8-piece endgame tablebases - first findings and interview!

[AdminX]

[quote="Chessbase"]
Endgame tablebases can only be described as "god chess" as they are perfect information on all positions they cover, and the terms 'better' or 'worse' cease to have any meaning. Marc Bourzutschky has trained and resolved a large number of the 8-piece tablebase positions and shared his results and findings, as well as insights in a video interview.

https://en.chessbase.com/post/8-piece-e ... -interview

[Ajedrecista]

Hello Ted:

Just for the record, there was a thread about the longest checkmate and 8-man EGTB few months ago:

The longest checkmate question

I pasted two links from other person at that topic. The checkmate in 584 moves was briefly discussed:

Re: The longest checkmate question.

[...]

https://www.arves.org/arves/index.php/e ... plorations

https://www.arves.org/arves/index.php/e ... n-endgames

[...]

Enjoy!

Regards from Spain.

Ajedrecista.

[kasinp]

Endgame tablebases can only be described as "god chess" as they are perfect information on all positions they cover, and the terms 'better' or 'worse' cease to have any meaning. Marc Bourzutschky has trained and resolved a large number of the 8-piece tablebase positions and shared his results and findings, as well as insights in a video interview.

https://en.chessbase.com/post/8-piece-e ... -interview

Inhumane chess?

[Collingwood]

There's a won position in the article that was tested on an NN clone engine with 6-man TB for a minute or so. It got the right first move but only with slight advantage.

The real question is whether SF with 7-man TB run for a long time can do better. If anyone here can try that out, please post the results.

[yurikvelo]

There's a won position in the article that was tested on an NN clone engine with 6-man TB for a minute or so. It got the right first move but only with slight advantage.

Will it secure win in actual gameplay?

[syzygy]

There's a won position in the article that was tested on an NN clone engine with 6-man TB for a minute or so. It got the right first move but only with slight advantage.

Will it secure win in actual gameplay?

50-move rule. An engine should be able to secure the draw it starts with

[jp]

Will it secure win in actual gameplay?

50-move rule. An engine should be able to secure the draw it starts with

As I've mentioned previously, I do think it's interesting seeing how engines go on such positions at very high depths with the 50-move rule turned off (i.e. with the code modified and then recompiled). As someone else pointed out previously, even a win in under but close to 50 moves is difficult for engines to find with the 50-move rule turned on. (It distorts their search tree.)

[syzygy]

Will it secure win in actual gameplay?

50-move rule. An engine should be able to secure the draw it starts with

As I've mentioned previously, I do think it's interesting seeing how engines go on such positions at very high depths with the 50-move rule turned off (i.e. with the code modified and then recompiled). As someone else pointed out previously, even a win in under but close to 50 moves is difficult for engines to find with the 50-move rule turned on. (It distorts their search tree.)

No need to recompile, just feed the engine each position with the movecounter set to 0 if you want to know whether it can win in "actual gameplay" in "chess without 50-move rule".

[jp]

No need to recompile, just feed the engine each position with the movecounter set to 0 if you want to know whether it can win in "actual gameplay" in "chess without 50-move rule".

No, "actual gameplay" is, to my mind, making it too easy (in general). The engine might (in a given example) just play the right moves by accident, all the while not knowing it's winning or thinking it's winning but having no idea about the best defensive moves.

What I'd like to see is whether an engine can calculate through (for some given example) and see the winning mainline with a winning eval, and at what depth this occurs.

[Lazy_Frank]

For me interesting is question: how long is longest forced win DTM for n-TB.
Someone may suggest: if 8-TB longest mate took 584 moves, then 9-TB can be ~ 584*2=1168 moves, 10-TB ~ 584*4=2336 moves and so on.
But i don't think so it works like than.
Why?
My suggestion is actually simple:
Side with advantage tend to get maximum mobility before conversion (capture or mate).

FEN "R7/8/8/8/7q/2K1B2p/7P/2Bk4 w - - 0 1"
Zeroing at move 584w.
Mate at move 592w.

White maximum mobility: 40 moves at move 256, 39 moves at move 489.
Black maximum mobility: 32 moves at move 103, 31 moves at move 177.

In other words: mate distance for side with advantage depends of how fast (or slow) can get maximum pieces mobility.

I don't think any forced win (DTM for any TB) ever exceeds 1000 moves.
Even 594 is a very big number.

But i need a patience ... to get answer.