Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Alpha/Beta adjustments via the hashtable or why Fritz 7 has messy PV's

Author: Gian-Carlo Pascutto

Date: 02:57:07 12/03/01

Go up one level in this thread


On December 03, 2001 at 05:28:39, Uri Blass wrote:

>On December 03, 2001 at 04:34:03, Gian-Carlo Pascutto wrote:
>
>>On December 03, 2001 at 03:55:56, Uri Blass wrote:
>>
>>>I do not understand it and I will be happy to see an example of a tree
>>>that leads to pruning the main line.
>>
>>It should happen in any branch where there was some form
>>of forward pruning leading to a lower bound on the score,
>>and where the branch was researched later on.
>>
>>There are probably other cases, but this is the obvious
>>example.
>
>I still prefer to see an example to understand and it means a chess position
>and the nodes that the program searches before showing the wrong line.

starting position (all score from white point of view)

before starting the search the program assumes the score will be within -0.10
<-> +0.10 (the aspiration window)

if the score is outside this window the program will fail low or high.

we start searching 1.e4 and the score for every reply is at least 0.10 for white

we store 'at least 0.10' in the hashtable and return

the score was >= 0.10 so we fail high

we research
----------------------------------------------------------------------

normal case:

we now search with +0.09 <-> +inf

we search all moves after 1.e4 and find a reply 1. .. e5 with score exactly
0.10

there are no better replies and we return (0.10 e4 e5)

----------------------------------------------------------------------

hash/bounds adjusted case:

we now search with +0.09 <-> +inf

we get to 1.e4 and find this position in the hashtable. the hashtable
says: at least +0.10

we change our bounds to +0.10 <-> +inf

we do not find a single move that is better than +0.10

we return (0.10 e4)

-----------------------------------------------------------------------

We now search the remainder of the white moves and find nothing better
than 0.10. While doing so, we overwrite the >=0.10 entry, so we will
never know what move produced this score.

This is perhaps a bit farfetched but I wanted to keep forward
pruning out of the example so as not to make it too complicated.

--
GCP



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.