[computer-go] 10k UCT bots

[Jason House]

On May 13, 2008, at 1:51 PM, Mark Boon <tesujisoftware at gmail.com> wrote:

>
> On 13-mei-08, at 14:10, Álvaro Begué wrote:
>> What others do is the right thing to do. Your method will introduce
>> some biases.
>
> Could you elaborate what bias it could lead to? I also do the same  
> as Jason. I did consider the possibility of a bias but couldn't  
> immediately think of one.

And I was thinking "let's not repeat this topic"...

The probability of picking a point is proportional to 1 + number of  
illegal points before it.

In practice, illegal moves are pretty rare until endgame. At that  
point, it's a trade off between bias and speed. Random number  
generation is not cheap. I have yet to see empirical proof that the  
pick and scan is bad. I've tried both methods in the past and saw no  
measurable difference in strength.


> What good does moving it to the end of the list do? Next time around  
> it's just as likely to be picked as when left in place. Or do you  
> only process the 'end' after the 'front' moves are all tried?

The range of the random number is reduced by one after each failed  
lookup. Shuffled data has no impact on future use of the array of  
empty points.


>
>
> Mark
>
> _______________________________________________
> computer-go mailing list
> computer-go at computer-go.org
> http://www.computer-go.org/mailman/listinfo/computer-go/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://computer-go.org/pipermail/computer-go/attachments/20080513/97a33cc3/attachment.htm