Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: DEEP BLUES AVERAGE PLY?

Author: Robert Hyatt

Date: 08:02:11 08/22/02

Go up one level in this thread


On August 22, 2002 at 02:13:49, Uri Blass wrote:

>On August 21, 2002 at 22:32:03, Robert Hyatt wrote:
>
>>On August 21, 2002 at 19:14:17, Vincent Diepeveen wrote:
>>
>>>On August 21, 2002 at 17:52:53, Robert Hyatt wrote:
>>>
>>>>On August 21, 2002 at 17:31:53, Uri Blass wrote:
>>>>
>>>>>On August 21, 2002 at 17:21:08, Robert Hyatt wrote:
>>>>>
>>>>>>On August 21, 2002 at 14:48:04, Vincent Diepeveen wrote:
>>>>>>
>>>>>>>On August 21, 2002 at 14:42:49, Robert Hyatt wrote:
>>>>>>>
>>>>>>>Bob if you don't read what they write,
>>>>>>>then please show us you can do math.
>>>>>>>
>>>>>>>Please quote what is the theoretic number to search FULLWIDTH without
>>>>>>>hashtables OR killermoves and WITH singular extensions a treesize
>>>>>>>of 18 ply..
>>>>>>>
>>>>>>
>>>>>>First, they don't claim to do "fullwidth" in the hardware.
>>>>>
>>>>>The 12.2 is software+hardware depth.
>>>>>It is very clear from their paper.
>>>>>
>>>>>see page 13 table 2
>>>>>
>>>>>iteration 12
>>>>>minimum software depth 8
>>>>>
>>>>>The explanation say that is it about the position before white's move
>>>>>in game 2 against kasparov.
>>>>>
>>>>>Uri
>>>>
>>>>
>>>>I don't begin to know how to interpret those numbers in light of the email
>>>>I have received from the DB group about the 12(6) issue.  IE do you assume
>>>
>>>We talk about an email around the year 2000 (in 1999 you still said
>>>12 ply) from a dude called Campbell , not the programmer of the thing
>>>Hsu. He probably referred to 'average' search depth. They have a big
>>>table later in the paper *average* search depth.
>>
>>
>>First, it wasn't from Murray, it was from Andrew, a different member of
>>the team.  You obviously don't know any of them so I won't go farther
>>there.  Second, if you recall the email, Andrew _specifically_ said he
>>talked to CB (Crazy Bird, AKA Hsu) to verify that it had not been changed
>>in meaning.
>>
>>What more can I say?
>>
>>
>>
>>
>>>
>>>Suggesting that deep blue 2 which was only slightly faster than deep blue 1
>>>(no more than a factor 2) getting suddenly 6 plies extra is not possible.
>>
>>I don't know what you mean.  I personally watched deep thought search 10-11
>>plies deep in the middlegame in 1989-1994 games at ACM and WCCC events.  I
>>_saw_ that.  Deep Though was credited with roughly 1-2M nodes per second
>>by Hsu and team.  DB was clearly a hundred times faster.  Which should
>>certainly produce 5 more plies at their 4.0 branching factor...
>
>4^5=1024 4^4=256
>
>so even with branching factor of 4 and being 100 times faster
>they can see only less than 4 more plies.
>
>If you consider the fact that the real branching factor was more than
>4 and they were less than 100 times faster(the 200M nodes by IBM
>was misleading and the effective speed was clearly smaller)
>then you get less than 3 plies.
>
>If you consider the fact that deeper blue di more extensions than
>deep thought then you can get less than 2 plies.
>
>Uri


I don't see anything that says that DB did more extensions than Deep Thought.
In fact the paper you have says exactly the opposite, that late deep thought
versions used the _identical_ search algorithm that ended up in deep blue.



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.