Author: Jan Pernicka
Date: 04:36:13 03/14/00
Go up one level in this thread
On March 13, 2000 at 09:27:45, Andrew Williams wrote:
>On March 13, 2000 at 08:20:11, Jan Pernicka wrote:
>
>>Hi,
>> I would like to know if you have (as a programmers)
>>some experience with search strategy MTD(f) (I will be able to write a brief
>>descrition of this strategy soon - if my bus is not coming - as now!!!!!)
>>
>
>Hi Jan,
>
>My program, PostModernist, uses MTD(f). There are at least two others
>I know of that use (or used to use) MTD(f): AnMon and Cilian.
>
>There's a little bit of information about my program at:
>
>http://www.doc.mmu.ac.uk/STAFF/A.Williams/postmodernist/postmodernist.html
>
>On that page there's a link to a page by Aske Plaat who explains MTD(f)
>in detail and gives some pseudocode.
>
Hi and thanks, but:
what is your experience with MTD(f) - is it much better(20%...)- in speed -
than "standard" techniques (as aspiration window,negascout...) and
how large hash tables should be used (especially interesting is
the ratio: #entries in hash table / #nodes (not only leaves) visited ).
Note: Isn't there a problem when your evaluation / selective
techniques depend on alfa-beta params? As in this case alfa-beta are
not "estimates" but only artificial parametres to speed up a search...
Well, this problem could also be noticed (or not?) in the techniques I
mentioned before (i.e. negascout....).
>
>> My second question is: How much faster is the incremental evaluation
>> vs. static evalutaion (i.e. in leaves) - if any.
>>Thanks,
>
>I'm not sure I understand this question.
>
>Andrew
So - to say it more precisely (I have more time today :-) ):
There are 2 possible ways how to evaluate your possition (and, of course,
their combinations):
1) Given the possition, calculate completely its score "from the scratch"
(I called it static eval...), i.e: scan through all the board and
asses the pieces you are going over.... OR:
2) Evaluate the starting position and after every move update the
score of it so you get score of the new position. This update
typically takes less time because it involve only small part
of the board (the piece it has moved + its neigbourhood...).
I wrote about this as incremental evaluation (dynamic eval. could be
also used... ).
Jan.
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.