[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [computer-go] Re: Sharing Secrets



You seem  to be claiming that  with truly random  numbers zobrist keys
are poorly distributed.  I don't believe this.
No.

With perfect random keys, the distribution is very, very, very good.
Extremely very good.

BUT the distribution with optimum Hamming distance is much better.
Orders of magnitude better, thousands of times better.

Which is is very relevant. Say you have collissions once per million or once per tenthousand. Big difference when looking at ten million nodes.

I call "thousand times worse" "piss-poor" compared to optimum Hamming distance.

I can't prove that it's 1000, but the chess cracks tell me it's much more than that.

My point was not really that optimum Hamming is better than random, my point was that using the maximum Hamming distance is a very bad idea. Better is to be a few notches lower.

I can't prove it but my experiments showed that absolute max. Hamming distance was an absolute disaster. Collisions all over the place.Perhaps because they have lost all their randommness, you are left with a small set of numbers that comply to a very strictly defined requirement.. I'm not a mathemagician.
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/