Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Tournament results (g/60, 2 comps, Fritz6, Century, Nimzo 7.32 and more)

Author: Ratko V Tomic

Date: 07:32:06 11/15/99

Go up one level in this thread


Thanks for a very insightful evaluation of these programs. I find these types of
tests (where a person, who has a good understanding of chess, plays the games
between programs and gets to think about the program's play move after move) and
the resulting descriptions more valuable than the output of some other mass
scale mechanized tests, devoid of common sense and human thought. There is a
greater, by several orders of magnitude, information content produced in a
single game than the 3 valued result win-draw-loss (i.e. 1.58 bits of
information per game) extracted from it by the mass scale mechanized test
method. After all, each ply delivers 5-7 bits of information, hence already in
each ply there is 3-4 times more information than the mindless tests will gather
from the entire game. And a 70-80 move game would generate around 1000 bits of
information, which is over 600 times more info than the 3 valued final result.
So a single humanly played and though through, move by move, game may extract as
much information about the program as several hundreds of mindlessly counted and
added up ones, one halfs and zeros.

> Another interesting thing about MChess is, that it's evaluation very
> often indicates the outcome of the game much earlier (often many, many
> moves) than the evaluations of the opponents.
...
> Unfortunately I have the feeling, that there is a small but severe bug
> in MChess's search function, because I have seen in two games, in which
> MChess was leading or at least equal, that it had a good PV-move with a
> good evaluation for a long time, then suddenly switched to another
> move, showing a worse evaluation than before, immediately played this
> move, and directly after playing this move discovered (while pondering)
> that this move was much worse than the move it had thought about before.

I have seen both of these phenomena, the positive and negative one, in Hiarcs
7.32. It is sad to see it sometimes waste away (sad, even though it may hand me
an easy point) an excelent or won position by some last moment switch into a
dubious line (which often it can't reproduce later in the replay). I hope it is
a bug (in MCP or Hiarcs) and not an inseparable side-effect of its algorithms. I
ordered MCP8 few months ago, but the store was out of stock. I'll probably try
it again, especially if the new version comes out, since the older MCP versions
(from perhaps 5 years ago, when I last updated it) were great fun to play
against (despite their crude UI at the time).




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.