Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Checks in the Qsearch

Author: Robert Hyatt

Date: 14:15:13 07/06/02

Go up one level in this thread


On July 06, 2002 at 11:32:12, Sune Fischer wrote:

>On July 06, 2002 at 10:25:44, Robert Hyatt wrote:
>
>>On July 06, 2002 at 01:28:07, Ricardo Gibert wrote:
>>
>>>
>>>When I read in CCC that the Deep Blue search had an EBF of about 4, my thoughts
>>>were, "Ugh! That means that in about 50 years of the expected hardware
>>>improvements, the PC programs of the day will be able to surpass Deep Blue even
>>>if Deep Blue were to get the commensurate hardware improvements."
>>
>>The problem is that you are making a _classic_ mistake.  The EBF has _nothing_
>>to do with how the two programs will compare.  What is important is the _tree_
>>that both search.  If one does a 20 ply search, and the other does a 10 ply
>>search, but they search the same tree, then they play equally tactically.
>>
>>Don't get hung up on a 20 ply search depth (iteration number).  It doesn't
>>mean a thing when compared to _other_ programs..
>
>Have you ever examined the tree, statisticly I mean?
>How many lines get pruned, how many extended, at what depths does this happen,
>at what depth do you the most hashtable hits, etc.?

For deep blue or Crafty?  I have done some analysis, although not in recent
years.  Others have also played with the idea of trying to graphically show
what the search is doing...  IE something to show where the nodes are most
numererous, or where the extensions are triggering the most, etc...



>
>I was thinking about picturing this in a histogram, so I would have an idea of
>the kind of distribution an X ply search makes.
>I guess it would take some bell shaped form, with few extended far and most
>ending around ply X.
>


I can think of several ways to show this.  But by "show" I mean "how I would
like it to look..."  But actually doing it on-the-fly seems challenging.



>Are there any papers on this?
>
>-S.



Not that I know of, but that doesn't mean there are not of course.



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.