Re: handicapping engines

Major Cat wrote: 
Johnny T wrote:

I suppose I was talking to the programmers, and taking the opportunity
to try and provide more clarity to the "handicapping" discussion.


That is what I thought. Thank you for the clarification.

There appear to have been 3 approaches to the handicapping problem.
1. Reducing the horizon.  Essentially anything to reduce the number of
positions that are seen by the computer in a certain time. (like messing
with search depth, or turning the permanent brain on and off.  This
leads to classic computer errors.  This will effect some engines more
than others.


What about restricting the program's time control? Unlike tinkering with
search depth and permanent brain, time control calibration is subject to
way less granularity. Moreover, the ensuing "blindness" is less uniform
(from move to move).

Unfortunately you get bad computer chess artefacts using fixed time control or fixed depth. Event horizon problems. During the opening book the computer can move instantly whilst inside the book, but in the endgame to avoid playing like a complete patzer it needs way more look ahead than a fast time control will ever allow.

Also, why are "classic computer errors" so different
from those, say, one comes across in games between class players?

In part because humans use pattern matching tricks to spot things that seem obvious to us but are very hard to simulate in software.

Finally, is engine code "out there" _that_ differentiated? 

What do you mean?

2. Equalization of material.  This is sort of an advanced version of
"odds".  This is what friend mode will attempt to do.  It will try and
get to an "equal" position by dropping material to you.  You can get the
computer to play REALLY poorly if you just sac a pawn or two.


This _simulation_ of human play may be fine for parlor/cafe chess. For
example, not being creatively forgiving of the blunders of a child may
be bad cultural form. Not to talk about a woman one wants to...bed! 8>)

Never really played friend mode. I find the lower level settings of Fritz 8 are usable somewhere in the range from 1640 (can hammer it if I concentrate properly) to 1900 (hardly ever win) on my 3Ghz Pentium. I often lose on time so setting assymetric time limits helps me a lot.

3. Unbalanced chess knowledge.  This is what Chessmaster essentially
does with it's "personalities" and other programs attempt with the
aggressive/passive settings.


Assuming, of course, that the "golden section profile" represented by
the default settings for strongest play is reflective of an _adequate_
simulation of some _user_desirable_ type of human play...

I found the Chessmaster user interface so incredibly ugly that I never really got into playing it. If there is any way to run it under the Chessbase interface I'd be a lot more interested. 

However, all of these seem to fail in either being unrealistically
difficult, or unrealistically easy.

Making it seem just right and like a human is technically very hard. The best I can imagine would be to use a very strong engine and allow a small probability that it sometimes misses certain parts of the game tree. Adding a few centipawns of coherent random noise to the terminal nodes might help. And then deepening the search progressively in the endgame stages. Human play is tricky to simulate well. 

Is there a consensus about this? Ok, posters often complain about the
FRITZ programs in regards to their presumed inability to generate truly
"intelligent mistakes".  CHESS_MASTER users seem to be way more pleased
with their personalities!

One thing that would improve simulations of more human-like weaker players would be to pay a bit more attention to the major pieces so that the machines mistakes are not such obvious blunders.

Regards,
Martin Brown

.

Relevant Pages

  • Re: 0xC00D1197: Cannot play the file
    ... Are you trying to play an Audio CD? ... -What do your Network settings look like in WMP? ... > The proxy server settings for the Player are not ...
    (microsoft.public.windowsmedia.player)
  • Re: MP3s - Some play - Some Dont
    ... "Sonia" wrote: ... > comparing the Registry settings, not the Win.ini and System.ini files discussed ... Then play the ... >> My recently ripped files play within PPT but do not produce sound. ...
    (microsoft.public.powerpoint)
  • Re: MP3s - Some play - Some Dont
    ... If your files play in mplay32.exe and not in PowerPoint when in slide show mode, ... everything to C:\Test and delete and reinsert the music files. ... > My recently ripped files play within PPT but do not produce sound. ... > There is no doubt that using RP has thrown some settings out of whack. ...
    (microsoft.public.powerpoint)
  • Re: Inserting sound files
    ... Try this: Right click the animation. ... > Animation settings: START- After Previous, ... The PLAY button at the bottom of the Custom Animation box, ... plays for only one second, then stops, no sound. ...
    (microsoft.public.powerpoint)
  • Re: I dont understand in part of western conference playoff spots...
    ... If they expanded 8x4 would work, but that looks very unlikely anytime on the horizon. ... Then every team could play 2 division games last of the season, ...
    (alt.sports.basketball.nba.la-lakers)