[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Sharing Secrets (was: [computer-go] Computer Go hardware)
All this "I'll show you mine if you show me yours" is like kindergarten.
If you want to share, do it without expecting any immediate payback,
otherwise you're already in a different ballgame from that which gave
you gnugo, linux, or the web in the first place.
> Still I feel comfortable to share because if someone would implement 64-bit
> Z-hashing it would would very badly without taking care of a certain aspect.
As one who also knows a thing or two about Zobrist hashing I'm curious
as to whether your "aspect" is indeed news to me or not. So here's my
not-all-that-secret Z-hash wisdom:
Antti Huima's trick of building mirroring, rotation, and color reversal
symmetries directly into a segmented Z-hash is neat, but his proposal is
flawed in that (as Antoine de Maricourt has shown) one can systematically
construct pairs of plausible Go positions that collide under Huima's scheme.
Fortunately there are alternative implementations that do not suffer this
defect, with as few as 6 (compared to Huima's 8) hash segments.
I have begun to write a paper on all this with Antoine but got bogged
down for too little time and too much perfectionism (wishing to be
mathematically precise about how such defects arise, how to prove that a
particular scheme is free of them, etc). If there's sufficient interest
I can try to bring what we have so far into presentable form and put that
up on a preprint server. All I ask is proper attribution of credit --
i.e., if you use it, cite it.
Now show me yours :-)
- nic
--
Dr. Nicol Schraudolph work: +61-2-612-51552
RSISE, bldg. 115, room B148 home: -51115
Australian National University fax: -58645
Canberra, ACT 0200
Australia http://n.schraudolph.org/
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/