Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Extended futility pruning and hashtables

Author: Vincent Diepeveen

Date: 03:43:01 01/02/00

Go up one level in this thread


On January 01, 2000 at 10:56:56, Gian-Carlo Pascutto wrote:

>On December 31, 1999 at 11:53:04, Vincent Diepeveen wrote:
>
>[...]
>>>b) I understand that just storing the values into the transposition table is
>>>deadly (e.g. having to research after a fail-low would yield garbage), and
>>>this is mentioned in the paper, but how should this optimally be handled then?
>>>Just not storing anything in the ttable seems rather radical.
>>
>>Indeed. The more efficient your search is (so the smaller your fliprate
>>which i defined as the chance that a node which was stored as < beta
>>in transpositiontable now becomes >= beta), the less likely such
>>dubious things as futitily pruning will work for you.
>
>I must admit I can't immediately see the link between the two ?

Make a minimax searcher without alfabeta which is highly inefficient.
Now any change will be considered an improvement.

Make a simple program, now most changes will be considered an improvement.

Make a program that searches very efficient,
now hardly anything will improve its search.

Darkthought is from the 'schach' times. Compared to other academic
research programs it's incredible well. See how well it usual performs
in the world champs.

However i feel that its evaluation and quiescencesearch are very simple
compared to many programs that are profitting bigtime from faster hardware.

I feel we must distinguish this discussion from the testmethod.

With a few search changes i can also solve hundreds of more positions at
the ECM testset with DIEP when running at 20 seconds a position.

Does my program improve solving more positions when running at 20
seconds a move at ECM, without further info?

So i'm talking here about 2 different points
  - testmethod
  - efficiency of search

>But yes, my fliprate (never heard this before, but sounds like something
>interesting to measure nonetheless) is probably quite high and that's
>not due to my search but rather caused by the nature of the game I'm
>working with.
>--
>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.