Author: Ratko V Tomic
Date: 12:59:38 11/26/99
Go up one level in this thread
> I thought you might be interested in a post that appeared out here from
> a gentleman by the name of Larry Serflaten. He propsed an arithemtic
> coding procedure that I thought was kind of interesting.
Yes, I had followed that thread. That encoding had different objective (i.e. to
minimize the average size of some "typical" positions using some ad hoc
probabilities for pieces). The schemes KarinsDad and I were discussing were
about minimizing the worst case encoding, i.e. providing a coding which
guarantees that for any legal position it won't exceed certain number of bits
(originally 160 bits, later 157 then 155 bits).
The arithmetic code I suggested in this thread would replace only the section 2
of the original 155-bit scheme. It would also operate on a given piece count
state (not a "typical" one but an arbitrary one, as obtained from the Material
Content part of the full code). The remaining parts of that 155-bit algorithm
would still produce the fixed size code (the same length for any position) with
predictable worst case behavior.
Additional difference between this "arithmetic code" and the one you cited is
that the earlier one operates on fixed probabilites while this one operates with
given exact counts which change (get decremented) as the encoding proceeds. That
allows arithmetic code to perform at the almost entropy level ("almost" since
due to the 32-bit arithmetic it uses, it wastes on average 1/2 bit per 32-bit
word compared to the optimal coding in the earlier algorithm).
This page took 0.22 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.