Chris Wirth's endgame tablebases

[Denis P. Mendoza]

As I'm fascinated with these different chess endgame databases/ I came across an older chess engine, ChesterfieldCL which uses Chris Wirth's endgame tablebases. Has anyone have an idea where to get his full generated tablebases in sdb format? His old sites were already obsolete.

[jkominek]

Now that's an interesting task of computer archaeology.

Quickly scanning his major publication on the subject ... Wirth, C. and Nievergelt, J. (1999). Exhaustive and Heuristic Retrograde Analysis of the KPPKP Endgame. ICCA Journal, Vol. 22, No. 2, pp. 67-80 ... there is no indication of how to get his data. I suppose you had to write personally. In the article it's given as wirthc@inf.ethz.ch, but that is 8 years old and is now most likely a bounce-back address.

With emule and Kirill's site doing such a good job of disseminating Nalimov tablebases, I'd be pleased to see all the older formats hosted as well -- for the sake of history and posterity. Having them vanish into the Great Bit Bucket in the sky would surely be a shame. I'm too new to this game to be of much assistance. Someone with better connections, or the old-timers of chess programming come to mind. Jonathan Schaefer is likely to have Thomspson bases; Robert Hyatt may have kept them too. The separately generated data of Edwards, Wirth, and Heinz will be harder to acquire.

As I see it, acquiring old chess programs is the hobby of collectors. Acquiring old endgame tables is the purview of museums. There is just something so ... so eternal about them.

john

[Kirill Kryukov]

Hi Denis, John! I'll be happy to help sharing any excavated endgame tables.

[Denis P. Mendoza]

Edwards, Wirth, and Heinz are also the best contributors in this endgame studies. The generated endgame tables of these individuals maybe too old yet still useful for study. I hope somebody has or knows where to get these egtbs of Chris Wirth. All the old archived sites and email address are already obsolete. These old egtb formats are priceless gems, not only for collection but to future developers of this sort.

Since this forum focuses on endgame tablebases, I assumed that this was the right place for my query. I hope the site also hosts archiving of other egtb/bitbase formats as well like that of Knightdreamer, Patzer, etc.. -as well as the generators.

Thanks John for the encouragement. We'll be waiting for the excavation of these tables Kirill I'll host the archives if we digged this treasure!

Denis

[guyhaw]

You could contact ETH in Zurich via http://www.ethz.ch/. They might have archived Chris' programs: the (DTC) EGTs themselves would not take long to generate.
I only have his files of maxDTC positions, zugzwangs and stats which he sent me years ago. Depth was measured in plies, losing no information.
Only one problem with the EGTs and derived stats: CW assumed that the winner always captures, which is not always the case. So, where the loser must force-capture, his ply-depths are one-out for some positions.
Guy

[jkominek]

If it could be tracked down that would be great, but I have my doubts about an institution holding on to such a thing, or if they have, knowing where it is. A Masters thesis yes, but a piece of software no. ... In a similar quest, it would be great to unearth R. Gasser's RETROENGINE which served as the starting point for Wirth's work.

After spending 2-3 hours google-hunting for Edwards tablebases , I have found a grand total of: three! krk, kbbk, and kqkb. They were part of a crafty rpm package, version 15.9-1. Too bad all were not included. The documentation says all 3-man and most 4-man tables exist(ed), and refers to ftp://ics.onenet.net/pub/chess for download. Which is of course extinct.

john

[guyhaw]

With respect to past computers of EGTs, I don't share jk's enthusiasm for hanging on to past EGTs. They were all noble achievements, but several had some rather quirky features which might only serve to confuse today:

- Ken Thompson's 5-man DTC/DTZ EGTs assumed the loser would capture (unforced) to get to a greater depth, thus not maximising DTC/DTZ,
- Wirth's EGTs' DTC-depth was in plies (for greater accuracy) but assumed the loser would never effect the move to the next phase - not true,
I think, now I am reminded of it, that this error stemmed from the RETRO-ENGINE code on which his work was based

Nalimov's EGTs do not recognise castling rights but his EGTs can be iterated to the correct ones by the addition of these relatively few positions.

What would be good is a core library of software so that others can build their own EGTs to their own metrics without having to start from zero again.

guy

[Denis P. Mendoza]

Thanks guy for the infos. It is really better nowadays to just acquire the generators of these databases as most hardwares nowadays are very fast. I just wanted to have CW endgame databases for my ChesterfieldCL project.

The old EGT's have some flaws, that's why Nalimov's EGT is still the most reliable though bulky. But i hope some open-source bitbases like Scorpio continue developing the 6-man databases for us users.

[jkominek]

Denis - While aware of the program Chesterfield, I can only guess at decoding "CL" and fess ignorance to ask: what is the ChesterfieldCL project?

From your perspective, are there any shortcomings to converting the Nalimov tables to bitbases?

john

[Denis P. Mendoza]

Denis - While aware of the program Chesterfield, I can only guess at decoding "CL" and fess ignorance to ask: what is the ChesterfieldCL project?

From your perspective, are there any shortcomings to converting the Nalimov tables to bitbases?

john

Honestly, this project is just a special interest of study on my part on some chess engines with chess learning. As it uses a rare endgame database, I was hoping that I could acquire a much larger archive (from CW) aside from the default files from Matthias Lüscher. I am also generating my own learning files from high quality games using Chesterfield's learning utility and see for myself the difference. This is only one of my odd projects at the moment.

I'm no expert on endgame databases, but the way I see it, nalimov tablebase is already an "oracle" of perfect endgame play, though needs a high demand for diskspace. Converting it to bitbase is losing some of its important features like DTM. Most engines using Nalimov EGTB benefit from this. With respect to bitbases like Scorpio egbb, I notice in the CCRL ratinglist that some engines lose some ELO points when bitbases are enabled. I'm not sure why is this so!

What we really need is how to find another alternative similar to nalimov but highly compressible like bitbases in the future. To me, 5-6 men tablebases/bitbases is just enough. The rest is already the job of the engine

[guyhaw]

The SHREDDER-centric work on BitBases shows that engines can be made much more efficient by 'fronting' Nalimov DTM EGTs with these Bitbases. Why go looking for a depth if the position is a draw? I anticipate Stefan/Eiko extending Bitbases to 6-man endgames at some time - why not?!
Past work done with Marc Bourzutschky's wonderful code to create DTC and DTZ EGTs shows that the DTZ EGTs compress to half the size of the DTM ones. This is because smaller numbers are involved and the compression process is more effective.
g

[Ray]

I anticipate Stefan/Eiko extending Bitbases to 6-man endgames at some time - why not?!

Stefan is certainly working on them last I heard, we may see them quite soon