Re: [gnugo-devel] arend_1_32.4: hash_(re_)init for new game (revision of gunnar_1_32.8)

[Heikki Levanto]

On Fri, Apr 12, 2002 at 12:56:12PM +0200, Arend Bayer wrote:
> > Maybe we should simply use a fixed seed for the Zobrist numbers,
> > instead.
> 
> Hmm I don't like this idea, but maybe I am not very rational about it.
> I wonder whether there is a way to avoid a "bad seed"? Say if there is a
> 1% of choosing a bad seed (i.e. making hash collisions more likely), I'd
> rather have a bad seed in 1% of the games than have a 1% chance of having
> a bad seed in all games.

I do like the fixed seed. 

Here is one way to check if a seed is bad: Generate a great number of moves
with different hash tables, and see if they produce the same moves. Even
better, verify that the top-10 moves are the same, with same values. If this
test works for 9 random seeds, and fails for one, we can assume (guess) that
the one seed is bad, and managed to produce hash collisions.

I guess one could also make some statistical calculations on the hamming
distances between all the hash values, and various combinations thereof. I
do not know how...

- Heikki

-- 
Heikki Levanto  LSD - Levanto Software Development   <address@hidden>