Null move depth reductions

[Tom King]

Hi all,

I'm back on the internet (hurrah!). I thought I'd let everyone who knows
me (from email discussions, Paderborn 1995 or Jakarta 1996) and reads
this group know my new email address is:

t...@hatbulb.demon.co.uk

So this posting is more interesting to computer chess folk generally, a
question about null moves. Has anyone any hard data (as opposed to
hunches etc.) about whether null move depth reductions where R is
greater than 2 are "useful", or does it just degrade the search too
much? My program, Francesca uses R=2, and last time I tried R=3 it
seemed a loser. What have you other programmers found? Has anyone tried
reducing by other (variable?) depths?

Regards,

--
Tom King

[Robert Hyatt]

Tom King (t...@hatbulb.demon.co.uk) wrote:
: Hi all,

: t...@hatbulb.demon.co.uk

: Regards,

: --
: Tom King

Bruce and I played with this. Bruce played a few games (R=3) vs crafty with
normal R=2 and we didn't notice any significant difference. Of course, these
were fairly fast games. R=3 becomes serious at deep depths because it
collapses the tree so quickly. R=3 might not be worth much at fast games
however.

In any case, I hate R=2 enough to have no intention at present of trying R=3
again. :)

Bob

[brucemo]

Tom King wrote:

> So this posting is more interesting to computer chess folk generally, a
> question about null moves. Has anyone any hard data (as opposed to
> hunches etc.) about whether null move depth reductions where R is
> greater than 2 are "useful", or does it just degrade the search too
> much? My program, Francesca uses R=2, and last time I tried R=3 it
> seemed a loser. What have you other programmers found? Has anyone tried
> reducing by other (variable?) depths?

I tried 4 (!) and mine still played something that was recognizable chess.
I had it play Crafty a couple of games like this and it did alright.

Fail-high fail-low lots of times.

bruce