Author: Ratko V Tomic
Date: 12:13:10 08/13/99
Go up one level in this thread
Thanks for the two links, the sites look interesting
and informative.
> Who wants the program to instantly repeat all the
> moves of a game it lost--how boring!
Even if it wins, it could have won later on tactics or
in endgame, even though in the opening it made a mistake.
It ought to go over the game even if it won, as long as
it has a large enough drop in its score at some point.
If a simple extension of search 2-3 plies (compared to
saved depth) doesn't show anything better, it would have
to back off further and simulate a longer play (against
itself with depth at least as large as the saved one),
but reusing the opponents moves as a skeleton (it would
try them out in the simulation if possible, e.g. it could
give them artificially higher positional score, so that
with all other things being roughly equal, it would
pick them out; or it could advance them on the move list
any other way). That way it could reach much larger depth
centered around the line of interest.
By gradually stepping back in this manner, it would
eventually find a place where its analysis picks
a different move than the one it picked in the game.
Once an alternative is found, it could auto-play one
or more game fragments with that move, to uncover any
potential downside beyond the regular horizon.
> Chenardonly stores positions up through move 15 of
> the opening.
Yes, I noticed. It shouldn't have such superficial limits.
Even in endgames, the positions can recur, especially when
playing the same opponent.
> The only chess program I know with offline
> learning is Hossa by Steffen Jakob:
That one looks interesting. He apparently hasn't quite
yet resolved all the issues of the learning technique,
but it certainly looks on the right track.
> Offline opening book learning is much more powerful
> than online learning. I'm rooting for this feature too.
And it saves great deal of user's time. Having only short
bits of time to play, it annoys me to sit and wait for a
program to spend minutes calculating only to come up with
the same move again. Entering manually a book, not only eats
up its own time (and it is quite tedious), but it is a
different process all together:
1. There is no element of surprise and associated
over-the-board problem solving (since I had input the line).
And that is the fun part of chess in the first place.
2. I would have to know from somewhere what the best line is
(another chunk of time spent trying to find out), and that
may not be the case, especially since these situations
would occur outside of the regular opening book (or even
in endgames).
3. The process lacks spontaneity, it loses the charm of
playing with a friendly foe. Thus the battle of minds
component from a human play is lost. With manually
entered opening lines, instead of battle you get a movie
battle with choreographed moves, not quite the same thing.
4. It is much too deterministic and mechanical -- all variety
comes from putting it into it in advance, thus it is not
a real variety. It's like a varity of plastic flowers, yes,
they vary but only as much as the designer chose to do so.
It is unfortunate that the commercial programmers have become
so concerned with beating competitors' programs and advancing
on SSDF, they forgot that it is at the end the users' enjoyment
of playing what keeps such products selling. Playing a life-like
opponent is an important part of that enjoyment. And the
"life-like" which really counts in chess isn't in the multimedia
gimmicks, or in any other superficial aspect, but in that which
makes chess an enjoyable game, the creative thinking, problem
solving and the battle of minds. It is in the abstract world
of thought not the world of senses. I guess that's too much to
ask from marketing gurus who noramally hype kids toys and
sports gear.
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.