Author: Robert Hyatt
Date: 16:53:14 09/22/99
Go up one level in this thread
On September 22, 1999 at 09:50:08, Robert Hyatt wrote: I have gotten the new tbindex stuff into crafty. I will try to release a new version later tonight that has this. It will probably have everything current except for the new majority stuff I am still tweaking. Latest version has three improvements of interest: 1. it outputs 'insane' (or redundant) FEN. :) Every rank has 8 squares specified, no trailing "/". 2. You can just cut and drop fen right into crafty without the 'setboard' on the front. It will take it just fine. If the position is a known tablebase position, it will give you the full mating/drawing line automatically (I may disable this later, see 3 below) 3. there is another use for the 'egtb' command. after using it to enable tablebases, you can say "egtb" and get the full (optimal) PV for the current position, if it is a TB position. If you say "egtb !" it will append a ! to any move where there is only one best choice and all others are inferior... More later... >On September 22, 1999 at 09:35:46, guy haworth wrote: > >>This note contains a request - for Robert Hyatt I think ... >> >>As CCC readers may be aware, Eugene Nalimov and Christoph Wirth have >>independently created a complete and self-consistent KQQKQQ endgame database. >> >>Both authors responded readily to the request of the World TEam (in GK-v-WT) and >>their achievements are without parallel. Many thanks and warmest >>congratulations to both. >> >>Eugene's is optimised to DTM (distance to mate)and Christoph's to DTC (capture >>or mate). However, they can clearly be used to second-source each other on the >>value of a KQQKQQ position. This gives more authority to both databases. >> >>A number of players involved in the Kasparov-v-WorldTeam event (and others) will >>now be thinking of using this KQQKQQ database and this leads to thoughts about >>the practicality of doing this. There appear to be two issues. >> >>The KQQKQQ database is a 6-man rather than a 5-man database and as such, I >>understand requires a modification to standard CRAFTY. >> >>It is also a very large database (407503Kb) and is presumably difficult to >>download in the standard way. >> >>May I ask Robert to define on this CCC board how to make the modifications to >>CRAFTY to use 6-man EGs ... and how the KQQKQQ database itself may be best >>acquired. >> > > >There are two parts to this. I haven't done them yet, because I am in the >middle of testing the pawn-majority code I have added, and I try to limit >changes to one area to avoid going nuts debugging. > >1. the actual source for crafty doesn't have to be changed. I fixed this >months ago. Eugene reports the number of pieces I should use to probe the >databases and I use this number. If you put a 6 piece file out there, I'll >probe it. > >2. egtb.cpp is the place where it must be changed. This needs to have the >old probe code of his removed (the tbindex stuff) and the new code added (the >new code is on the ftp machine in tbgen_nw.zip). That is the only real change >and it should be easy. It will hopefully become a part of the next version if >I have time to merge it in, since it doesn't slow anything down (it will take >a special compile to enable 6 pieces probably, I don't know what including their >support will do memory-wise if you don't have any. Eugene?) > > > > > > > >> >>As a footnote, both Eugene and Christoph 'warmed up' their codes on KNNKNN and >>KRRKRR. Eugene later produced a KBBKBB EG-db as well. All four of Eugene's >>datbases are on ftp://ftp.cis.uab.edu/pub/hyatt/TB/. >> >>As a technical footnote, Eugene used a Compaq Alpha procesor with 4GB RAM and >>ran the problem 'in store'. Christoph had 1GB RAM and therefore sustained some >>100GB of 'paging' disc traffic. Both databases are post-hoc checked to ensure >>that the value/depth of a position is consistent with that of its successors. >> >>Again, thanks and congratulations to both Eugene Nalimov and Christoph Wirth.
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.