Current status of 7-men tablebases?

[ZeroOne]

I cannot seem to find any up to date information so I decided to ask. What's the current situation with 7-men tablebases? All 6-men tablebases have been created so I think the creation of 7-men tablebases should be in good progress, but the situation appears to be completely opposite. I have gathered that a few have already been created but not published. I also got the impression that the Nalimov format does not support more than 6-men? Everyone also keeps talking about some "tbgen.exe" program, which, however, does not seem to be available for download anywhere.

I have been downloading some tablebases recently and would like to donate some computing power for calculating some 7-men tablebases if I only knew how.

[vb4]

I cannot seem to find any up to date information so I decided to ask. What's the current situation with 7-men tablebases? All 6-men tablebases

Hi ZeroOne well not all 6 piece EGTB have been done, just the 3-3 and 4-2 with and without pawns. 5-1 has never been done and there seems to be no interest in getting it done.

have been created so I think the creation of 7-men tablebases should be in good progress, but the situation appears to be completely opposite. I have gathered that a few have already been created but not published. I also got the impression that the Nalimov format does not support more than 6-men? Everyone also keeps talking about some "tbgen.exe" program, which, however, does not seem to be available for download anywhere.

I have been downloading some tablebases recently and would like to donate some computing power for calculating some 7-men tablebases if I only knew how.

I will look through my records and send you what I have and who is working on them. I just cant remember the name at the moment. I am glad someone is trying to push the developement of the 7 piece EGTB.

Take care

Les

[guyhaw]

Yakov Konoval has a new super-fast code that computes 7-man EGTs to the DTC(onversion) metric. Marc Bourzutschky has production-run this code to create several P-less endgames' EGTs and a few P-ful ones too.
I think that further development of the code, to create P-ful endgames' EGT to the DTZ metric, P-slice by P-slice, is on the back-burner at the moment, but maybe there's some more production-running going on.
g

[jshriver]

What's the current situation with 7-men tablebases?

Also curious of the status. Need to feed the data monster

-Josh

[guyhaw]

g

[ZeroOne]

Yakov Konoval has a new super-fast code that computes 7-man EGTs to the DTC(onversion) metric.

What if one is not interested in the metrics but just the status of a position -- whether it's won, lost or a draw? Anyone developing any new bitbases? I'd think that's a faster process than developing full Nalimov sets. 5-man Scorpio bitbases take about 350 MB, compare that to 7 GB Nalimov bases. If the trend continues, 6-man bitbases would take about 60 GB, which is perfectly manageable unlike the 1.2 terabytes that Nalimov tables take. So, are there any 6, 7, 8 or more man bitbases in sight?

[guyhaw]

Bitbases certainly have their uses, as the SHREDDERBASES demonstrate. Don't know if there are 6-man Sh'bases.
I think bitbases are also a useful 'opening gambit' when creating EGTs some metric like DTC(k), DTM(k), DTR or DTZ(k). I suspect they can speed up the process, but it's only my surmise. Maybe Yakov or Marc would say I was wrong.
g

[jshriver]

Just bumping this up, to see if there is still continued interest. Who is doing what with 7men, what type of generator, and what the status is. Any info is appreciated

With dual core and SMP machines being common now adays even 64bit, generating ought to be a little easier
-Josh

[Dhanish]

Hi all,

Is anybody working on the 7 men bases? At least selected ones?

I would be interested in krppkrp!

Regards,
Dhanish

[Codeman]

Some 7 men endgames do exist (http://www.gothicchess.com/javascript_8 ... dings.html)
but as far as I know, none including pawns have been generated yet.

You should be thinking about using Freezer, it is perfect for endgames with blocked pawns, as this could then be simplified a lot.

Codeman

[Arpad Rusz]

Some 7-men tablebases with pawns have been generated more than a year ago. See the following link:

http://www.vlasak.biz/tablebase.htm

But there is a problem: no news since then...

[Dhanish]

Hi Codeman & Rusz,

Thankyou for the interesting links.

It looks like krppkrp is quite far away. I have a correspondence game going on with this material!

Regards,
Dhanish

[jkominek]

Some 7 men endgames do exist (http://www.gothicchess.com/javascript_8 ... dings.html)
but as far as I know, none including pawns have been generated yet.
Codeman

A caveat here -- those super-long 7-man checkmates are due to the work of Konoval and Bourzutschky and are with respect to the DTC (distance to conversion) metric, not DTM (distance to mate). Often they will be equal, but in general there is the potential for a shorter mating line, disregarding the 50-move rule.

You might enjoy this announcement of a year ago from Marc B. himself.
http://kirr.homeunix.org/chess/discussi ... ?f=6&t=805

john

[h.g.muller]

It looks like krppkrp is quite far away. I have a correspondence game going on with this material!

KRPPKRP is actually very close. Calculating the P-slices you need for that correspondence game is almost trivial with current computer technology. I don't know exactly where these Pawns are on the board, but even if they are all passers on their initial rank, they have only 6 positions, so 216 Pawn conformations for the total set. The 4-men (KRKR) P-slices are only 16M entries. So with one byte per entry (enough to hold most DTMs), there whole thing fits in 3.3GB. Can even be done on a 32-bit machine!

[Codeman]

It looks like krppkrp is quite far away. I have a correspondence game going on with this material!

KRPPKRP is actually very close. Calculating the P-slices you need for that correspondence game is almost trivial with current computer technology. I don't know exactly where these Pawns are on the board, but even if they are all passers on their initial rank, they have only 6 positions, so 216 Pawn conformations for the total set. The 4-men (KRKR) P-slices are only 16M entries. So with one byte per entry (enough to hold most DTMs), there whole thing fits in 3.3GB. Can even be done on a 32-bit machine!

Aren't you missing all the sub-endgames for promotions and captures?
KQRPKRP
KQQRKRP
KRPPKQR
KQRPKQR
KQQRKQR

KRPKRP
KPPKRP
KRPPKR
KRPPKP

Codeman

[h.g.muller]

Indeed I was assuming the daughter EGTBs were available. They only have to be available as a bitbase.

The daughter EGTB after capture are all much smaller, and don't pose a significant challenge. After promotion is a bit more tricky. I doubt if you would need all the EGTBs you mention, though. I can imagine you could easily prove that KQQRKRP is generally won without having KQQPKQR. (Some wins might take a bit longer, but for building a bitbase that is not relevant.)

The number of positions that will remain undecided with the assumption that KQQRKQR is always draw or lost (a very unrealistic assumption, of course) will be extremely small, and two-ply searches from them (probing KQQRKR) will decide most of those as well. For what is left you can probably easily afford deeper searches to decide them. (If you need them at all; note that for building KQRPKRP you only need the positions in KQQRKRP where a Q is on the 8th rank.)