Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: About random numbers and hashing

Author: Sven Reichard

Date: 11:35:57 12/06/01

Go up one level in this thread


Just to clarify one thing:

I didn't want to say that checking Hamming distances doesn't work. I was just
saying that I don't see any reason *why* it should work, and that the idea seems
counterintuitive to me. If it works in practice, fine.

The exposition above neglects a couple of things important for chess (like that
there can't be two pieces on the same square). This can make a lot of difference
in the analysis.

My suggestion for finding good hash codes:
- Generate a set of hashcode sets by random;
- Let each of them analyze a large set of games at a reasonable fixed depth
(e.g., d = 10);
- See which set visits the fewest nodes.

Of course this is quite time consuming. Oh, and while you're at it: Include a
set constructed by the Hamming criterion. Please report any significant
differences :)

As for the theoretical analysis, I'll check the literature as to how good a set
we can hope to construct, i.e., the largest n such that there is a set of 768
vectors in F2^64 no n of which are linearly dependent. (I have a couple of
references that might help.)

Bis die Tage,
Sven.



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.