Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: CM8000 - an analysis anomaly

Author: John Merlino

Date: 18:00:17 12/07/00

Go up one level in this thread


On December 06, 2000 at 19:25:32, Uri Blass wrote:

>On December 06, 2000 at 15:57:00, John Merlino wrote:
>
>>On December 04, 2000 at 23:27:21, William Penn wrote:
>>
>>>CM8000 - an analysis anomaly
>>>
>>>This is an analysis after 1.Nf3 Nf6 2.c4 c6 3.Nc3 d5 4.e3 e6 5.d4 Nbd7 6.Qc2 Bd6
>>>7.Be2 0-0 8.0-0 e5 9.cxd5 cxd5 10.Nb5 Bb8 11.dxe5 Nxe5 12.Bd2 Bg4.  The position
>>>after 12...Bg4 is as follows:
>>>
>>>BR  BB  --  BQ  --  BR  BK  --
>>>BP  BP  --  --  --  BP  BP  BP
>>>--  --  --  --  --  BN  --  --
>>>--  WN  --  BP  BN  --  --  --
>>>--  --  --  --  --  --  BB  --
>>>--  --  --  --  WP  WN  --  --
>>>WP  WP  WQ  WB  WB  WP  WP  WP
>>>WR  --  --  --  --  WR  WK  --
>>>
>>>This is a known position with 17 example games found in the literature and
>>>databases.  14 games played 13.Nbd4 which may be considered the book move, 2
>>>games played 13.Bb4, and 1 game played 13.Rfc1.  Nobody tried 13.Nfd4.  Using
>>>CM8666 Deep (as described previously, analogous to Shep's CM6666 Deep for
>>>CM6000), analysis yielded the following information in the Thinking Lines
>>>window:
>>>
>>>    after 12...Bg4
>>>Time  Depth Score Positions Moves
>>>0:28     9  0.37     1M *   Bb4         (* rounded off to nearest 1000)
>>>1:00    10  0.18     3M     Bb4
>>>2:27    10  0.31     8M     Nfd4
>>>5:58    11  0.23    19M     Nfd4
>>>17:31   12  0.14    58M     Nfd4
>>>23:07   12  0.28    78M     Bb4
>>>41.43   13  0.12   142M     Bb4
>>>2:11:00 14  0.13   448M     Bb4
>>>
>>>So my CM personality thinks 13.Bb4 is the best move in this position with a
>>>+0.13 score at 14 ply.  It didn't find the book move 13.Nbd4.  Why?  Well, maybe
>>>the book move isn't as good.  To test this theory, I made the book move.  If
>>>it's not as good, the resulting analysis should give poorer scores (smaller
>>>positive values) than 13.Bb4.  Here's what it shows:
>>>
>>>    after 12...Bg4 13.Nbd4
>>>Time  Depth Score Positions Moves
>>>0.16     9  0.56     1M     Nxf3+
>>>0.36    10  0.44     2M     Nxf3+
>>>1:33    11  0.43     5M     Nxf3+
>>>6:00    12  0.37    20M     Nxf3+
>>>17:12   13  0.37    59M     Nxf3+
>>>1:05:00 14  0.29   220M     Nxf3+
>>>
>>>Surprisingly 13.Nbd4 gives higher scores than 13.Bb4.  I thought that might be
>>>because the Selective Search was set too high in CM8666 Deep.  So I searched
>>>with smaller Selective Search settings (including a setting of zero, among
>>>others) but it still couldn't find the book move, 13.Nbd4.
>>>
>>>Clearly 13.Nbd4 produces better scores, so why wasn't it found and selected as
>>>the best move?  This is an unexplained anomaly, and I don't know the answer.
>>>Does anyone?
>>>WP
>>
>>I forwarded your post on to Johan, and here is the gist of his reply:
>>
>>-- The GUI supports selective settings of 1 to 12 because those are the
>>"reasonable" settings. The higher you go past 12, the less useful the data is.
>>He basically stated the setting of 32 is pretty much a "garbage in/garbage out"
>>scenario.
>>
>>-- The engine does NOT do any checking for "illegal" values in personalities
>>(meaning values outside of the ranges available in the GUI). Therefore, your
>>setting of 32 WAS used for CM8666 Deep.
>>
>>-- He also stated "This guy is still confused about the score being displayed
>>for side to move. So the second line is just -0.37 for white, meaning the engine
>>likes Nbd4 less than Bb4."
>>
>>-- Therefore, there is no bug, and this is completely acceptable behavior.
>>
>>jm
>
>Preprocessing can be also a reason for analysis anomality that the poster
>described.
>
>I know that a lot of chess engines including some top programs do preprocessing
>and it means that the score is based also on the root position.
>
>I know that Gandalf does not do preprocessing and I am interested to know if
>Chessmaster8000 is supposed to have the same behaviour.
>
>Uri

Here is Johan's response to the question of preprocessing:

"The King changes the material and centrilzation tables gradually from opening
to middle game to end game. These add only a small amount to the total
evaluation, though the changes *are* visible after castling and queen
(re)captures."

jm



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.