Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Crafty EGTB Bug

Author: Brian Katz

Date: 17:20:40 01/02/04

Go up one level in this thread


On January 02, 2004 at 17:47:03, Luis Smith wrote:

>On January 02, 2004 at 17:02:31, Robert Hyatt wrote:
>
>>On January 02, 2004 at 14:10:59, Luis Smith wrote:
>>
>>>Someone posted this position in the Winboard forum
>>>
>>>[D]3r4/1K6/5N2/8/8/4k3/8/8 w - - 7 98
>>>
>>>This is what Crafty displays under analysis mode.
>>>
>>>depth=0 -327.39 1. Nh7 <EGTB>
>>>Nodes: 0 NPS: 0
>>>Time: 00:00:00.01
>>>
>>>If you disable EGTB's it displays one of the drawing moves.  However if you
>>>force it to play as black it plays Ng4+ one of the drawing moves.
>>
>>

>>I don't know what could cause this.  I set the position up and I get
>>the expected answer in analysis mode:
>
>I think I might have incomplete Tablebases.  Is there a utility that checks for
>you or something?
>
>>
>>White(1): 3r4/1K6/5N2/8/8/4k3/8/8 w - - 7 98
>>1. Nh5 Rd1 2. Nf6 Ra1 3. Ng8 Rb1+ 4. Kc8 Ra1 5. Ne7 Rb1 6. Ng8 Ra1
>>White(1): an
>>Analyze Mode: type "exit" to terminate.
>>end-game phase
>>              clearing hash tables
>>              time surplus   0.00  time limit 22.50 (3:30)
>>              depth   time  score   variation (1)
>>                1     0.00  -0.58   1. Nh7
>>                1->   0.01  -0.58   1. Nh7
>>                2     0.02     --   1. Nh7
>>                2     0.03 -Mat14   1. Nh7 <EGTB>
>>                2     0.04     ++   1. Ng4+!!
>>                2     0.05   0.01   1. Ng4+ <EGTB>
>>                2->   0.06   0.01   1. Ng4+ <EGTB>
>>                3     0.06   0.01   1. Ng4+ <EGTB>
>>                3->   0.06   0.01   1. Ng4+ <EGTB>
>>                4     0.06   0.01   1. Ng4+ <EGTB>
>>                4->   0.06   0.01   1. Ng4+ <EGTB>
>>              time=0.06  cpu=0%  mat=-2  n=92  fh=100%  nps=10K
>>              ext-> chk=9 cap=0 pp=0 1rep=0 mate=0
>>              predicted=0  nodes=92  evals=1
>>              endgame tablebase-> probes=13  hits=13


Hello Luis,

If you have the 4 Disc Set of Nalimov Tablebases (in the TBGEN Folder on the 1st
Disc of the Nalimov 4 disc set) or the Tablebases from the Fritz6 CD, which are
in the TBS folder in the Fritz6 CD. (Fritz 7 shows this folder but it is empty)
You will also see an ICON for Tablebase Generator. "TBGEN"
You can actually let this application show you what endgames are missing using
the Subendgames or the Suggestion Button.

Perhaps I can share a Tablebase problem that I encountered after I loaded the
Nalimov Tablebases 4 Disk Set a few years ago, which by the way have All 3 and 4
men Tablebases. It has 5 men Tablebases, but not all of them are on the Disk. It
does not say All 5 men are on the disk in the instructions either.
 IN THE FOLLOWING POSITION:
With the 4 Disc Set of Nalimov Tablebases as of 3 or 4 years ago, Fritz 6 and
Fritz 7 would not promote the pawn at first, and chooses to check the Black King
for a few moves, due the fact that some of the the 5 man Tablebase files are not
included on the disk.

[D]r7/6P1/5K2/6R1/3k4/8/8/8 w - - 0 1

In the above position, and also similar positions, the easiest and fastest
winning move is obviously to promote the pawn with 1.g8...to a queen or Rook.
With the 4 disc set  using the Fritz 6 and Fritz 7 engine, you will find that
rather than promote the pawn, Fritz checks with 1.Rg4+... and will continue to
check the King for a few moves before finally promoting the pawn, which of
course can be taken by the a8 Rook and then White recaptures and wins easily.
Since the KRRKR endgames are missing and some others such as KRNKR, which would
be the case if White UNDERPROMOTED the pawn, the program then does not know that
particular endgame perfectly. If the Black Rook does not give itself up for the
promoted pawn, the Program is still in a 5 piece endgame that it does not have.
Since it does not know it perfectly it refrains from giving up the pawn for a
Rook so fast.
Once I found which endgames were the missing ones for that particular endgame,
the engine played that endgame perfectly. Some Engines like Crafy were smart
enough to play the correct move with the missing Tablebases anyway. It appears
that Fritz 8 will play the correct move but it shows that it does not have that
endgame in the Tablebases, because it does not show a full list of moves, it
just shows evalutions without showing mate.
Interestingly, Fritz 6 and Fritz 7 would play the Correct move without the
tablebases.

Also, if you used the Tablebases which are included on the Fritz6 CD, it will
also play this endgame perfectly.
I do not know if the later Endgame turbo's from Chessbase cover this full
endgame.

Perhaps this is why your program is not playing that particular endgame
perfectly. You may be missing those files.
I now have a 9 disc set which by the way was also missing a few of the 5 man
endgames. I used the Tablebase Generator to generate those files, and now I have
all the 5 men Tablebases.
You will need a FAST PROCESSOR AND PLENTY OF RAM TO DO THIS, HOWEVER.
Hope you don't mind my rambling on like this.
Hope this helps you in some way.
Brian



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.