7-men Syzygy attempt

Discussion of chess software programming and technical issues.

Moderators: hgm, chrisw, Rebel

User avatar
Ozymandias
Posts: 1536
Joined: Sun Oct 25, 2009 2:30 am

Re: 7-men Syzygy attempt

Post by Ozymandias »

Nordlandia wrote: Thu Nov 29, 2018 11:53 amYou mean HDD space... money and time ;)

10 TB HDDs is available at 300-350 dollars at the time being. Wdl ~ 10 TB.
Yes, time too. Games finish earlier with the same adjudication parameters; the bigger the TB set, the shorter the games are.

I was thinking more along the lines of 8TB HDD + 2TB SSD. I already have a Seagate Archive, I would only need to offload the most relevant content to Blu-Ray discs (already cheaper than HDD space) and delete the rest. The Crucial MX500 is also around 300 for the 2TB model, which is a good price for such a large SSD.
Sesse
Posts: 300
Joined: Mon Apr 30, 2018 11:51 pm

Re: 7-men Syzygy attempt

Post by Sesse »

I still wish we had a 7-man-capable version of Fathom :-/
User avatar
Nordlandia
Posts: 2821
Joined: Fri Sep 25, 2015 9:38 pm
Location: Sortland, Norway

Re: 7-men Syzygy attempt

Post by Nordlandia »

yurikvelo wrote: Wed Nov 28, 2018 10:25 am
Nordlandia wrote: Wed Nov 28, 2018 8:03 am Is it waste to purchase 10 TB HDD for storing WDL for simple adjudication with cutechess?

The idea is that engines probe 5-men during search + automatic 7-piece adjudication once 7 pieces remains on the board.

It will speed up endgames somewhat compared to 6-piece adjudication. But is worth the investment for just passive adjudication?
Syzygy author explains that WDL is not working correctly without WDZ. Compression scheme require all files (including 3-4-5-6-men)
I would love to make a torrent and add to my http://oics.olympuschess.com/tracker site. However I do not have the storage available to even host a local copy as of yet.
full Nalimov-6 set couldn't be converted to *.torrent, because of limit, so it was split in parts and uploaded to rutracker.org
"Syzygy author explains that WDL is not working correctly without WDZ. Compression scheme require all files (including 3-4-5-6-men)"

I don't get it. DTZ is only probed at root.

I'm downloading RPP vs RP WDL and my initial plan is that Stockfish 10 might hold this endgame by consulting 7-piece egtb while H6 is only limited to 6-piece consulting.

This position is John Nunn's test position for rook endgames, only that it's white to move this time.

[d]8/ppp1rkpp/5p2/8/8/8/PPPR1KPP/8 w - - 0 1

So what is the likely outcome if SF 10 probes 6-piece + 7-piece egtb while H6 is limited to 6-man?

Is RPP vs RP egtb knowledge enough to turn the tide into draw?

Position is evaluated at ~ -1.00
jdart
Posts: 4378
Joined: Fri Mar 10, 2006 5:23 am
Location: http://www.arasanchess.org

Re: 7-men Syzygy attempt

Post by jdart »

Sesse wrote: Thu Nov 29, 2018 3:21 pm I still wish we had a 7-man-capable version of Fathom :-/
I have looked recently at merging the CFish version of tbprobe.c into my fork of Fathom, but there a lot of diffs, so that it is kind of a tall order. And I don't have 7-man bases to test it with. Pull requests are welcome, though.

--Jon
syzygy
Posts: 5647
Joined: Tue Feb 28, 2012 11:56 pm

Re: 7-men Syzygy attempt

Post by syzygy »

yurikvelo wrote: Wed Nov 28, 2018 10:25 am
Nordlandia wrote: Wed Nov 28, 2018 8:03 am Is it waste to purchase 10 TB HDD for storing WDL for simple adjudication with cutechess?

The idea is that engines probe 5-men during search + automatic 7-piece adjudication once 7 pieces remains on the board.

It will speed up endgames somewhat compared to 6-piece adjudication. But is worth the investment for just passive adjudication?
Syzygy author explains that WDL is not working correctly without WDZ. Compression scheme require all files (including 3-4-5-6-men)
I don't know where you read that, but WDZ does not exist.

WDL for N pieces requires WDL for (N-1) pieces and below.
DTZ for N pieces requires WDL for N pieces.

So WDL without DTZ is perfectly possible. As far as cursed/blessed wins/losses are concerned (so whether the 50-move rule makes a difference), the answers you get will be accurate only if you probe immediately after a capture or pawn move.

When adjudicating 7-piece positions, you will probe only immediately after the capture that brought the game from 8 to 7 pieces. So WDL will give the correct answer.
User avatar
Ozymandias
Posts: 1536
Joined: Sun Oct 25, 2009 2:30 am

Re: 7-men Syzygy attempt

Post by Ozymandias »

syzygy wrote: Sat Dec 01, 2018 1:27 pmWhen adjudicating 7-piece positions, you will probe only immediately after the capture that brought the game from 8 to 7 pieces. So WDL will give the correct answer.
In that scenario, I take it we can make do without DTZ completely, as well as without 6 men (and below) sets (either WDL or DTZ).
syzygy
Posts: 5647
Joined: Tue Feb 28, 2012 11:56 pm

Re: 7-men Syzygy attempt

Post by syzygy »

Ozymandias wrote: Sat Dec 01, 2018 2:21 pm
syzygy wrote: Sat Dec 01, 2018 1:27 pmWhen adjudicating 7-piece positions, you will probe only immediately after the capture that brought the game from 8 to 7 pieces. So WDL will give the correct answer.
In that scenario, I take it we can make do without DTZ completely, as well as without 6 men (and below) sets (either WDL or DTZ).
Please read carefully:
WDL for N pieces requires WDL for (N-1) pieces and below.
If you want to understand why, read the code. If you don't want to read the code, just accept what I wrote.
User avatar
Ozymandias
Posts: 1536
Joined: Sun Oct 25, 2009 2:30 am

Re: 7-men Syzygy attempt

Post by Ozymandias »

syzygy wrote: Sat Dec 01, 2018 2:27 pm
WDL for N pieces requires WDL for (N-1) pieces and below.
Got it, thanks.
User avatar
Nordlandia
Posts: 2821
Joined: Fri Sep 25, 2015 9:38 pm
Location: Sortland, Norway

Re: 7-men Syzygy attempt

Post by Nordlandia »

I created this thread on cutechess github page for sorting tings out.

https://github.com/cutechess/cutechess/issues/461

It seems cutechess from time to time adjudicates 7-man positions.

With KBPPvKPP WDL set on M.2 SDD, game continues when wK captures the pawn on h-file. What's going on?

For example i tried this position: [d]2k5/1p6/pP6/P7/8/B6K/7p/8 w - -
syzygy
Posts: 5647
Joined: Tue Feb 28, 2012 11:56 pm

Re: 7-men Syzygy attempt

Post by syzygy »

Ozymandias wrote: Sat Dec 01, 2018 3:32 pm
syzygy wrote: Sat Dec 01, 2018 2:27 pm
WDL for N pieces requires WDL for (N-1) pieces and below.
Got it, thanks.
What is not needed for WDL are the tables that can be reached through promotions.

So if you are only interested in KRPPvKRP and only need WDL, then you must download just KRPPvKRP and all 6-piece WDL TBs (or at least those that can be reached through captures, but you're not going to gain much by being selective with 6-piece tables if you are downloading the much bigger 7-piece tables). You don't need all KRXYvKRZ if you are only interested in adjudicating KRPPvKRP positions or in analysing positions for which KRPPvKRP is important.

(But if you want to play out KRPPvKRP positions to mate reliably, you need all the tables that can be reached through promotion and the corresponding DTZ tables.)