Difference between revisions of "Capture Extensions"
GerdIsenberg (talk | contribs) |
(stockfish capture extensions) |
||
| Line 7: | Line 7: | ||
A lot of todays programs, more focused on [[Reductions|reductions]] rather than extensions in the [[Scout]] part of their [[Principal Variation Search|PVS]], most often don't reduce captures, which makes it a kind of implicit extension relative to reduced late [[Quiet Moves|quiet moves]]. It also seems worth a try to reduce even captures with negative SEE while performing [[Late Move Reductions|LMR]]. | A lot of todays programs, more focused on [[Reductions|reductions]] rather than extensions in the [[Scout]] part of their [[Principal Variation Search|PVS]], most often don't reduce captures, which makes it a kind of implicit extension relative to reduced late [[Quiet Moves|quiet moves]]. It also seems worth a try to reduce even captures with negative SEE while performing [[Late Move Reductions|LMR]]. | ||
| − | However, it seems important to extend captures if a transition to certain late endgames occurs. [[Ed Schroder|Ed Schröder]] for instance extends '''three''' [[Ply|plies]] in [[Rebel]] if entering the [[Pawn Endgame|pawn endgame]], and the static, [[Incremental Updates|incremental updated]] [[Score|score]] dominated by [[Material|material]] is not outside a +-3 pawn window around zero: The extension is very powerful, it will often avoid REBEL entering a lost ending and vice versa. | + | However, it seems important to extend captures if a transition to certain late endgames occurs. [[Ed Schroder|Ed Schröder]] for instance extends '''three''' [[Ply|plies]] in [[Rebel]] if entering the [[Pawn Endgame|pawn endgame]], and the static, [[Incremental Updates|incremental updated]] [[Score|score]] dominated by [[Material|material]] is not outside a +-3 pawn window around zero: The extension is very powerful, it will often avoid REBEL entering a lost ending and vice versa. |
| + | |||
| + | =Modern Capture Extensions= | ||
| + | |||
| + | Introduced by [[Taras Vuk]] and subsequently simplified into its current form. Stockfish extends a move that captured the previously moved piece by one play in a [[Node_Types#PV-Nodes|PV-node]] when [[History Heuristic#Capture History|capture history]] score of that move is high. | ||
=See also= | =See also= | ||
Latest revision as of 00:25, 24 July 2024
Home * Search * Selectivity * Extensions * Capture Extensions
Capture Extensions were used early in brute-force chess programs to minimize some horizon effects. Due to the self-restricting nature of captures in Chess (max. 30 pieces may be captured per game), one may consider this a good idea. However, extending every capture was discarded by most programmers, and some constrains or conditions were applied, i.e. to extend only forced recaptures on the same target square.
Hans Berliner suggested to extend only those (re)captures, which restore the material balance at the root. Furthermore, capture extensions are often restricted to PV-nodes, or some programs assign fractional extensions to different kinds of captures, considering SEE, distance of target square to the opponent king, recapturing on the same square, and material balance as mentioned.
A lot of todays programs, more focused on reductions rather than extensions in the Scout part of their PVS, most often don't reduce captures, which makes it a kind of implicit extension relative to reduced late quiet moves. It also seems worth a try to reduce even captures with negative SEE while performing LMR.
However, it seems important to extend captures if a transition to certain late endgames occurs. Ed Schröder for instance extends three plies in Rebel if entering the pawn endgame, and the static, incremental updated score dominated by material is not outside a +-3 pawn window around zero: The extension is very powerful, it will often avoid REBEL entering a lost ending and vice versa.
Contents
Modern Capture Extensions
Introduced by Taras Vuk and subsequently simplified into its current form. Stockfish extends a move that captured the previously moved piece by one play in a PV-node when capture history score of that move is high.
See also
Publications
- David Levy, David Broughton, Mark Taylor (1989). The SEX Algorithm in Computer Chess. ICCA Journal, Vol. 12, No. 1, pp. 10-21.
- Don Beal and Martin C. Smith (1995). Quantification of Search-Extension Benefits. ICCA Journal, Vol. 18, No. 4, pp. 205-218.
- Yoshimasa Tsuruoka, Daisaku Yokoyama, Takashi Chikayama (2002). Game-Tree Search Algorithm based on Realization Probability. ICGA Journal, Vol. 25, No. 3, pdf, pdf
- David Levy (2002) SOME COMMENTS ON REALIZATION PROBABILITIES AND THE SEX ALGORITHM. ICGA Journal, Vol. 25, No. 3
Forum Posts
- Capture/recapture extensions by Andrew Williams, CCC, December 14, 1998
- Search extensions at promising trajectories by Reinhard Scharnagl, CCC, May 28, 2008
