Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Smarter programs not bigger TBs

Author: Robert Hyatt

Date: 17:56:19 04/13/00

Go up one level in this thread


On April 13, 2000 at 11:21:11, blass uri wrote:

>On April 13, 2000 at 09:22:44, Robert Hyatt wrote:
>
>>On April 13, 2000 at 00:43:02, Dave Gomboc wrote:
>>
>>>On April 12, 2000 at 23:18:38, Robert Hyatt wrote:
>>>
>>>>On April 12, 2000 at 22:27:22, Dave Gomboc wrote:
>>>>
>>>>>On April 11, 2000 at 16:44:59, Robert Hyatt wrote:
>>>>>
>>>>>>On April 11, 2000 at 12:18:46, Jason Williamson wrote:
>>>>>>
>>>>>>>On April 11, 2000 at 09:31:51, Robert Hyatt wrote:
>>>>>>>
>>>>>>>>On April 11, 2000 at 03:07:48, Jouni Uski wrote:
>>>>>>>>
>>>>>>>>>On April 10, 2000 at 17:22:09, Arndra L. Sharp wrote:
>>>>>>>>>
>>>>>>>>>>On April 10, 2000 at 15:52:18, Robert Hyatt wrote:
>>>>>>>>>>
>>>>>>>>>>>On April 10, 2000 at 11:20:51, Arndra L. Sharp wrote:
>>>>>>>>>>>
>>>>>>>>>>>>First of all let me state that I am not a programmer, just someone who enjoys
>>>>>>>>>>>>computer chess immensely.  I have seen a lot of posts that put down the Endgame
>>>>>>>>>>>>Turbo disks because they do not contain all of the TBs, in particular those TBs
>>>>>>>>>>>>after the pawn queens.  I think it a good idea to reduce the hard drive space
>>>>>>>>>>>>required for TBs by pruning those TBs that any good chess program can figure out
>>>>>>>>>>>>if its brain was not disabled.  It seems the real problem is that those programs
>>>>>>>>>>>>that use TBs turn off the permanent brain once the program is in a TB position
>>>>>>>>>>>>and then the program gets confused if after a pawn queens and the now simple win
>>>>>>>>>>>>(for good chess programs) is not in the TB folder.  This is something that the
>>>>>>>>>>>>programmers probably did not anticipate originally everyone has seen games where
>>>>>>>>>>>>this impacts the result. Now that this has been identified, why can't the
>>>>>>>>>>>>programmers tell their programs to follow the TB tree to pawn promotion and
>>>>>>>>>>>>reset the permanent brain at that point.  Many people have reported that the
>>>>>>>>>>>>programs that blunder with missing 5 man TBs play the same ending fine with 3
>>>>>>>>>>>>and 4 man TBs.  It just seems that the programs don't know how to think again
>>>>>>>>>>>>after they start down a tree and the tree ends before checkmate.  A bug fix by
>>>>>>>>>>>>the programmers would be more preferable than taking up another 5 gigs of hard
>>>>>>>>>>>>drive space.
>>>>>>>>>>>>
>>>>>>>>>>>>Arndra
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>Crafty does this correctly.  But with the price of disk drives, holding all the
>>>>>>>>>>>3-4-5 piece files is now trivial...  40 gigs for 250 bucks is typical now.  You
>>>>>>>>>>>only need 8 gigs for _all_ the 3-4-5 piece files (compressed).
>>>>>>>>>>
>>>>>>>>>>Does anyone sell all the TBs on CDs?
>>>>>>>>>
>>>>>>>>>Try http://mitglied.tripod.de/ChessBits/index.html! They sell 10 CDs or one
>>>>>>>>>harddisk for 3-5 piece.
>>>>>>>>>
>>>>>>>>>Jouni
>>>>>>>>
>>>>>>>>
>>>>>>>>Unfortunately it isn't all the 3-4-5 piece files.  They add up to almost 8
>>>>>>>>gigs compressed, which won't come close to fitting on 10 CDs...
>>>>>>>
>>>>>>>All the TB - the 6 piece ones from your site come to 7.05 gigs (7,580,368,539
>>>>>>>bytes).  This is all the 5 piece ones correct?  Or perhaps I missed a few...
>>>>>>
>>>>>>
>>>>>>the 7.5 gigs is right...
>>>>>>
>>>>>>which won't fit on 10 cd rom disks...
>>>>>
>>>>>But it will fit on 3 or 4 DVD disks.  Maybe Chessbase will use DVD if they print
>>>>>another run of these in the future.  All of the 3-4-5 piece databases would fit
>>>>>on a single DVD.
>>>>>
>>>>>Dave
>>>>
>>>>
>>>>I don't follow.  I wasn't aware that a single DVD could hold almost 8 gigs,
>>>>which is the total size of the 3-4-5 piece files (compressed).
>>>
>>>My mistake.  I thought the 7.5 gigs included the six-piecers that had been
>>>constructed.  I obviously didn't read all of the quoted text carefully! :(
>>>
>>>DVDs hold 2.something gigs each, I think.  So I guess it would take 3 or 4 DVD
>>>disks just for the 3-4-5-piece databases.  That seems reasonably doable.
>>>
>>>Dave
>>
>>
>>The problem is that Eugene slipped in another 2 gigs of 5 piece files on my
>>ftp site.  We now have _all_ 5 piece files, including the 4 vs 1 configurations.
>>Almost 8 gigs just for 5's now.  I had not even noticed these until I copied
>>the things over to one of the 8 quad xeons I have been playing with, and I
>>wondered "why is this now 7.5 gigs when it was under 6?"  I found out.  :)
>
>4 vs 1 are the less important tablebases because I believe that programs can
>always find the right moves by search without tablebases in 4 vs 1 endgames.
>
>It can save some space to use only win/draw tablebases without distance to mate
>in 4 vs 1 endgames.
>
>Uri


You will mis-evaluate lots of positions, if you start in a 5 piece file but
trade into a position you have no database for.  IE KNN vs KP has lots of
wins, and lots of draws.  Ditto for endings like KRP vs KR...  without the
3/4 piece files you don't see that your opponent is trading into a drawn
position or lost position when he was winning before, because you don't know
how to evaluate the 4 piece files.  If you are given a 4 piece ending to
start with, you probably will win most of the winnable cases, even without
the 4 piece files.  But the transition from 5 to 4 becomes clumsy...



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.