[computer-go] thoughts on 100,000 cgos games

[David G Doshay]

On 7, May 2006, at 7:14 PM, steve uurtamo wrote:

> you
> know that you made an incorrect move sometime
> before that evaluation.  since this is something
> that you could stash while you're playing a game,
> it might be worthwhile as a way to decide when
> to check deeper down some previously-pruned
> branches.
>
> i guess that this presumes the following, and
> i'm curious if it sounds reasonable:
>
> given enough depth of search, even a halfway
> decent board evaluation function will (eventually)
> come up with the right move.  given that depth
> of search on a big board is expensive, better board
> evaluation functions are (generally) worth far
> more than a slight increase in depth of search.

As in my last post, I am not convinced that your statement must
be true. It could also be that the board just evolved into a game
that confuses the evaluation functions, and that we are likely
to make our mistake very soon. Thus my comment that I agree
that this can be used to point out a likely place to inspect ...
but it still must be manually by a human.

Your presumption is interesting, and one we would hope to be
true, but I don't think it is ... unless all evaluations functions in
use now are not "even halfway decent."

Also in your assumption is pruning of the lookahead tree. To
date SlugGo does fixed lookahead width and depth in a game.
The only pruning we do is the selection of the moves for look
ahead. Different branches are not inspected to different depths,
the most normal definition of pruning.

But in as far as your "with enough depth" comment, I can assure
you that our results show very clearly that deeper search results
in weaker play with the evaluation functions we are using. The
problem seems not to be that depth of search is expensive, but
that width of search is expensive. However we take care of that
to a great degree with the cluster, but there again we find that
when we add more levels of branching ... again, we play worse.

But lastly, I agree completely with your observation that

> better board
> evaluation functions are (generally) worth far
> more than a slight increase in depth of search

It is just that those better evaluation functions have been elusive.

Cheers,
David