[computer-go] Explanation to MoGo paper wanted.

[Magnus Persson]

Quoting Don Dailey <drd at mit.edu>:

> On Wed, 2007-07-04 at 11:34 +0200, Magnus Persson wrote:
>> but what really will make a
>> difference is in the quality in the playouts.
>
> I would like to suggest a more abstract view of things.  In the purest
> form of the algorithm there isn't an artificial distinction between the
> tree and the play-outs.  The algorithm is applied as if the whole tree
> already exists (conceptually) and nodes are updated to the end of the
> game.

I agree with you as usual. And in fact Valkyria do the same things in both
playouts as in the serch tree. And in addition it uses more tricks in the tree
than in the playouts since that is not time critical.

Inspired by this discussion however I pulled out my first Valkyria 
version which
uses alpha-beta search. Moveordering is crude so it is really not a fair
comparison, but it uses exactly the same playout code as Valkyria2.6Bk rated
about 2200. With crude alpha-beta search ValkyriaAB2.6 seems to get 
around 1700
only.

I could probably improve the alphabeta search a lot, but 500 rating 
points seems
to be a lot. Maybe it is possible using all possible techniques for improving
alfa-beta but it would be some really hard work compared to UCT. So to clarify
my position. UCT makes a go program really strong by itself. Go knowledge
manually entered or learned of course could however improve on that without
upper bounds in principle.

Just to disturb the vision a strong go program without hardwired go 
knowledge I
currently think that there are some really important things in Go that are
really hard or even impossible to learn with for examples patterns. The ideal
program would need to learn procedural skills (algorithms).

-- 
Magnus Persson
Berlin, Germany