Re: [computer-go] Re: Sharing Secrets

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

> You didn't challenge my claim that custom 60 bits is no better than random 
> 64 bits.
I can't give hard numbers.
I have no idea, frankly.
What I have done is asked some top-level chess programmers.
They said what I just said (do not use random, ever).

> But people have already done this work for you. Think libraries.
> No one has to code a PRNG from scratch... as opposed to your custom 
> values.
I have replaced Delphi's out-of-the-box random (that even neglects entire 
bits..) with such a "library" random. But such "library" randoms that you 
are so happy to trust, usually use some form of long shift register, and a 
scheme of XORing part of that register into another position into itself 
(and a bit more more than XOR of course).

Some of our good chess programming friends have done a wealth of research in 
this area and they found that the best random generators give good random 
when drawing dots on the screen or prediction the stock market, but when you 
use their "random" values to xor with other "random" values they produced, 
you are asking for trouble, because a main ingredient of most random 
generators is XORing..

You could try letting a random generator run for a while (time-dependent 
with few % margin of error, like using a Windows timer) and not take the 
next values.

Depending on the kind of random generator, I expect that you end up with 
only 60% of the effective bits, sometimes much less, due to the "XORing" 
nature of many random generators.


> I guess I'm gonna join Erik in asking you to quantify the benefit of 
> custom values...
Better idea: Tell Erik his Hamming-optimization was not clever and ask him 
why he did it.. Tell him that he should have used "random". He will tell you 
exactly the same as I did, that "random" is not reliable and optimizing 
Hamming distance is the way to go.

I found that throwing computer power at the problem of MAXIMIZING Hamming 
distance was a mistake however. Optimum <> Maximum.

And that is non-intuitive, not even for the people who acknowledge that 
Hamming distance is a main consideration of Zobrist keys. 
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/