KPPKPP and the 'incomplete tablebase problem'

[Mig]

Is it essential to have all the 3 vs 3 bases installed in order to get accurate performance from the KPPKPP positions? That is, in order for it to display the number of moves to mate for different moves, etc. This would make sense since KPPKPP can become any other 3 vs 3 position, but I wanted to check. Are they all underpromotion aware? My various engines seem to do fine announcing mates after a few seconds of calculation, but they clearly suffer from not having all the 3 vs 3 bases. I'm curious at how much of a practical problem this is or if engines are smart enough to figure things out if some bases are missing. Would getting just the various queen bases help with this issue?

Thanks, Mig

http://www.chessninja.com

[guyhaw]

Mig: interesting to see you turning up on this board.
Aaron Tay discusses the 'incomplete EGT holding' problem at http://www.aarontay.per.sg/Winboard/egtb.html. The consequences of having a 'hanging EGT' rather depend on the intelligence of your chess engine.
The position r7/2K1P3/8/4R3/2k5/8/8/8 w - - 0 1 which he uses may indicate the intelligence of any specific engine in the context of a present KRPKR EGT but an absent KQRKR EGT. Does it go for P=Q although it cannot get a definitive EGT-depth for the resultant position, or does it prefer, e.g., 1.Re4+.
Aaron thinks that CRAFTY and YACE are clever enough to handle the 'missing EGT' problem [and presumably they still are] but that AMY, COMET, FRITZ, GANDALF, THE CRAZY BISHOP and TIGER are not [but might be ok now].
You will be able to see the DTM depth of a KPPKPP position using just the KPPKPP EGT - with WILHELM for example.
Others may have more detail on the behaviours of specific chess engines as of now: I think there was discussion of this much earlier on this board.
g

[Mig]

Thanks for the reply. Yah, I saw that page, but was wondering if the complexities of promotion and underpromotion opened a different can of worms. That is, not having 5-man and other smaller bases is a clear problem. But does not having, say, KBBKBB screw up KPPKPP results? Or would having KQPKPP KQPKQP KQQKQP and KQQKQQ be enough for effective usage? Is this also engine dependent? Some engines, even top ones, don't 'do' underpromotion, which might actually make things simpler. Or not, if this bargles up tablebase results.

I'm doing an article on tbs and freeware and software in general that I want to include practical guides. Downloading the terabyte isn't practical. So I'm wondering if having a few 6-man bases with pawns can actually make things worse when it comes to engine play and analysis. I wouldn't want to recommend just grabbing a few of the useful-looking ones if that's the case.

Saludos, Mig

[guyhaw]

Not having KBBKBB would put the KPPKPP results at risk from a creation process not using KBBKBB, but the results would be much more at risk from not having something 'one remove' (rather than 4 removes) from KPPKPP, and something more likely to be relevant like KNPKPP.
It rather depends on whether your 'intended usage' is in the context of the pursuit of perfect knowledge, the analysis of Studies and Maximals, or just the study of games. Tim Krabbe has a website (and even a book) on underpromotions required in games: they are very, very few but always fascinating. Underpromos turn up much more in Studies where they are 'a theme', and often as the reason for maxDTM positions.
I think you might experiment with a few positions and more than one engine to see the effect.
g

[Wulff]

Hi!

Just tested the position above with all 3+4 man and krpkr available to Gandalf. It wants to promote to a queen, and reports a mate in 15 after eq=Q Rxe8 Rxe8. How is the problem supposed to show up?? Do I need to remove the 3 & 4 man bases too, and only have krpkr available, or am I completely missing the point here ?

Additional question: might that be a GUI problem, if the GUI does the TB access instead ?

Edited, I accidently wrote 3+5 man, it is 3+4 man and krpkr.

[Arpad Rusz]

Today the incomplete database problem never occur. You can get perfect results with the KPPKPP even with no 5men databases installed. The engine tries to calculate if a position from a missing database is reached, but every KPPKPP position is correctly evaluated.
So even having only some of the 6men databases (e.g. KRPPKR, KRPKRP) helps a lot. Probably the important ones weights only around 100GB.

[guyhaw]

I think Wulf is missing the point. The scenario to test is the one where the KRPKR EGT is available but the KQRKR EGT is not available.

I'd be surprised if Arpad is right: I think some engines are 'tricked out', and more likely in earlier versions of these engines, by the fact that [as in the above] an EGT is available but a 'later' one, e.g. KQRKR vis-a-vis KRPKR, is not.

g

[Wulff]

Hello Guy!

I think Wulf is missing the point. The scenario to test is the one where the KRPKR EGT is available but the KQRKR EGT is not available.

I'd be surprised if Arpad is right: I think some engines are 'tricked out', and more likely in earlier versions of these engines, by the fact that [as in the above] an EGT is available but a 'later' one, e.g. KQRKR vis-a-vis KRPKR, is not.

g

That is excactly what I tried, all 3+4 man TB were available, along with krpkr. Gandalf wants to promote to a Queen, and expects the opponent to exchange the rook for the queen, yielding a mate in 15.

I think it is not a problem because Gandalf sees that white can win faster than 15 moves with the queen on the board, whereas a GUI does not do a search, thus it would not see it.

Maybe I'm still barking up the wrong tree.

[guyhaw]

I misread your '3-4' as '3-5': I'll blame jetlag at this end
Ok - it sounds like GANDALF is being intelligent enough to overstride the fact that it does not have the KQRKR EGT.
g

[Arpad Rusz]

It seems that the incomplete tablebases are good only for analysing with an engine. The engine can complete the list with evaluation of the moves analysing itself the missing positions which occur after transformation.
But when is playing a game, the GUI may choose another move even when the engine knows that the Q transformation is better.
So my previous post is true only for endgame analysis.
A solution for this problem could be the following advice: if you have a tablebase with P you should have the same tablebase with Q instead P too.