CCRL Discussion Board

Well, also here holds: If you play engines like Clueless 1.4, you might as well play micro-Max 4.8, as that beats Clueless easily in a 10+0 Nunn match (70% score).

So far I did not realize this: for the purpose of including the exact consequences of the 50-move rule, it would be necessary to maintain the DTZ in ply, rather than in moves. This would require a different treatment of BTM positions where black is checkmated, and where black is forced to make a los...

Broken positions are positions that according to the mapping between positions and index have two pieces on the same square.

Perhaps it would be good to explain my approach to speeding up EGTB building. In calculating the index for accessing the TB, I avoid doing any conversions between board positions and index, and vice versa. In stead, I generate board positions from board positions, and indices from indices, in parall...

It seems this big a file immediately brought me over the mail-server limit (which is only 15 MB), as my inbox was already pretty full with junk. I cleaned up, and now there should be 5 MB of empty space again.

So if I understand you correctly the actual EGTs generated by FEG are useless to anyone except ChessMaster, as the format is not disclosed. So people generate it only for the purpose of generating the statistics file? I was not aware of this state of affairs. This makes it all the more important to ...

I know little about EGTB formats used by others, but why would it be difficult to extract these zugzwangs from FEG files? Is it because you have to compare the .wtm and .btm files with swapped colors, so that the indexing is completely different for opposite sides to move? If so, it seems not so awf...

Well, perhaps I am overdoing it. Seems KQKR does not have any stalemates at all. With non-bare black King I guess the number of stalemates will be very very low in general. So it is silly to worry about them. The nice thing about counting black's remaining non-losing moves (as negative numbers) in t...

OK, thanks for the references. I had to pause a few days because of the drudgery of normal life (tax declarations and such... :? ). I am currently doubting the wisdom of not counting the black conversion moves. Basically the problems I am facing are due to the fact that the WTM positions in my build...

I tried marking the positions where black is in check as won (for white with white to move) through stepping along Queen lines from the Queen square in the black King dimension, for every constellation of white pieces and the black Rook. (Doing this at a time when all the involved entries are in the...

One of the areas where I achieved considerable speedup compared to my previous algorithm is the counting of the number of moves in each position. The availability of such a move count for the losing side can speed up the building process greatly: to determine if the precursor of a position that was ...

I created this thread to post new developments in my project to write an EGTB builder that is fast enough to do its work 'behind the board' (e.g. in ponder time). Apart from the obvious advantage that this would not make an excessive demand on disk storage capacity, the real benifit occurs in end-ga...

Surely it wouldn't be so difficult to extract the required statstics if you have the corresponding TB? If you have one kind of statistics you can even greatly speedup the process by just probing for the difference. Only 21 out of the 462 King-pair positions have both Kings on the dagonal, so you wou...

Well, I am back from a great vacation, so you will hear from me soon. I came back with lots of ideas. First I had to implement those concerning my engine Joker, though, which is competing in OpenWar2, and coud use some improvement... :? Now that is done, I am going to code my latest EGTB algorithm f...

Indeed. I only discovered this forum recently, but am also a little bit concerned by the low level of posting frequency. On Talkchess.com there is generally more response, even on a specialist subject like EGTBs.

OK. I will try the e-mail. My guess why escape counting was not working for Ken, while it seems to work for me, is that he was building the TB on disk rather than in memory. Redoing a calculation 4 or 5 times, even a complicated one like a one-ply search, is always much faster than a disk access, bu...

The 'random' accesses are actually not very random if you design the indexing scheme properly. I use the black piece locations to derive the low-order index bit. So during potential-loss verification you sample the TB very locally, (4KB blocks for 2 black pieces, 256KB for three, nicely fitting in t...

This paper, 6-Piece Endgames is in the ICCA Journal, v19.4, pp215-226 . Do you happen to have an URL for this paper? I could not find it on the web. I did find the paper by Wu and Beal, that discuss the bit-map strategy. I guess this is very similar to what I do, except that I hide the bit map as o...

OK, thanks for the reference, I will have a look what Ken Thompson has to say on this. For me it so far is the fastest way I found for end-games that are predominantly won. Only in end-games that are virtually without wins it is a big waste of time to count moves in positions that you will never vis...

My generator experiences a large speedup as the number of won states dry up, as it then can skip all cache lines that do not contain any won states. Before I did that, a single cycle through a 5-men symmetrized TB (128MB) took at least 1.2 sec, even if there was almost nothing in it to process. Now ...

I calculate currently DTM. The advantage of DTC would be that I only need win/loss information on the TBs it converts to. For capture conversions it doesn't really matter, as my current algorithm automatically calculates all TBs with a subset of the pieces as well. (This is included in the quoted ti...

Why do you expect that to be faster? I don't see how having the bitbase could shorten the building process.

It might help to have a bitbase for the TB with Queens to which one with a Pawn corresponds after promotion, though. Then it cn easily look up if a promotion is winning or not.

As I want to have my engine construct EGTBs 'behind the board' as it needs them (so that I can handle 9-10 men end-games with Pawns without needing gigabytes of pre-computed information), speed is of the utmost importance. My own EGTB generator for pieces-only, (i.e. applying 8-fold symmetry) now ge...