[computer-go] 10k UCT bots

[Jason House]

On May 13, 2008, at 3:02 PM, Don Dailey <drdailey at cox.net> wrote:

>
>
> Álvaro Begué wrote:
>> On Tue, May 13, 2008 at 2:28 PM, Mark Boon  
>> <tesujisoftware at gmail.com> wrote:
>>
>>> On 13-mei-08, at 15:08, Jason House wrote:
>>>
>>> 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.
>>>
>>> OK, I understand now why a point at the end (or beginning) is a  
>>> little less
>>> likely to be picked. Although I still have doubts whether that  
>>> will lead to
>>> a noticable bias, I'll try to think about it.
>>>
>>
>> I don't care much about it being noticeable. This thread is about
>> putting bots on CGOS that use a reproducible algorithm, to help  
>> people
>> detect bugs in their implementations. As part of specifying what  
>> these
>> bots do, we should all pick the next move in a playout using the same
>> criteria. If we agree to use uniform distribution among empty
>> non-eyeish points, that's what should be implemented.
>>
> I agree with this.   If we get too clever we lose confidence that we  
> are comparing apples to apples by risking bugs and making  
> assumptions that may not hold.


I take an opposite approach: the data will tell us when we are  
comparing apples and oranges. It will help us identify the major  
factors leading to strength. There's probably more subtle variation  
than we realize...


>
>
> - Don
>
>
>>
>>> I would imagine moving an illegal point towards the end and only  
>>> start
>>> including it when the other 'legal' moves run out can lead to  
>>> terrible bias
>>> however because they may not remain illegal for very long and  
>>> actually
>>> become important points to play. A ko-point is probably the most  
>>> extreme
>>> example of that.
>>>
>>
>> I don't think you understood the algorithm. The eyeish point is
>> removed from the lottery only for picking this particular move, not
>> for the rest of the playout.
>>
>>
>>> Anyway, I don't bother to order the empty-point-list or scramble  
>>> them in any
>>> way prior to the game. So the first point is the 1-1 point and the  
>>> last is
>>> the 19-19 point (or whatever boardsize you're playing) so I have  
>>> no qualms
>>> about those moves being a little less likely to be played. Or even  
>>> a lot
>>> less. I think it would actually be beneficial.
>>>
>>
>> Reproducibility was the point, not strength of the bot.
>>
>>
>>> If this asymmetry really bothers you, you could very easily fix  
>>> this by
>>> wrapping the search around. There's no asymmetry in a circle.
>>>
>>
>> That doesn't fix anything.
>>
>>
>> Álvaro.
>> _______________________________________________
>> computer-go mailing list
>> computer-go at computer-go.org
>> http://www.computer-go.org/mailman/listinfo/computer-go/
>>
>>
> _______________________________________________
> computer-go mailing list
> computer-go at computer-go.org
> http://www.computer-go.org/mailman/listinfo/computer-go/