[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Value of a group
Mousheng Xu wrote:
>> A group with 10 stones is worth 0 points if it lives, 20 points to enemy if
>> it's captured (Japanese scoring).
>
> The value of a group may mean it's own value + the territory it
>holds + the potential to obtain more territory and the potential to kill
>more enemy stones. The above and the below algorithms are a bit
>simplified.
> This problem involves life/death, territory counting, potential
>calculation. It really is the key of go programming.
Yes, I suppose so.
I want it mainly for choosing goals and priorities in the searches, so it
does not need to be exact. A few more thoughts:
1.Is this a good application for a neural net/genetic algorithm?
2.Is it easier if we think about about relative rather than absolute values?
Eg. I might not be able to work out the exact values for chains A and B,
they are both in semeai with similar sized groups, but A has more stones and
territory so A is bigger. Or B is in a semeai with two enemy groups, so even
though it has fewer stones and territory we give it the higher priority.
> An extra question:
> Has anybody fingured out the hightest value for the first stone
>(the smallest group) on board by a NATURAL algorithm?
This is too difficult I think. However, on a 5x5 board the first move is
worth 24 points, because the second player can't live. I don't know if that
counts as a 'natural' algorithm though.
David Mechner wrote:
>A fundamental concept in go strategy is that of an "imporant" (cutting)
>stone. Important stones separate groups that might be attacked,
>unimportant stones separate nothing, or separate live groups. So on top
>of immediate territory value assocated with a group you have to add the
>expected profit from attacks against the groups the group in question
>separates.
How about this way of judging cutting stone importance:
1.Pretend it has absolute life. What is the score?
2.Pretend it is dead. What is the score?
If a big difference it's important. Maybe this method could be used to judge
the value of any group? It sounds slow though!
Finding the value of attacks on a group can maybe only be done with search?
Darren