Gaviota Tablebases vs. Nalimov

[Dirt]

As far as I have read the Gaviota and Nalimov tablebase contain exactly the same information, at least for now. Is this correct?

[michiguel]

As far as I have read the Gaviota and Nalimov tablebase contain exactly the same information, at least for now. Is this correct?

I never used the Nalimov tables, but, as far as read, yes. Both are distance to mate and take care of ep info. I do it on the fly, I do not know how Nalimov does that. However, I plan to expand Gaviota's. At one point I will incorporate castling and allow to read Distance to zeroing 50, if the user wants it.

Miguel

[ernest]

and allow to read Distance to zeroing 50, if the user wants it.

Distance to zeroing 50 is quite useless, if you want to find the shortest path to Mate, with the 50-move rule.
To do that, for instance with KNNKP, you need to construct the 100 KNNKP files (100 ply to zeroing, to 1 ply to zeroing) for White and the 100 KNNKP files (100 ply to zeroing, to 1 ply to zeroing) for Black.
A friend of mine just did that, but of course today's chess programs cannot use those 200+ KNNKP files (including the files needed after capture or promotion in the KNNKP set). He had to write a special purpose program, to solve that shortest path to Mate (with the 50-move rule) problem.

[michiguel]

and allow to read Distance to zeroing 50, if the user wants it.

Distance to zeroing 50 is quite useless, if you want to find the shortest path to Mate, with the 50-move rule.
To do that, for instance with KNNKP, you need to construct the 100 KNNKP files (100 ply to zeroing, to 1 ply to zeroing) for White and the 100 KNNKP files (100 ply to zeroing, to 1 ply to zeroing) for Black.
A friend of mine just did that, but of course today's chess programs cannot use those 200+ KNNKP files (including the files needed after capture or promotion in the KNNKP set). He had to write a special purpose program, to solve that shortest path to Mate (with the 50-move rule) problem.

I believe they would be quite powerful if used in combination with DTM

Miguel

[ernest]

I believe they would be quite powerful if used in combination with DTM

Indeed, how so?
Can you give an example?

[michiguel]

I believe they would be quite powerful if used in combination with DTM

Indeed, how so?
Can you give an example?

There are positions in which it is mate in 60, but it is actually a draw because you have to go through 50 moves with no captures or pawn moves. DTM is blind to that, and it will trade to finally end in an endgame that is a draw. If DTM sees mate in 60, it should consult DTZ 50 first to see whether it is a draw or not before trading. Just an example.

Miguel

[hgm]

and allow to read Distance to zeroing 50, if the user wants it.

Distance to zeroing 50 is quite useless, if you want to find the shortest path to Mate, with the 50-move rule.
To do that, for instance with KNNKP, you need to construct the 100 KNNKP files (100 ply to zeroing, to 1 ply to zeroing) for White and the 100 KNNKP files (100 ply to zeroing, to 1 ply to zeroing) for Black.
A friend of mine just did that, but of course today's chess programs cannot use those 200+ KNNKP files (including the files needed after capture or promotion in the KNNKP set). He had to write a special purpose program, to solve that shortest path to Mate (with the 50-move rule) problem.

I don't understand this. I believe the problem you point out exists in DTR. But I cannot see it in DTZ50. If you have determined all wins within the 50-move rule of all successor EGTs, then building a normal DTZ table seeded by those winning zeroing events, and discarding the positions with DTZ > 50, should give you what you want. It would give you the shortest path to a winning conversion.

That path could not be shorter if you started it with a non-zero 50-move counter, and were in a greater hurry. You do already the best you possibly can, and if that is not below 50 when added to the current counter, it is a draw, and using another EGT cannot change that. Plus that there really is no reason why you would have a non-zero 50-move counter when you started the longest path to winning zeroing. It would mean you wasted time, and why would you do that?