Re: [computer-go] Pattern matching - example play

[Vincent Diepeveen <diep@xxxxxxxxx>]

Nullmove already reduces in combination with hashtables for 19x19 the b.f.
to under 7.0

Even a brain dead implementation R=2 and hashtables at 19x19 gives b.f.
10.0 to start with. From there it only goes down and down.

The square root thing is from the start of the 70s or so.

I didn't mention even further selective methods yet here too much yet, but
at least 2 ideas i already did. I'm 100% sure of that. 

Huge advantage in go to keep branching factor low is hard fact that many
moves are with way over 99% sureness prunable. So i don't mean *hard*
prunable, but reducable in one or another way.

That really takes down branching factor a lot.

Then assuming you get a bit reasonable big depths, the hashtable of course
works real great in 9x9 go.

Especially at 9x9 that's the reason why b.f. effectively will be real real
low. Way under 3.0

Vincent

At 00:44 2-12-2004 -0500, Chris Fant wrote:
>> > > 9 ply full width on 9x9?  Certainly not near the beginning of the game.
>> >
>> >It depends on the time settings. For tournament play it was probably a
>> >bit less in the opening and a bit more towards the end.
>>
>> With nullmove R=3 and some selectivity and good hashtables you should get
>> far deeper depending upon the total nodes you see.
>
>Obviously when you start pruning you can get deeper.  Did I just
>imagine the phrase "full width"?  No, I think it was actually in the
>original email.  Sqrt(81 perm 9) = 308E6 leaf nodes to evaluate for
>the first move of the game.  Symetry and a transposition table both
>reduce it by some uncalc-able (by me) amount.  But my point is that
>"full width" does not mean "add your favorite move selection heuristic
>for tree pruning".  I am by no means an expert on search, I just
>believe in preserving the meaning of a given terminology.
>_______________________________________________
>computer-go mailing list
>computer-go@xxxxxxxxxxxxxxxxx
>http://www.computer-go.org/mailman/listinfo/computer-go/
>
>
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/