H. G. Muller wrote on Tue, Jan 2, 2018 11:01 AM UTC:
@V.Reinhart: there is a big difference between exactly knowing which positions of an end-game with a certain material composition are draws or wins in an unclear end-game (such as KPK), and knowing whether in general the material combination offers only very slim chances to be winnable (such as KRKB). (KBNK is always won, btw, if the initial position does not forcibly lose B or N bay shallow tactics.) To answer the first question you would need End-Game Tables. These can be built by retrograde analysis, starting from all possible checkmates with (subsets of) the given material. For end-games with 4 or 5 men, it would even be possible to do such analysis 'on the fly', instead of a forward search that engines use for normal play. They can also be pre-calculated, and stored on disk, so the engine can probe them.
Here we are talking about the simpler, and more general issue of recognizing the cases where the simplistic addition of piece values gives a misleading measure of the advantage. This typically occurs for positions without Pawns (or other pieces with decisive promotions). or with just 1 Pawn that can be destroyed by a sacrifice. E.g. KBPKN is pretty hopeless, because black can sac the Knight for the Pawn to convert to a dead draw.
In the Pair-o-Max fork of Fairy-Max I used the following general heuristic:
If the strong side has no Pawns, divide the score by 2
If in addition, his advantage in non-Pawn material is less than 350cP, divide the score by 8 instead
If the strong side has only a single Pawn, and will be less than 350cP ahead after the opponent sacrifices his weakest non-Pawn for it, divide the score by 4
These rules can be slightly modified by properties defined in the game definition:
If the strong side has only a single piece, and it is color bound, it will be assumed to be worth less than 350cP for application of rules 2 and 3 , even if its normal piece value is much higher. (E.g. Bede (BD) or Adjutant (BDD))
Pieces worth <350cP can be defined as 'mating minors', and the <350cP ahead rules will not be applied to a player that still has such a mating minor. E.g. Commoners (K) or Woody Rook (WD).
Pieces can be defined as 'defective pairs', and having a pair of those and nothing else will be considered as worth less than 350cP. (E.g. Kight)
Pieces can be defined as 'strong defenders', and then will be assumed to be able to hold a draw (dividing the score by 8) against a single Queen-class piece even through they are worth less than 350cP (so that the advantage of the Queen side can be >600cP). (E.g. Commoner)
@V.Reinhart: there is a big difference between exactly knowing which positions of an end-game with a certain material composition are draws or wins in an unclear end-game (such as KPK), and knowing whether in general the material combination offers only very slim chances to be winnable (such as KRKB). (KBNK is always won, btw, if the initial position does not forcibly lose B or N bay shallow tactics.) To answer the first question you would need End-Game Tables. These can be built by retrograde analysis, starting from all possible checkmates with (subsets of) the given material. For end-games with 4 or 5 men, it would even be possible to do such analysis 'on the fly', instead of a forward search that engines use for normal play. They can also be pre-calculated, and stored on disk, so the engine can probe them.
Here we are talking about the simpler, and more general issue of recognizing the cases where the simplistic addition of piece values gives a misleading measure of the advantage. This typically occurs for positions without Pawns (or other pieces with decisive promotions). or with just 1 Pawn that can be destroyed by a sacrifice. E.g. KBPKN is pretty hopeless, because black can sac the Knight for the Pawn to convert to a dead draw.
In the Pair-o-Max fork of Fairy-Max I used the following general heuristic:
These rules can be slightly modified by properties defined in the game definition: