the same strength independent of the hardware or system load.

Being tinkering again about my "naive AI" bot, I'm having a hard time imagining how I'll implement both time control and hardware independence. This system is not based on a search tree but rather on a shapeless network of tasks at hand, with the priority of each task not being clear, and each task being likely to spawn another bunch of new tasks the amplitude of which can't be known in advance. :)

So the options of the program for enforcing hardware independent strength will probably have exotic names, such as "concentration", "obsessiveness", "versatility" instead of "number of plys". What kind of crazy boat have I embarked onto? :/

Title: Re: hardware independent constant strength of a bo
Post by Fritzlein on Dec 8^th, 2007, 9:42am

Time control and hardware-independent play are indeed antithetical. The idea of hardware-independent play is to assume that you have enough time and limit you bot in some other fixed way. For example, you have a hard cap on the number of "tasks" you refer to above, so your bot stops thinking after a certain amount of thinking, not after a certain amount of time.

Title: Re: hardware independent constant strength of a bo
Post by leo on Dec 9^th, 2007, 12:09am

I'm striving to do so, but when it comes to implementing it everything suddenly looks complicated. :) In time control mode it's difficult to estimate the time each task will take because they are not fixed format operations, they depend on board state and the current plans of the AI; and in constant strength mode it's difficult to estimate the consequences of each task, and the depth of thought they will appear to have.

I'm interested to know how human players limit their strength when playing against a weaker player in a tutor way. At least for me, it seems there's a lot of "no, I'm not going to play that, it would be wicked" kind of thoughts, so there is a sense of what is "simple" and what is "wicked". But it's afterward trimming, the playing mind doesn't limit itself to the more simple moves, it's the tutor mind that marshals the moves...