Chess Endgames - Tilburg University

Download Report

Transcript Chess Endgames - Tilburg University

Some Notes on Chess
Endgames
Guy Haworth
Reading, Berkshire, UK
<[email protected]>
Chess Endgames, 2001
• Current Situation
• Endgame Table Statistics
– %-results, maximals, and mutual zugzwangs
• “Discarding Like Pieces”
– an ‘established’ principle
– statistical evidence not convincing so far
• Depth by the Rule
– correction to previously defined algorithm
– better news about the computability of DTR EGTs
The Current Situation
• All 3-to-5-Man DTC and DTM EGTs
– http://chess.jaet.org/endings/
EGT service c/o John Tamplin and Thomas Lincke of ETH(Zûrich)
– ftp://ftp.cis.uab.edu/pub/hyatt/TB/ ... EGTs c/o Rob Hyatt
– http://www.chessbase.com/ ... CDs from ChessBase
• Some 6-Man DTC EGTs
– http://cm.bell-labs.com/cm/cs/who/ken/chesseg.html
• Some 6-Man DTM EGTs
– http://chess.jaet.org/endings/ … EGT service c/o John Tamplin
– ftp://ftp.cis.uab.edu/pub/hyatt/TB/ ... EGTs c/o Rob Hyatt
EGT Statistics
• Useful for …
– highlighting rare phenomena
• maximals, mzugs, under-promotion, draws by minority force
– comparing one EGT with another
• ad hoc removal of unreachable positions is not in fact helpful
• Nalimov, Wirth, and Karrer work
– reported at this workshop on their behalf
– % of wins, draws and losses, wtm and btm
– maxDTC and maxDTM results, wtm and btm
•
•
overlap of maxDTC and maxDTM positions not computed
some ‘fringe’ oddities thrown up by the DTC metric
– mutual zugzwangs: ww, bw, fpz
twin-sourced results on distinct positions in full agreement
“Discarding Like Pieces”
• guideline: 6+-man positions can usually be evaluated by
ignoring like pieces? Does this mean:
– look at the statistics, or
– remove the like pieces from the board and evaluate the position?
• full stats information exists on the 2-4 and 3-5 comparison
– but theory is covering these endgames already
• some data available on 4-6 comparisons
– statistics do not seem to support the guideline
• Wirth provides DTC data for 3-5-man endgames
– enabling shallow wins to be discounted
– what about the uncomputed ‘shallow draws’?
Depth by The Rule
• Depth by the Rule (DTR) = dr =
the least k such that best play avoids a k-move draw claim
• “Constrained Optimisation….”, ICGA J v23.1
–
–
–
–
proposed revival of DTZ EGTs
proposed use of DTC, DTM and DTZ EGTs in combination
defined the DTR metric above
defined the incorrect DTR algorithm AL1 using dz
• !! The ‘standard’ retrograde analysis algorithm
– works for computing DTR EGTs
– requires the use of the ‘phase count’ instead of dz
Endgame Challenges
• create more efficient codes for generating DTM EGTs
– c.f. Wu/Beal, Info. Sciences, Vol. 135 (June 2001), pp. 207-228
• create EGTs
– to the DTZ and DTR metrics and/or with correct depths in plies
– identifying ‘finite-depth’ draws forced by one side or the other
– derived from basic EGTs, e.g., bitmap win/no_win EGTs
• mine EGTs for unique moves, studies, problems
– various definitions of ‘unique move’
– is Walter’s KBNK “Mate in 31” wKh2Ba3Nb1/bKd4+w beatable?
•
•
•
•
•
annotate studies (a subset of the “annotation challenge”)
classify studies by theme
define more sophisticated strategies based on EGTs
cite instances of deep endings won over the board
progress Awari, Chess (Chinese, Western, Losing, Mini), other
… add value to, rather than displace, human experience