High-rated ambiguous problems.

[richard]

Tmr,

Unfortunately I think that is the case.  Previously the mates out numbered the non-mates so it was mostly the mates people complained about, now that a) there are more non-mates and b) the mates are much better than the old set (although not yet perfect) the alternative material winning positions that got past the filter (a lot more than I expected) are getting attention.  The new problem selection while definitely an improvement also exposes more of the problematic positions, especially to higher rated players who are much better at seeing the issues.

I estimate that > 70 % of the problems being reported as having alternative winning lines are due to a single problem in the generator. I now have a new version of the generator which will address a large number of the alternative winning line issues, as soon as I've tested it a little more I'll start a verification run on the current problem set (which takes a few weeks unfortunately. I've been meaning to buy a new multi-core machine to dedicate to problem set crunching so I'll try and get my hands on that which will cut down the verification time a bit).

Regards,
Richard.

[drahacikfm]

11867:  Four different first moves evaluate more than +3.0 by Rybka.  After all, White is up a whole piece in the starting position.

[drahacikfm]

24694:  After the problem's 1....Rxg3 2.Rcd1, Rybka gives:

2...Rxh3 +5.97
2...Rxf1 +5.97
2...Qh4 +5.77
2...Bxh3 +4.62
etc.

And the problem forces you to play 2...Qh4.

[richard]

24694 is a bit unfortunate in that toga originally sees Qh4 as 3 pawns better than the next move.  It then switches to Rxf1 but then after a bit longer switches back to Qh4, but by only a small margin after close to 10 mins Rxf1 overtakes Qh4 again, but only by 1/100th of a pawn.

Toga really disagrees with Rybka assessment of Rxh3 which just scrapes in as 4th best move from Toga's point of view.

I'd be really interested if anyone else reading this thread could plug problem 24694 into fritz and see what it says about this position.  If toga is getting evaluations wildly wrong I'll have to look more seriously  at switching engines.
Toga's analysis after around 10 mins:
2..Rxf1 +8.42
2. Qh4 +8.41
3. Bxh3 +5.71
4. Rxh3 +4.54

Of course the best 4 moves all win here so again the alternative winning line code never kicked in (due to the "pre-tactic evaluation" bug).  I haven't run this one past the new generator but I expect it will reject it due to having too man alternative lines.

[drahacikfm]

Fritz 9, depth 15, 3 minutes on Pentium D dual core 3.2 GHz:

1. -+  (-12.55): 41...Rxf1
2. -+  (-10.67): 41...Qh4
3. -+  (-10.10): 41...Rxh3+
4. -+  (-9.77): 41...Bxh3

Fritz has Rxh3 as not much worse than Qh4, but Toga says it's much worse.

I ran Rybka for 6 minutes this time, same move ordering as before, only difference is that now the top 3 moves are all the same at -5.97 and Bxh3 is -5.35.

Hard to compare on different CPUs.  Maybe your computer is much faster and Toga is looking much deeper than Fritz or Rybka, so that Rxh3 is indeed not so good.  Or maybe your computer is much slower and with longer search Rxh3 would go up.

[richard]

Toga's dislike for Rxh3 doesn't seem CPU related, it hates it from early on and then keeps hating it.  My CPU is in the same ballpark as your fritz hardware. I tried running toga for longer to see if it would find what the others were seeing, but it kept on hating :-)


EDIT: I ran toga again on this for 25 mins and FINALLY it liked Rxh3 , scores:
Rxh3 9.14
Rxf1 8.92
Qh4 8.89
Bxh3 7.34

This seems much more in line with Rybka and Fritz now, I wonder why Toga took so long to come up with a similar eval for Rxh3.


It is somewhat concerning from a UCI portability point of view how different the actual evaluation amounts are between the three engines, you said Rybka was conservative but Fritz seems particularly non-conservative. I had only tried with one other free engine in the past and it didn't have evaluations that differed much from Toga's, I'd have to be very careful switching to another engine.

It might be interesting at some stage to do a verification run with a different engine and see how much of the current set it would disagree with.

Regards,
Richard.

[tama]

31645:  1.Qxe5+ is a knight-up endgame, easy win, more than +3.0 Rybka.

U are Incorrect because after winning the Kight the correct way the queens are still on the board and in fact wight will win the exchange or he will give a mate very soon.

[xjamesbx]

31645:  1.Qxe5+ is a knight-up endgame, easy win, more than +3.0 Rybka.

U are Incorrect because after winning the Kight the correct way the queens are still on the board and in fact wight will win the exchange or he will give a mate very soon.

Er, what?

Hopefully I didn't misunderstand your somewhat broken English, but you are given the problem from the eyes of white.  After Qxe5, white is definitely up a piece, no matter what happens.  And after that, white's going to be able to trade the queens off regardless of black's next move.

At that point, white has an easily won game, just as drahacikfm said.

[drahacikfm]

Yes, I did not say that 1.Qxe5+ is better than the move the problem wants you to make.  I only said that it is completely winning.  The goal of the problems on this site is to not have more than one completely winning move.  Because then it is an "ambiguous problem".

[drahacikfm]

19329:  Alternate move 1...Rxe4 is completely winning, up lots of material. No need to check Rybka to see that.

[tama]

This problem is in fact an 1800 problem. It is a very easy Mate in one.

I know for a fact this problem should be rated around 1200.

Can someone back me up?

[richard]

This problem is in fact an 1800 problem. It is a very easy Mate in one.

I know for a fact this problem should be rated around 1200.

Can someone back me up?

It is indeed a mate in 1, but it seems it is not quite that obvious to everyone.  Lots of people are playing for a mate in 2 with  1...Qf4   2. Kh3 Qg4

Richard.

[tama]

Ok yes but y?

one of the main runs for solving tactics is If u c a good move look for a better one! I didn't c the mate in 2, the first check i saw was QxP++.

[richard]

Ok yes but y?

one of the main runs for solving tactics is If u c a good move look for a better one! I didn't c the mate in 2, the first check i saw was QxP++.

Hard to say why one person sees a move and others don't...I think players have different blind spots, and different strengths.  Maybe that type of position is one you do particular well at (a number of higher rated players than you played Qf4, so maybe this is an area of strength for you :-)  ).

Richard.

[drahacikfm]

14201:  Very nice problem. 1.Qxd6 is an unusual tactic.  But after the problem's 1...Qd7, there seem to be 3 winning moves:  The problem's 2.Qxe7, plus 2.Qxd7 and 2.Qxf6+.  Hope the new generator keeps this problem, with the new feature that allows "takebacks" for the winning alternate moves.