I understand you don't like to be my homonymousguyhaw wrote:[DW: EN may relay the EGTs via one of our colleagues shortly. Meanwhile, Marc may convert more of 'the 16' to EN form in his own time. Not ETAs. I've collated MB FEG.lof stats with EN.tbs stats with 100% alignment as expected.]
Knocking out 'unreachable' positions completely, afaik, is an unsolved problem. A position may be retro'd any number of plies before it is realised that it is unreachable, e.g. wBa1, wPb2. [It is not now possible to convert a P to a man of the other colour! ] Therefore, any subtractions on this basis are rather arbitrary.
Guido's [not 'homonymous' but sharing a linguistic root] point about "not computing btm loses in m because it was done before" is fair enough, but in the more efficient programs (which only need bitmaps in RAM), DTM wins in m are not backed up until all DTM losses in m are being looked at.
This 'deferral' is not, I imagine, particularly convenient, and the advantage of DTC and DTZ EGT-generation is that all backing up from subgames can be done immediately 'up front'. Nalimov's indexing-scheme is reversible but is relatively slow to compute as there are lots of integer divisions.
!But you would lose to share the name with Guido from Arezzo, inventor of the musical notes, and I with Guy de Maupassant:-)!
Coming back to chess and EGTBs I agree with you about unreachable positions, but in the kpk ending probably the
computation with one back move is sufficient and theoretically exact, evidence to the contrary excepted. I put this option in the program to see if the reduction of positions could be useful to reduce the cpu time (answer: no at least for my programming skill) and for pure curiosity.
In the retrograde algorithm I compute all the capture/promotion moves in cycle 0, solving definitively at this level only the cases when all the legal moves are of such type, then for the other cases I maintain a bit file used during the normal cycles to compare results from capture/promotion moves and normal moves in order to choose the best.
Your information about reversibility of EN indexing-scheme is interesting and more plausible than the negative answer of
Bob Hyatt remembered by me.
About this problem proposed by you:
"
KBNKPP: 1 btm loss in DTM=122 but no wtm wins in DTM=122 (or 121).
I wonder what the position is? Conjecture: quite likely to be a promotion, and because of the extreme depth, an underpromotion.
"
I agree with you about a promotion, but if it is an underpromotion, we should suppose that a promotion to queen would cause a stalemate (a loss seems unlikely but nothing is impossible) and this is not too much plausible as the white has three men. If it is an underpromotion, it is in general to a knight, but also this is not sure.
Unfortunately I cannot generate this ending at least for now, differently I would have no problem to solve the question.
Remaining in this subject I remember the famous underpromotion with white to move in kbpk:
white: Ka6, Bb6, Pd7
black: Kc6
This is a win in 31, while black has no loss in 30.
The same situation of underpromotion exists if we move all the men of one or two or three square to right, where white wins in 28, 22 and 25 moves respectively, but in these cases there are losses of the black in 27, 21 and 24 moves, even if there is no connection between these losses and the above mentioned victories. The back move should start from positions of the kbnk and not of the kbpk ending.
Ciao
Guido