Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: [DB] Some data from the logfiles

Author: Robert Hyatt

Date: 20:31:57 08/27/02

Go up one level in this thread


On August 27, 2002 at 20:36:30, Dave Gomboc wrote:

>On August 27, 2002 at 06:14:03, Uri Blass wrote:
>
>>On August 27, 2002 at 04:34:16, Dave Gomboc wrote:
>>
>>>On August 25, 2002 at 23:39:29, Robert Hyatt wrote:
>>>
>>>>On August 25, 2002 at 21:56:03, Dave Gomboc wrote:
>>>>
>>>>>On August 22, 2002 at 16:09:07, Uri Blass wrote:
>>>>>
>>>>>>On August 22, 2002 at 15:51:28, Jeremiah Penery wrote:
>>>>>>
>>>>>>>On August 22, 2002 at 06:47:34, Gian-Carlo Pascutto wrote:
>>>>>>>
>>>>>>>>That does not make sense - it only does when you take the first number as
>>>>>>>>the nominal ply depth and the second number as the part of that that was
>>>>>>>>done by the hardware searches.
>>>>>>>
>>>>>>>
>>>>>>>So what does it mean when you have searches like this,
>>>>>>>
>>>>>>>-->  17.   Be3 <-- 23/113:12
>>>>>>>---------------------------------------
>>>>>>>Guessing Qc7
>>>>>>> 3(4) 25  T=0
>>>>>>>qd1d2 Pc5c4 pb3c4P
>>>>>>> 4(5) 25  T=0
>>>>>>>qd1d2 Pc5c4 pb3c4P
>>>>>>> 5(5)[Qd2](25) 25  T=1
>>>>>>>qd1d2 Pc5c4 pb3c4P
>>>>>>> 6(5)[Qd2](25) 25  T=2
>>>>>>>qd1d2 Pc5c4 pb3c4P Qc7c4p
>>>>>>> 7(5) #[Qd2](28)##################################### 28  T=4
>>>>>>>qd1d2 Re8b8 nf3e5P Pd6e5n
>>>>>>> 8(6) #[Qd2](28)##################################### 28  T=12
>>>>>>>qd1d2 Re8b8 bc2d3 Pa6a5 pc3c4
>>>>>>> 9(6)<ch> 'ng6'
>>>>>>>---------------------------------------
>>>>>>>--> Ne7g6 <--
>>>>>>>---------------------------------------
>>>>>>> 28  T=19
>>>>>>>qd1d2
>>>>>>> 3(4)[Qd2](30) 30^ T=1
>>>>>>>qd1d2 Pc5c4 pb3c4P Pb5c4p
>>>>>>> 3(5) 35  T=1
>>>>>>>qd1d2 Qd8c7 pb3b4 Pc5c4 be3h6P
>>>>>>> 4(5) 35  T=1
>>>>>>>qd1d2 Pa6a5 pa2a3
>>>>>>>
>>>>>>>
>>>>>>>where you have depths like 3(4)?  They can't have 3 nominal plies, where 4 of
>>>>>>>those plies come from the hardware, because obviously that's impossible.
>>>>>>
>>>>>>A good question.
>>>>>>
>>>>>>I do not understand the meaning of the second mnumber
>>>>>>but the first number is clearly the brute force depth based on their paper.
>>>>>>
>>>>>>Maybe the second number is about some limit about the extensions but OI do not
>>>>>>know.
>>>>>>
>>>>>>Uri
>>>>>
>>>>>Uh, is that what you guys are all discussing _again_?
>>>>>
>>>>>Sheesh.
>>>>>
>>>>>The first number is the depth of the software search.  The second number is the
>>>>>depth of the hardware search.  I posted this _years_ ago after asking a member
>>>>>of the DB team directly: check the archives.
>>>>>
>>>>>Dave
>>>>
>>>>That is what I was told also.  However, a fairly new paper really clouds the
>>>>issue in that they mix depths between DB2 in the 1997 match, DB Jr on slower
>>>>hardware, etc...
>>>>
>>>>I think that the only explanation for the (x) number is the one given by the
>>>>team to me.  And apparently to you as well, and probably others that simply
>>>>don't post here...
>>>
>>>Often when they refer to their search tree they refer to the software depth
>>>only.  Which paper is causing the kerfuffle?
>>>
>>>Dave
>>
>>A paper by Murray Campbell,Joseph Hoane Jr and Feng-hsiung Hsu (august 1 2001)
>>
>>In that paper they said the following in page 5:
>>
>>"A three minute search on deep blue would reach a full width depth of 12.2 on
>>average."
>>
>>"The estimate is based on a linear least squares fit on all the iteration,log
>>time data points from the 1997 match against kasparov."
>
>I suppose you could take the information from the logs, compute this, and see if
>12.2 is the average software search depth or the average total (software +
>hardware) search depth.
>
>>They also say in page 13 the following about deep blue Jr:
>>"For a given iteration i,the software is assigned i-4 ply which represent the
>>minimum depth search in software."
>>
>>They never said that iteration means different things in deep blue Jr and deep
>>blue so it is logical to assume that if iteration is software+hardware in deep
>>blue junior it is also the case in deep blue.
>
>All the same, I think what they just meant was that in Deep Blue Jr., the
>hardware was set to search four ply, independent of the (software) search depth.
> I don't believe they're using "iteration" here to mean exactly the same thing
>as when they used "full width depth" in the other quote above.
>
>>I have some questions for them that I did not understand from the paper:
>>I will be happy if they answer only by yes/no when they can answer the last
>>question by a number.
>>
>>1)Does iteration mean the minimal depth that they could not miss tactics(In
>>other words in the worst case they could miss tactical line of 13 plies when
>>they searched iteration 12)?
>>
>>2)Does iteration mean the software search in deep blue II?
>
>From the one quote, it sounds like iteration means software + hardware.  But in
>another context they may use iteration differently?
>
>>3)Did they use only selective search in the hardware(they say in comment 3 in
>>page 5 that their experiment showed that deep blue sacrificed 2 plies
>>of full width searrch in order to execute the selective search algorithm)?
>
>Their software search includes selectivity (extensions!).  ASAICR, the hardware
>chips can't do singular/dual/etc. extensions.
>
>Dave


Actually, based on this recent paper, the DB2 chip apparently had SE added
when it was re-designed...  I was not aware of it specifically, but then I had
not asked either since the original deep thought stuff did not do SE in the
hardware.

whether or not it did the double-SE and whatever is not clear at all from the
paper, however...  just that the software was doing that was all that was
clearly stated.



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.