Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: For Bob Hyatt(or other): a question about deeper blue logfiles

Author: Robert Hyatt

Date: 07:39:53 07/18/00

Go up one level in this thread


On July 18, 2000 at 08:10:03, blass uri wrote:

>Here is part of the logfile for game 4
>
>---------------------------------------
>-->  45.   Rb4 <-- 15/70:34
>---------------------------------------
>hash guess Rc2a2,Guessing Ra2
>...
>...
>14(7) #[Kb1](-32) -32  T=610
>ka1b1 Ra2a3p ra7a3R Kc5b4r ra3h3 Rf2e2 ph4h5 Re2e5p ph5h6 Re5e8 ph6h7 Re8h8
>kb1c2 Kb4c4 rh3h4 Pd5d4 rh4h5 Rh8h7p
>
>I do not understand the last Rh8h7
>

This is not uncommon.  They can't get their PV the way most of us are doing
it, namely by backing up the PV along with the score.  Since the hardware chips
do searching, they would have to be capable of passing back dozens of words of
data containing the PV, in one cycle. This is impossible.  As a result, they
grab moves from hash tables, and that leads to strange output at times.

I generally try to ignore their PV moves.  I tried this same thing when I was
playing with mtd(f) and saw some really strange moves sneak into the PV thru the
hash table.  It has no effect on the program, because the program really didn't
consider that move best.  It was just left in the hash table somehow, or was
entered, when it didn't make any sense (ie a best move at a fail-low node is
meaningless and often is garbage).





>The position on the board:
>[D]8/R7/2p5/2kpP3/1R5P/P7/2r2r2/K7 b - - 0 1
>
>Deeper blue expected 45...Ra2 and at depth of 21 plies brute force calculated
>the main line:
>
>46.Kb1 Rxa3 57.Rxa3 Kxb4 58.Rh3 Re2 59.h5 Rxe5 50.h6 Re8 51.h7 Rh8 52.Kc2 Kc4
>53.Rh4+ d4 54.Rh5 Rxh7
>
>I simply do not understand the idea behind the last move
>
>The move 55.Rxh7 is only 19 plies deep with no extensions and if you assume
>14+7=21 plies with no extensions you can see that it is wrong.
>
>Tablebases can tell that black is mated after 55.Rxh7 and even without
>tablebases material count can tell easily that it is wrong.
>
>Uri


This is a peril of getting your PV from the hash table.  Which is a peril of
doing special-purpose hardware where you can't back the PV up with the score as
most of us do...

The only thing that is important is the score and move at the root, because this
is always right since it doesn't come from the hash table.



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.