Re: [computer-go] Re: Sharing Secrets

[John Tromp <John.Tromp@xxxxxx>]

Frank de Groot wrote:
> > I've never seen any evidence or discussion of this. I certainly cannot
> > imagine that specially chosen 60-bit numbers are better than randomly
> > chosen 64-bit values. So I see no basis for your claim above.
> I have never said that hand-picked 60-bit values outperform 64-bit 
> random values.
No, you said:
>So, herewith I have basically disclosed part of the "secret": DO NOT, UNDER 
>ANY CIRCUMSTANCES, USE RANDOM VALUES AS HASH KEYS. Never ever. Not even when 
>you use white noise from a resistor or diode. It simply gives sub-par 
>performance.

So you made it sounds like random values are MUCH worse than some custom 
values, something to be avoided at all cost.

> I said that hand-picked 64-bit values outperform 64-bit random values.
> And already 2 people confirmed this just now.
> What's so difficult to understand about that, and do you think Erik & 
> Arthur are doing nonsensical things when they bother with the Hamming 
> distance of their Hash keys?
And now you say that some complicated method of picking custom values
might be equivalent to 1 or 2 extra random bits.
For me, this lends strong support to the advise to use true random
or excellent pseudo random values, since many attempts at creating
custom values (ignoring all the software engineering headaches)
will inadvertantly lead to a *worse* collision probability, maximizing
hamming distance being a good example.

Why go through all that effort when random values, which are trivial to
code, are so close to optimal???

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