Author: Anthony Bailey
Date: 12:35:36 09/16/99
Go up one level in this thread
On September 15, 1999 at 15:35:05, guy haworth wrote: > You are right that KQQKQQ is not the first EG-database that might be relevant > to GK-v-WT ... but as it doesn't have Ps, is symmetric and has two pieces > the same on each side, it's a good first run out for any 5-man code on a 6-man > problem. Sure. These facts make it a completely solvable endgame using only small modifications to existing software and lotsa computer power. Whatever else anyone attempts, Eugene Nalimov's production of this tablebase will be extremely useful, and it he wasn't doing it, we'd seek to persuade someone else to try! I will be interested to hear how it turns out... > You are also right that the WT (Black) will probably make its 'last stand' > when White gets its pawn to g7. And any KQP(g)KQP database has to tackle g7 > before g6. So KQP(g)KQP(b7-2/d6-2) would be good. ...the (maybe obvious) point I was hoping to underline was that it not only has to tackle g7 before g6; it can tackle g7 in its entirety before even considering g6. KQP(g7)KQPP(b7-2/d6-2) can be generated before KQP(g6)KQPP(b7-2/d6-2). This halves the size of the tablebase one has to generate at any one time. (I also suspect we should add to the above positions those where one of the two black pawns has been captured. This does not increase the size of the tablebase by an impossible amount (42/30) and keeps many more of the possible game paths inside the tablebase.) > I'm assuming here that the Black 'b', 'd' and 'g' pawns will not divert off > the file because: a) any Q exchange will drop the position to 5-man and > b) neither side is going to sacrifice its Q to a P Absolutely. The same goes for queening on squares other than g8, b1 and d1. Also underpromotions seem unlikely to feature in critical lines (i.e. if the distinction between the game being winnable or drawable from the current position well back down the road is decided only by a line involving underpromotion.) > However, modifying existing code isn't so easy; this would require removing > the 'capture sideways' rule from the program. Yes, it is possible to introduce bugs through such special cases. And rewriting new special case code is even more likely to produce bugs. Still, it might be worth doing. > I tried to find a way of using the KQPKQ-db in the KQPKQPP context - to > create an algorithm which discovered when the bPs were irrelevant to the > result. This was basically about finding: > a) forced wins for White that work regardless of the Pawns > b) draws that work regardless of the Pawns Yes, I read your interesting post on that topic on the match BBS. > I don't think I bottomed out a proper algorithm but it's a 'teaser' you might > like to think about. It seems problematic. Every time a possible move forward for White is invalidated by the presence of a pawn, the current position evaluation becomes useless. Therefore I think you probably have to go back to square one and work backwards, regenerating the tablebase taking the black pawns into account for White moves. In other words, you generate are pretty much generating a KQPvKQPP tablebase where the Black pawns are considered to be unmovable statues on b7 and d6... in which case you might as well generate said tablebase. Of course, this could in itself be a worthwhile thing to do. If Black can draw without moving his pawns, that'll do very nicely. However the human analysis so far seems to suggest that Black has to at least advance the d-pawn a bit for the draw to be held. > You should also considering using possibly inaccurate heuristics which > Wirth/Nievergelt mention in their ICCA J v22.2 paper. If White has 2Qs, Black > has won and there is no capture, mate, stalemate or promotion for 3 ply, then > one might assume White wins. Not strictly true but 'almost true'. ...I believe that a concrete suggestion using approximately this approach formed the main body of the post you were replying to. Probably it wasn't as well expressed as it could have been. Is the Wirth/Nievergelt paper available on-line? - 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.