Author: Eugene Nalimov
Date: 22:00:20 09/06/99
Go up one level in this thread
Theoretically something similar can be done. You can generate only part of the TB, as you know ranks with the pawns, and you don't need huge parts of the TB. It's not necessary to think about en passant. More, it looks that DTC, or even W/D/L TB will be sufficient. Even better, resulting TB can be large, it's not necessary to use complex indexing schema. So you can use Stiller's backward analysis algorithm that is much faster than more traditional forward ones (it heavily depends on uniformity of simple indexing schema, so it can use bit vectors; unfortunately more complex schemas do not have such properties, that's why I did not use that algorithm). I have no time nor resources to do something myself - I am moving into other product development group, so I am very busy. More, on the new position I'll have no access to large servers I used for 5-man TBs generation. Maybe you should contact Stiller. I'll send you his e-mail address tomorrow, from my office. Eugene On September 06, 1999 at 19:46:56, Anthony Bailey wrote: >Disclaimer: I'm a programmer and chess fan, but am ignorant of many of the >techniques used in computer chess. So forgive me if this is a dumb idea... but >I'd like to check out the plausibility of producing specialised tablebase-based >solutions for solving particular endgame positions where a plausibly small set >of positions that must be passed in order to achieve an interesting result can >be calculated. > >In particular, there are some king, queen & pawn endings rapidly approaching in >the Kasparov vs the World match <http://www.zone.com/kasparov/> that, although >they are beyond the current five piece tablebase limit, seem they might be >amenable to a solution along these lines because of some assumptions that could >be made regarding the position. In fact, a guaranteed solution, though >desirable, should not be necessary; analysis with known holes that could be >inspected by human analysts, or analysis that is merely "very likely" to be >correct, would be a great boon to the World team, who are currently hoping to >rescue a draw from what has been an extremely complicated and enjoyable game >against the world champion. > >I made a post about a particular position and its perceived amenability to a >solution by working backwards from positions of known value on the bulletin >board used by the World team to discuss current analysis; you can read it at ><http://bbs.msnbc.com/bbs/kasparov-team/posts/ts/61327.asp>. It was suggested on >that forum that people here would be very well placed to judge the plausibility >of the idea (and maybe some would even like to help out with its >implementation.) I understand that some tablebase experts such as Dr Nalimov >read this board. I'm sure that if they can spare the time, they can tell me >quickly just how ridiculous my ideas are. (c: > >The basic idea is that we have a six piece position, > . .q. . >.KQ . .P > . . . . >. .p. . > . . . . >. . . . > . . . . >. .k. . >with Kasparov as White to play that we are hopeful we can force the game to. >(There are also some other similar positions in the current analysis tree that >contain an extra black b-pawn which we can almost certainly reach; but I want to >look at this one first since solving a six piece ending seems a better first >step to tackle than solving a seven piece one!) > >The key observation I make is that would seem to be possible to calculate a >probable solution (or at least the probable absence of a win for White) by >listing a set of positions through one of which the game would have to pass in >order to be resolved. These would be: a known position from any five piece >tablebase following the capture of a piece; an immediately evaluable position >involving all six pieces (such as a checkmate or stalemate); or a position where >the white pawn promotes on h8 without some catastrophe such as checkmate or >queen loss swiftly befalling White (say, within a search of the following game >tree to a depth of five or six ply.) > >Because of the limited number of squares on which the Black pawn might be found >(and the perceived unlikelihood of it queening; if White's only way to win >involves the black pawn promoting or other peculiarities such as a White >underpromotion this would seem very surprising) it seems that the number of >finishing and possible intermediate positions is not unreasonably large; it >seems to me it would be smaller than, say, the number of positions in a KQPvKQ >tablebase. > >Therefore, is it conceiveable to calculate a plausible set of finishing >positions and backtrack from there in what I understand to be the usual manner >of building tablebases in order to build at least a partial tablebase for this >ending? >And given the existence of a backwards-built tablebase of wins, could one work >forward from the starting position to attempt to derive a drawing variation? > >Your insight would be appreciated. > >Thanks, > - Anthony.
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.