Re: [computer-go] Minimax with random evaluations

[Darren Cook <darren@xxxxxxxxx>]

> > >   I think these findings are very intriguing. But how about putting this
> > >   a bit more to the test, and compare '5 ply random' against '5 ply
> > >   legal-move-count'?   
> > What would you conclude if say ...
> >
> >    1) move count did much better.
> That the reason random evaluation is improved by reading is in fact only 
> caused by choosing lines of play leading to bigger move counts.
I think there is a correlation between black territory and available moves 
for black. This is one way of thinking about the game: playing moves in 
order to be able to get more intersections to fill than the opponent. And as 
others have pointed out, random evaluation tree should return a higher 
evaluation for branches where there are more leaf nodes.

So, as Mark says, it will be very surprising if legal-move-count did not do 
at least slightly better than random-evaluation.

What would be very interesting is to see the effect of search depth on a 
more intelligent, but still simple, algorithm, such as the one David Fotland 
(I think) suggested a few months back of always defending the weakest chain 
(again I think that was it). But I guess that is a couple of orders of 
magnitude slower.

Darren
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/