Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Shirov strong tactical test position

Author: Chris Whittington

Date: 05:32:05 10/04/97

Go up one level in this thread



On October 03, 1997 at 19:33:06, Bruce Moreland wrote:

>
>On October 03, 1997 at 14:25:14, Chris Whittington wrote:
>
>>Was the 'sniff' word-concept a Bruce invention ?
>>
>>Because, if so, its such a neat idea, that we owe him at least some wine
>>in paris for inventing it.
>
>Ok, I'll claim to have invented this :-)  Finally I discover somthing
>new.
>
>>Incedentally, this idea of sniffing, I understand it because I see it,
>>but I wonder if it isn't more of a fast-eval-searcher concept. liek my
>>program tends not to sniff things, but to either find them of not. I
>>guess that since fast evals are getting many of their positional
>>concepts from the search, you'ld expect to see a gradual 'sniffing' as
>>the plies deepen; while a knowledge prg would just tend to grab it on
>>finding it .... ?
>
>Maybe.  I use the term to refer to cases where you find the answer to a
>tactical test for positional reasons, which could also be referred to as
>"tactical instinct" or perhaps, uncharitably, "luck", but perhaps you
>could also sniff a great positional move the same way.
>
>A sniff is when you find a Bxh2+ sacrifice but your score is +0.04.

No, no - the great sniff concept goes much further than that.

You find a move, but then the score drops each iteration, a little each
time, till you then fail low, and then find a substitute move.

The bad consequences only appear bit by bit, you haven't found them
fully, its something to do with search + nieve positional in the eval.

What must be happening is that (either finding a good move, or getting
to dislike a bad one) slowly, slowly bits of the tree that the PV was
getting its score from are being eliminated (bad move), or desired (good
move), but the tree keeps providing alternatives with slightly reduced
(bad move) or or increased (good move) eval regions.

A 'total' knowledge program, or oen with partial knowledge in the field
that counts, is going to find stuff on or off, but I guess still suffers
from sniff.

Chris

>
>bruce



This page took 0.26 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.