[computer-go] hybrid monte carlo / TD neural network position evaluator approach

[Mark Boon]

Yes, you'e right, although it's probably the only one?

It shows how careful you must be with heuristics in Computer-Go.  
Although I probably wouldn't mind if my program missed killing the  
group in this case. But heuristics and exact results don't go together.

Mark


On 25-aug-2006, at 3:59, Karol Golab wrote:

> Mark Boon wrote:
>> On 24-aug-2006, at 16:06, Don Dailey wrote:
>>> I'm curious about when that definition can fail, in the sense  
>>> that it's
>>>
>>> actually the best move to play in some situation.  Does anyone  
>>> have an
>>>
>>> example?
>>>
>> The definition used is very restrictive for an eye. Even I would  
>> doubt playing there would ever be a good move, and I'm usually  
>> very careful in saying something like that.
>
>   One may fail to create a nakade in a situation pictured below:
> .xxo.x
> xxoo.x
> o.o..x
> ooo..x
> xxxxxx
>
>   If X does not play at (1,1) the O group lives.
>
>   Regards,
>   Karol
>
> -- 
>    e-mail: Karol.Golab at tls-technologies.com
>       www: http://www.tls-technologies.com
> signature: not found
>
> _______________________________________________
> computer-go mailing list
> computer-go at computer-go.org
> http://www.computer-go.org/mailman/listinfo/computer-go/