Computer Chess Club Archives




Subject: Re: Position Evaluation

Author: Dave Gomboc

Date: 05:31:42 11/23/99

Go up one level in this thread

On November 22, 1999 at 23:04:18, Ratko V Tomic wrote:

>>I think that including initiative as an evaluation term is sufficient to
>> provide this behavior.  The initiative term need not be too low.  A rule
>> of thumb I learned once is that "90% of the time, the initiative is worth
>> a pawn."
>> Obviously, it would take some work to translate this into an eval function.
>> Simply testing for initiative and adding 1 to a score would produce horrible
>> results.  But the main idea is clear enough.
>Easier said than done. Initiative, like beauty or elegance, is somewhat
>intangible. It is a quality, not a quantity like material or mobility. Giving a
>check with Bb5 or attacking a knight on g5 with h6 doesn't an initiative make,
>even though initiative does often involve pushing the opponent's pieces back.

I think that initiative can be successfully quantified.  It is easy enough for a
human to look at a position, consider it for a bit, and decide "white has a
clear initiative", "there is a struggle for the initiative", "this position is
dead", et cetera.  Obviously, it would take some work, but all the same it
doesn't seem like a terribly difficult problem to come up with something
reasonable, and tune it as experience is acquired.  As for how to collect this
information, it could be done while the tree is being traversed, or computed at
the tips: programmer's choice.  Some people might even ignore it as an eval term
per se, but shape their tree in recognition of the importance of initiative.
That could work too, if you get enough depth where it's important, I suppose.

Not only do I think it can be done, I also think that it's important that it be
done.  Granted, it may be argued that I enjoyed Dynamic Chess Strategy (Mihai
Suba) too much!

>Additionally, initiative isn't free, one often has to unbalance the position,
>even weaken some aspect of ones own position or give material. For something
>that is hard to quantify (if possible at all)? Would a program playing without a
>book, and with your hypothetical initiative term, come up with Evans gambit or
>Kings gambit? (I know, some programs have hand-crafted instances they recognize
>in which they will give an exchange or a piece for 1 or 2 pawns to get a strong
>attack without seeing an explicit compensation within the horizon. But those
>things, while working sometimes, more often hurt them than help. I think Hiarcs
>7.32 has some of such hand-crafted logic and what some people here (including
>myself) have noticed as occasional flakeness in Hiarcs' play, may be due to the
>backfiring of this kind of evaluation override.)

Nothing is free in chess.  Pawn structure isn't free, either, yet it is
routinely exchanged for better piece placement, or vice-versa.  The exchanging
of one advantage for another is something that minimax can cope with very well.

>>Quantifying the amount of initiative one or the other opponent has
>>The frequent repetition of "static evaluation" ignores that dynamic terms can
>Did something get mangled and chopped off at the end of your message? Of the
>little that came through, it sounds interesting. Could you post the rest of it?

Hmm, weird!  Maybe I started writing more and forgot to finish? :)

I have too much work to get done, I've got to get back to that.  I think what I
wanted to say was that the dynamism of a position is important, and it can make
sense to include dynamic terms when assessing what direction of play is best.


This page took 0.01 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

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