Re: [computer-go] Re: Sharing Secrets

["Frank de Groot" <frank@xxxxxxxxxxxxx>]

From: "John Tromp" <John.Tromp@xxxxxxxxxxxxxxxxx>

> uses 64 bit Zobrist hashing to detect superko. In fact, I store Zobrist 
> hashes
> of all preceding positions in a Java HashSet.
>
> Is there supposed to be an issue with this setup?
> Or does the issue only come up with search and transposition tables and 
> the like?
No, that will work just fine for an editor.
The "issue" is about hash collision avoidance.
When you start thinking ahead (or when you have a huge pattern database),
you will need millions of hashes and you'll get collisions even with 64-bit
hashing if you have not "taken care of the issue :)".

I have never heard of a hashing class in any programming language, neither
of any literary source that satisfactorily solves this "issue" best. The
very rare sources that hint at a really good solution are incomplete and in
fact even lead astray.

I think such "secrets" are interesting because they lay at the foundation of
everything. Without a good foundation, even the best ideas simply don't
perform at their best.

Other major secrets IMO are ways to speed things up. My ladder code takes a
few milliseconds for ladders that crisscross the board and it is
un-optimized and find even a bit more than pure ladders. A fast ladderer can
find ladder breakers in less than a second for all ladders on the board
worst-case, simply by trying all legal moves and recalculating all ladders.

If someone has found a way to speed up a ladderer in a genius way (for
example that a progressing position delta-updates ladder status, something
that appears impossible to me now but I have a feeling it *might* be
possible..), this would be an A-superplus secret I would be very interested
to hear about.

_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/