46848

[glennwilson]

http://chesstempo.com/chess-problems/46848
46848 should accept 1. Re3 as a solution (or at least an alternate) but it is not even listed.  Fritz and Nalimov tables agree.  These seems different that the other move generator bug I've seen mentioned (no lines are listed twice but this line is omitted).

[drahacikfm]

Yes, any future generator runs should have the endgame tablebases hooked up to Toga.  Does Toga allow that?

And maybe a run-through should be done on all positions in existing problems that have 8 pieces or less.  The tablebases have only 5 pieces (or 6 if you have tons of disk space) but are used to evaluate the variations, which starting from 8 could easily get down to 5 very quickly.

Note that "pieces" when talking about tablebases includes kings and pawns.  The problem mentioned here has 6 "pieces".

In general though on Chess Tempo now I find very few problems that have winning moves that are marked wrong.  So few that it's not much of a worry at all.  A big improvement from May when I first came here.

For those of you who are new, in May you had to find the best move, and you were marked wrong for all other moves!  Then later, you had to play a move that evaluated at +2.75 or more to not get marked wrong.  Now it's +1.75, which seems to be about right for most problems.

[glennwilson]

To clarify, Fritz 6 finds this "instantly" w/o using the Nalimov tables.  It is not that hard (although it was plenty hard enough for me).  So, while using Nalimov tables as appropriate would be great I suspect that is not the source of this bug, per se.  Of course, I am guessing a bit at the source of the error....(I am guessing that the engine found this move but it somehow did not show up in the output or was inappropriately parsed out somehow....)...

[richard]

Toga supports endgame tablebases via the Scorpio bitbases by Daniel Shawul , however I'm not sure if this is currently supported on non-windows platforms.  There are probably a few positions that are certain wins that toga doesn't find due to not having tablebase lookup (although in practice engines seem to be fairly good with endgames even without tablebases, I've seen estimates of tablebases only adding about +50 elo to an engine's rating). I'll see if I can get the scorpio support going..

Regards,
Richard.

[drahacikfm]

Richard:

I just got this problem wrong, playing the completely winning 1.Re3, and lost 11 rating points.  Too bad I didn't remember this forum discussion.

Anyway, there is really an urgent need for manual disabling of problems.  It's extremely frustrating to play a forced win and get marked wrong, especially when you lose a lot more for wrong problems than you get for correct problems.

There aren't many of these problems, but it's really needed to just disable them now.  No coding is needed, I don't think.  You mentioned you don't have the code to keep the problems disabled after a future generator run.  Well, pencil and paper can keep a list of problems that should remain disabled, without any coding

Rybka, with NO tablebases, says +4.92 for 1.Re3 after 5 seconds of thinking.  I don't know what the problem is with Toga.  Tablebases are not needed for an engine to see that 1.Re3 wins.  And certainly I hope we don't have to wait for a new generator run with tablebases before the endgame problems with errors get disabled.

[richard]

Ok, this is the very first problem to get "manually disabled".  I've added the code to check for manual disabling to the updater, so hopefully it will not get automatically added back next time (I'll probably add tablebase support to the next run anyway so It shouldn't matter either way).  Toga does eventually see Re3, but it takes several minutes, so it is probably not an outright bug but probably an issue of candidate move calculation not being very optimal in this case (compared to Rybka).

Richard.

[drahacikfm]

Great, thanks!   Maybe we can have a new topic where we give you problem numbers that need to be manually disabled.  I don't think there will be many.

The other one I am aware of right now is 41438, analyzed by milly and me to have several moves winning by force on move 1 that are not allowed alternates.

Now the trick I have not figured out yet is how to get problems disabled before I do them!

[richard]

Ok, I've killed 41438 also.

I had a quick play around with using star ratings of high rated players to auto-kill bad problems and there is still too much disagreement on which problems are 1 star worthy to use this method at the moment.  I could add a third "problem comment type", for "this problem should be killed immediately" but that is likely to require careful filtering anyway, so posting them to the forum is probably still the best way of getting rid of the truly atrocious problems still lurking in the set.  Hopefully some can be found by others before you have to see them :-)

Richard.