Re: [computer-go] Territory & nfluence

[Chris Fant <chrisfant@xxxxxxxxx>]

> I would like to get a discussion started about Influence/territory instead.
> My idea is that such a thing is not required to be accurate.
> 
> For example it is rather irrelevant that it knows how much territory a
> player really has, as long as the PROPORTIONS between the players are
> correct. So that an engine can do a search to optimize territory or to
> optimize a percentual increase or optimize an absolute increase or optimize
> a difference with the opponent or keep the territory at least the same etc.
> 
> The actual correct territory, is it really of paramount importance?
> I ask this because my territory algo is only LIKE Bouzy 21/7, the results
> are in fact a bit different.
> So I wonder how much time I should spend on making it more accurate. It
> seems that it is accurate "proportionally", but not "absolutely". Bouzy's
> 21/7 neither is completely accurate and as far as I understand an order of
> magnitude slower to implement.

The tricky thing about the evaluation function is that any errors it
has will be sought out by the search process.

Once you have your current round of learning complete, I have another
stat for you to calculate that I think would be interesting.  When
ordering your moves, how many moves must be checked for each turn to
ENSURE that the pro's move is always considered?  Also, how many moves
on average must be checked?  You will probably have some cases when
the pro's move is ordered quite low by your matcher.  Could you maybe
post a couple of these positions?  My guess is that they would be
interesting, unique tactical situations.
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/