Forums

Tablebase for stockfish question

Sort:
Pl4y_4_fun

Hi,

I`ve just downloaded syzygy tablebases for 6 pieces (and less, they were included). I downloaded both dtz and wdl and configured stockfish to read from that directory. And now I started testing it.

Easy position KBNvK is automatically +152 with 14 TB hits:

5f23a5864d84e5bdc9141bd2de56213b.png

And that is actually a checkmate in 27 moves:
800807f7b84fcf919978ab8ea378ff89.png

Without those tablebases it was +3 for first 3 minutes of analysis and later stockfish was finding a checkmate.

But now I try to test an actual 6-piece position, that is present in my folder:

6a4e4c6a449e29c7f102ddeddfc1435b.png

And the result is:

06c95480b00d529149632ea916a043b7.png

Although theoretically it`s also a win:

0feb7af6a6d8b55a875e7b4cb12e4e35.png

Analysis continues and it has more and more TBHits but it doesn`t see, that it`s a clear win (not so clear for me, but should be clear for TB).

So what is the reason for that? When the stockfish will finally figure out, that it`s a win? Will it actually work in situation, where it`s 7+ piece and in the end of a line is 6-piece position?

taychoe

Can you post the settings for Stockfish 14.1 ?  You may have to change some settings in Arena specific to the engine you are using.

Pl4y_4_fun

4aa3a2f11458e452e75df42c5c4edca7.png

and here I tried with both clicked and unclicked tablebases:

49df3e7244ce8dbc49b2b439672d5db8.png

 

taychoe

You have two subfolders inside E:\TB, so on the SyzygyPath setting of Stockfish, you can try "E:\TB\dtz;E:\TB\wdl" without the quotes.  See if that works.  

Pl4y_4_fun

I`ve copied all files into E:\TB 

Pl4y_4_fun

2ca282e9b4c2435d1d42f8bf94aba224.png

Pl4y_4_fun

And I just tried, that doesn`t work even for easy KBNvK situation, so probably it doesn`t understand that expression.

taychoe

It appears the tablebase is being read, but you probably need to move the pieces further on before Stockfish will see a clear win, or a mate. 

taychoe

According some internet sources I've been reading, syzygy bases use dtz (distance to zero), the number of moves to get to a point where a capture or a pawn move would occur (thus resetting the 50-move counter to zero) and still have a winning position.  So in your case, Stockfish is satisfied with Ke5 or Kd5 and doesn't investigate further. 

Perhaps if you uncheck the tickbox that says "Syzygy50MoveRule", then Stockfish will continue investigating beyond what the syzygy tablebase recommends and it should be able to provide better scores.

 

James_J_Henderson

@Pl4y_4_fun Sorry I'm late to the discussion.  If Stockfish is properly reading your Syzygy tablebase directory, it will instantly evaluate that position as +152.65 for White.  Here's what I get running Stockfish from the command line:

Stockfish 290122 by the Stockfish developers (see AUTHORS file)
ucinewgame
setoption name MultiPV value 1
setoption name Hash value 1024
setoption name Threads value 1
setoption name SyzygyPath value /usr/local/Six_Man_TB
info string Found 510 tablebases
setoption name Use NNUE value false
position fen 4k3/8/8/5B2/r2BK3/6N1/8/8 w - - 0 1
go depth 1
info string classical evaluation enabled
info depth 1 seldepth 1 multipv 1 score cp 15265 nodes 15 nps 7500 tbhits 17 time 2 pv e4d5
bestmove e4d5

So, even at depth 1, Stockfish sees a win (+15265 centipawns, with 17 tablebase hits, after 2 milliseconds).

I am not familiar with the GUI you're using as a frontend to Stockfish, but in your engine management window, I don't think the Winboard features are relevant, as you're using UCI, not Winboard.  One possibility is that you need the .txt files as well as the .rtbz and .rtbw files in your Syzygy directory.  You might try copying the Syzygy files to a directory without any subdirectories, containing only .txt, .rtbz, and .rtbw files.  One other possibility is that your Syzygy directory is missing some of the five-piece files (or Stockfish isn't seeing them), such as KBBvKR.rtbz, that could result from your six-piece position after one of the pieces gets captured.  This last possibility might be why there wasn't a problem with your king-knight-bishop position but there was a problem with your six-piece position.

Pl4y_4_fun

@James_J_Henderson Thanks for your response. I have discovered, that I was missing some 5-piece tablebases. I don`t know how that happened, but I downloaded them and now everything works great and evaluation is +15265 centipawns even with depth set to 1.