Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Evaluation-based Reductions and/or Extensions

Author: Tom Likens

Date: 16:24:17 12/28/03

Go up one level in this thread


On December 28, 2003 at 17:34:27, Uri Blass wrote:

>On December 28, 2003 at 13:32:05, Tom Likens wrote:
>
>>Hello Everyone,
>>
>>I've been experimenting recently with using the evaluation function to shape
>>the search tree.  Specifically, I've been using the static evaluation of
>>the current position and the previous position to determine if a move should
>>be extended or reduced.  I also have been making allowances for moves that
>>increase or decrease the pressure against the king, attack hung pieces,
>>save hung pieces etc.
>>
>>So far the results have been exciting, but also potentially frustrating.
>>The main problem I've encountered is that any pruning or extensions based on
>>the previous node's score cause hashing problems because this becomes path
>>dependent. In a way, I suppose this isn't much different then making these
>>type of decisions based on the value of alpha or beta as well, but these new
>>effects have (at least for my program) seemed more detrimental.
>>
>>My (obvious) question, how do other programmers deal with this phenomenon?
>>I suppose ignoring it is one option, but I'm hoping there is a better
>>solution.
>>
>>regards,
>>--tom
>
>I simply use hash tables only for order of moves.
>I believe that the potential gain from using the evaluation function to shape
>the search tree is simply too big to miss it.
>
>I do not believe that I am close to using the evaluation in the optimal way to
>shape the search tree and I do not use part of your ideas(I do not evaluate
>today attacking hung pieces or saving hung pieces and I believe that it is one
>of the things that should help me if I get the energy to write the relevant
>code).
>
>Uri

Hey Uri,

It's interesting, email discussions with Tord started me thinking about new
ways to use the evaluation function for extending and pruning the tree.  After
I started looking at my evaluation function it quickly became apparent that I
had a number of these terms already quantified but wasn't taking full advantage
of them in other parts of the program.

I think one of the reasons my program isn't stronger is that it is getting
outsearched.  My next round of improvements (of which this is the start) will
concentrate on new ways to reduce and prune useless branches.  It's a fine
line though, too aggressive and the baby goes out with the bath water.  Not
aggressive enough and the "improvements" are less than worthless.

regards,
--tom



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.