Author: guy haworth
Date: 06:26:20 10/08/99
Go up one level in this thread
I think the prospect of generating subsets of KQPKQP relevant to the GKasp-World game is remote because of the availability of the key EGT-workers, not because of the intractibility of the problem. It's therefore worth discussing 'in principle'. What is required is a set of EGTs which together include all the positions from the current position with the wP on g5 and the bP on d6. If the pawns stray off the g and/or d file, a capture has taken place, 6-men has become 5-men, and the position is in an EGT that already exists. I agree that it is probably not too dangerous to assume P=Q (so that's MOD 1 to any existing code). The problem is that the EGT will not represent 'absolute truth' or have enduring value for the chess world. The eight KQQKQP(di: i=2..6) and KQQKQP(bj: j=2..4)[alias KQP(g5-7)KQQ] EGTs are needed, and maybe d6 becomes d5 by tomorrow night. That's 7/24ths of KQQKQP, quite a lot, but I suspect there are a lot of quick wins which helps - see kqkp.tbs on Robert Hyatt's ftp server. Each database needs an index scheme to ensure that the range of position-index numbers is not much greater than 4x that for KQQKQ. Just as the chef's secret is in the sauce, the key to efficiency - which Eugene Nalimov has mastered - is in the programming of these indexes. Even better, each KQP(gi)KQP(dj) database - there are 18 - should be creatable with an index not much greater than 4x that of KQKQ. Therefore I suspect that all the ingredients are computable as instore problems for a 1GB RAM machine using Eugene Nalimov's code. KQQKQP(d2) is needed before KQQKQP(d3) etc. KQP(gi)KQP(dj) is needed before KQP(gi)KQP(d,j-1) and KQP(g,i+1). If the necessary KQP(gi)KQP(dj) EGTs can't all be produced in time, a subset is still better than nothing. So, in summary, I believe that the relevant subset of KQPKQP is computable with today's 32-bit architecture and available computers - but the coding effort is not. guy h
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.