It's on my todo list and has been for a long time. But I'm moving it up close to the top because I AT LEAST want to understand exactly what is happening here since it could have a real impact on the strength of the program and also because many if not most Komodo users would want to see better behavior. As a Komodo user myself I don't like to see that.lucasart wrote:I'm glad I'm not the only one to have this problem. I never bothered to investigate it as it never had anymore than a cosmetic effect (still plays the right move). Something to put far away in the todo list...Don wrote:There is some bug in Komodo that also makes it do that sort of thing. It will see a mate but report it as something like a mate taking 100 moves or more. I think it is a hash table interaction with the maximum mate depth Komodo can report - so with the hash table Komodo perhaps strings together a mate in some ridiculous number of ply and then moves on to the next iteration.hgm wrote:Something is a bit suspect here: just before it finds the mating line at 34 ply, with score +M23, it gets a '33+' line with a score of +150.93. I don't see how it could ever get such a score from normal (material + PST + mobility) evaluation, and it is not a mate score either. So it seems to have some pattern, and considering the root material, only KBNK patterns can match.
Best was to Recognize Know Endgames?
Moderators: hgm, Rebel, chrisw
-
Don
- Posts: 5106
- Joined: Tue Apr 29, 2008 4:27 pm
Re: Best was to Recognize Know Endgames?
Capital punishment would be more effective as a preventive measure if it were administered prior to the crime.
-
hgm
- Posts: 27790
- Joined: Fri Mar 10, 2006 10:06 am
- Location: Amsterdam
- Full name: H G Muller
Re: Best was to Recognize Know Endgames?
The "sepcial knowledge" here being that it would know which corner is the deadly one?syzygy wrote:Without special knowledge (and tablebases off) my engine takes 2 minutes to find a mate in 29 which then quickly goes down to mate in 21.
With special knowledge a mate in 25 is found in less than 8 seconds and mate in 21 in 14 seconds.
-
hgm
- Posts: 27790
- Joined: Fri Mar 10, 2006 10:06 am
- Location: Amsterdam
- Full name: H G Muller
Re: Best was to Recognize Know Endgames?
Don wrote:There is some bug in Komodo that also makes it do that sort of thing. It will see a mate but report it as something like a mate taking 100 moves or more. I think it is a hash table interaction with the maximum mate depth Komodo can report - so with the hash table Komodo perhaps strings together a mate in some ridiculous number of ply and then moves on to the next iteration.
I see that here in this ending - so I have a good test case and I will try to fix this glitch. Because of aspiration at the root I will get fail high scores that look like modest wins similar to the 150 pawn score you see here.
Well, 100 moves or more is sort of normal, and should not bring you out of the range of mate scores. Even whnen you want scores to fit in 16 bits you should have plenty of room for a few thousand mate scores. It would be truly remarkable if Komodo could see mates in over a thousand moves.
What do you mean with "maximum mate depth Komodo can report"? Is that where it switches from mate to cp notation in the protocol driver? It should be pretty obvious whether a cp score was intended as a mate score, when you know the internal score Komodo uses for checkmate.
To get a score of 15093 cp, as we see here in Quazar, requires a really strange mate score, or an unbelievable DTM. So I still doubt it can have a similar explanation.
-
Don
- Posts: 5106
- Joined: Tue Apr 29, 2008 4:27 pm
Re: Best was to Recognize Know Endgames?
If the score is high enough Komodo assumes it's a mate and converts it to mate score for UCI. In theory a score of 10 centipawns could be a mate in thousands of moves of course but no program would see those.hgm wrote:Don wrote:There is some bug in Komodo that also makes it do that sort of thing. It will see a mate but report it as something like a mate taking 100 moves or more. I think it is a hash table interaction with the maximum mate depth Komodo can report - so with the hash table Komodo perhaps strings together a mate in some ridiculous number of ply and then moves on to the next iteration.
I see that here in this ending - so I have a good test case and I will try to fix this glitch. Because of aspiration at the root I will get fail high scores that look like modest wins similar to the 150 pawn score you see here.
Well, 100 moves or more is sort of normal, and should not bring you out of the range of mate scores. Even whnen you want scores to fit in 16 bits you should have plenty of room for a few thousand mate scores. It would be truly remarkable if Komodo could see mates in over a thousand moves.
What do you mean with "maximum mate depth Komodo can report"? Is that where it switches from mate to cp notation in the protocol driver? It should be pretty obvious whether a cp score was intended as a mate score, when you know the internal score Komodo uses for checkmate.
It can also be aspiration window glitches. It may be bounded by a window in other words.
To get a score of 15093 cp, as we see here in Quazar, requires a really strange mate score, or an unbelievable DTM. So I still doubt it can have a similar explanation.
Capital punishment would be more effective as a preventive measure if it were administered prior to the crime.
-
hgm
- Posts: 27790
- Joined: Fri Mar 10, 2006 10:06 am
- Location: Amsterdam
- Full name: H G Muller
Re: Best was to Recognize Know Endgames?
I guess you would have to use some form of fail-hard for window bounds to transfer back to scores. But would it really make sense to set one of the limits of an aspiration window at 150 Pawns?
-
syzygy
- Posts: 5563
- Joined: Tue Feb 28, 2012 11:56 pm
Re: A clarification
A trivial modification to idea #2 does allow this, as already explained here.sje wrote:A clarification:
As I see it, there are two distinct material distribution ideas presented here.
1. The material distribution key is produced by Zobrist hash construction.
2. The material distribution key is produced by adding or subtracting a single bit with the bit position defined by the piece being added or deleted.
Idea #1 allows the use of traditional transposition table techniques, but the key itself doesn't have any encoded data.
Idea #2 doesn't allow the key to be used as a transposition table index, but it does contain clearly identifiable and useful data.
No, the hit rate is 99.9% so there is nothing to gain. (And if the hit rate were much lower there would still be nothing to gain.)A transposition table keyed by a material Zobrist key is much like a pawn structure table; both contain a lot of data which can be used not only from search to search, but also from game to game. Would it be worthwhile to save these tables after each program run and then reload them the next time around?
-
syzygy
- Posts: 5563
- Joined: Tue Feb 28, 2012 11:56 pm
Re: Best was to Recognize Know Endgames?
I have a bonus for getting the losing king closer to the right corner, and a penalty proportional to the distance between the two kings.hgm wrote:The "sepcial knowledge" here being that it would know which corner is the deadly one?syzygy wrote:Without special knowledge (and tablebases off) my engine takes 2 minutes to find a mate in 29 which then quickly goes down to mate in 21.
With special knowledge a mate in 25 is found in less than 8 seconds and mate in 21 in 14 seconds.
On this position my engine without tablebases and without special heuristics starts to report impossibly large mate values after 25 seconds (due to TT interactions). Mate in 45 after 42 seconds, mate in 33 after 94 seconds.Don wrote:Here is a KBNvsK position that requires 33 moves, which is the longest this should take with perfect play:
[d]K1k1B3/8/8/8/8/8/7N/8 w - - 0 1
I'm going to run this overnight - it should take a lot longer than the last one took.
Don
With special heuristics for KBNK it takes 31 seconds to get a mate in 45 and 61 seconds to get to mate in 33. No impossibly large mate values.
Just for fun the two PVs I get:
1.Ka7 Kd8 2.Ba4 Kc7 3.Nf3 Kd6 4.Kb6 Kd5 5.Bc2 Kd6 6.Bb3 Kd7 7.Kc5 Ke7 8.Kd4 Kf6 9.Ke4 Kg6 10.Ke5 Kg7 11.Kf5 Kh6 12.Kf6 Kh7 13.Ne5 Kh8 14.Bc4 Kh7 15.Ng6 Kh6 16.Bg8 Kh5 17.Ne5 Kh4 18.Kf5 Kg3 19.Ng4 Kg2 20.Bc4 Kf3 21.Bb5 Kg3 22.Bc6 Kh3 23.Kf4 Kh4 24.Be8 Kh3 25.Ne3 Kh4 26.Ng2+ Kh3 27.Kf3 Kh2 28.Kf2 Kh3 29.Bd7+ Kh2 30.Ne3 Kh1 31.Bf5 Kh2 32.Nf1+ Kh1 33.Be4#
1.Ka7 Kd8 2.Bb5 Kc7 3.Nf3 Kd6 4.Kb6 Kd5 5.Bd3 Kd6 6.Bc4 Kd7 7.Kc5 Ke7 8.Kc6 Kf6 9.Kd6 Kf5 10.Bd3+ Kf6 11.Ne5 Kg7 12.Ke7 Kg8 13.Kf6 Kh8 14.Bc4 Kh7 15.Ng6 Kh6 16.Bg8 Kh5 17.Ne5 Kh4 18.Kf5 Kg3 19.Ng4 Kg2 20.Bc4 Kf3 21.Bd3 Kg3 22.Be2 Kh4 23.Nf2 Kg3 24.Nd3 Kh3 25.Kf4 Kh4 26.Ne1 Kh3 27.Bg4+ Kh2 28.Kf3 Kg1 29.Bh3 Kh2 30.Kg4 Kg1 31.Kg3 Kh1 32.Bg2+ Kg1 33.Nf3#
-
sje
- Posts: 4675
- Joined: Mon Mar 13, 2006 7:43 pm
Table persistancy across program restarts
Actually, the higher the hit rate, then the more the gain. A more complete analysis would cover the per entry generation time and the whole table save/restore time.syzygy wrote:No, the hit rate is 99.9% so there is nothing to gain. (And if the hit rate were much lower there would still be nothing to gain.)sje wrote:A transposition table keyed by a material Zobrist key is much like a pawn structure table; both contain a lot of data which can be used not only from search to search, but also from game to game. Would it be worthwhile to save these tables after each program run and then reload them the next time around?
-
michiguel
- Posts: 6401
- Joined: Thu Mar 09, 2006 8:30 pm
- Location: Chicago, Illinois, USA
Re: Best was to Recognize Know Endgames?
I managed to improve this significantly with the following two changesmichiguel wrote:I let Gaviota run on this position (AMD Phenom(tm) II X4 965 Processor × 4 3.4 Ghz), hash = 4 Gb, of course, no TBs.Don wrote:Actually, this took less than 5 minutes to solve too.
Don wrote:Here is a KBNvsK position that requires 33 moves, which is the longest this should take with perfect play:
[d]K1k1B3/8/8/8/8/8/7N/8 w - - 0 1
I'm going to run this overnight - it should take a lot longer than the last one took.
Don
After 2 min 40 seconds (160.0 s in this log) a mate is announced
At depth 45, ~16 minutes, a mate under 50 is announced.Code: Select all
333150264 35 144.5 +5.85 1.Bh5 Kd7 2.Ng4 Ke6 3.Ne3 Ke5 4.Nc2 Ke4 5.Kb7 Ke5 6.Kc6 Ke4 7.Kc5 Ke5 8.Nd4 Kf6 9.Kd6 Kg7 10.Nf5+ Kf6 11.Ne7 Kg7 12.Ke5 Kh8 13.Kf6 Kh7 14.Bf7 Kh8 15.Ng6+ Kh7 16.Kg5 Kg7 17.Bb3 Kh7 18.Kf6 Kh6 19.Nf8 [>] 354260831 35 153.9 :-) 1.Bb5 354264379 35 153.9 :-) 1.Bb5 364415388 35 160.0 +Mat_69 1.Bb5 Kc7 2.Ka7 Kd6 3.Kb6 Kd5 4.Bc6+ Kd4 5.Nf3+ Kc4 6.Be4 Kb3 7.Ne5 Kc3 8.Kc5 Kb3 9.Kd4 Kb2 10.Kc4 Ka2 11.Kc3 Ka1 12.Ng4 Ka2 13.Bd3 Ka1 14.Ba6 Ka2 15.Kc2 Ka3 16.Kc3 [>]
At depth 60, ~98 min, finally mate in 33 is announced.Code: Select all
2158546628 45 981.1 +Mat_39 1.Ka7 Kd8 2.Bc6 Ke7 3.Ng4 Ke6 4.Be4 Ke7 5.Kb6 Kd6 6.Kb5 Ke7 7.Kc6 Ke6 8.Bg6 Ke7 9.Kd5 Kf8 10.Ke6 Kg7 11.Bb1 Kg8 12.Kf6 Kh8 13.Ba2 Kh7 14.Ne5 Kh8 15.Ng6+ Kh7 16.Bb1 Kg8 17.Ba2+ [>]
Code: Select all
12212789059 60 5902.6 +Mat_33 1.Bg6 Kd7 2.Kb7 Ke6 3.Kc6 Ke5 4.Bb1 Kd4 5.Nf3+ Ke3 6.Ng5 Kd4 7.Ne6+ Ke5 8.Kd7 Kf6 9.Kd6 Kf7 10.Ke5 Ke7 11.Ng5 Kf8 12.Kf6 Ke8 13.Bf5 Kf8 14.Be6 Ke8 15.Nf7 Kf8 16.Ne5 Ke8 17.Bf7+ Kd8 18.Ke6 Kc7 19.Be8 Kb6 20.Kd6 Ka5 21.Kc5 Ka6 22.Nc6 Kb7 23.Nd8+ Kc7 24.Nf7 Kb7 25.Kb5 Kc7 26.Ka6 Kc8 27.Kb6 Kb8 28.Bd7 Ka8 29.Bb5 Kb8 30.Ba6 Ka8 31.Ne5 [>] 12218733597 60: 5905.7 +Mat_33 1.Bg6 Kd7 2.Kb7 Ke6 3.Kc6 Ke5 4.Bb1 Kd4 5.Nf3+ Ke3 6.Ng5 Kd4 7.Ne6+ Ke5 8.Kd7 Kf6 9.Kd6 Kf7 10.Ke5 Ke7 11.Ng5 Kf8 12.Kf6 Ke8 13.Bf5 Kf8 14.Be6 Ke8 15.Nf7 Kf8 16.Ne5 Ke8 17.Bf7+ Kd8 18.Ke6 Kc7 19.Be8 Kb6 20.Kd6 Ka5 21.Kc5 Ka6 22.Nc6 Kb7 23.Nd8+ Kc7 24.Nf7 Kb7 25.Kb5 Kc7 26.Ka6 Kc8 27.Kb6 Kb8 28.Bd7 Ka8 29.Bb5 Kb8 30.Ba6 Ka8 31.Ne5 [>] 12232467096 61 5917.9 +Mat_33 1.Bg6 Kd7 2.Kb7 Ke6 3.Kc6 Ke5 4.Bb1 Kd4 5.Kd6 Kc4 6.Nf3 Kb4 7.Kd5 Kc3 8.Nd4 Kb2 9.Bh7 Kc3 10.Kc5 Kb2 11.Kc4 Ka1 12.Kc3 Ka2 13.Nb3 Ka3 14.Bb1 Ka4 15.Nd4 Ka5 16.Kc4 Kb6 17.Nb5 Kb7 18.Bf5 Kc6 19.Be6 Kb6 20.Bd7 Ka5 21.Nc7 Kb6 22.Ne8 Ka5 23.Kc5 Ka6 24.Nd6 Ka5 25.Bc6 Ka6 26.Bb5+ Ka7 27.Kc6 Kb8 28.Ba6 Ka7 29.Bf1 Kb8 30.Kb6 Ka8 [<] 12235965572 61: 5919.8 +Mat_33 1.Bg6 Kd7 2.Kb7 Ke6 3.Kc6 Ke5 4.Bb1 Kd4 5.Kd6 Kc4 6.Nf3 Kb4 7.Kd5 Kc3 8.Nd4 Kb2 9.Bh7 Kc3 10.Kc5 Kb2 11.Kc4 Ka1 12.Kc3 Ka2 13.Nb3 Ka3 14.Bb1 Ka4 15.Nd4 Ka5 16.Kc4 Kb6 17.Nb5 Kb7 18.Bf5 Kc6 19.Be6 Kb6 20.Bd7 Ka5 21.Nc7 Kb6 22.Ne8 Ka5 23.Kc5 Ka6 24.Nd6 Ka5 25.Bc6 Ka6 26.Bb5+ Ka7 27.Kc6 Kb8 28.Ba6 Ka7 29.Bf1 Kb8 30.Kb6 Ka8 [<]
Miguel
1) entries with mate values take priority in the hash table
2) Do not do LMR if beta is in the mate value range, that is, no reason to reduce when all moves lead to a checkmate.
1 --> makes sure mates are not overwritten in the hashtable
and 2 avoids the annoying jumping up and down in mate scores reported.
with that i got a mate in 2 min 20 sec, and mate it 33 in 5 min 20 sec.
Code: Select all
311424055 38 139.1 +5.87 1.Ka7 Kd8 2.Bc6 Ke7 3.Ng4 Ke6 4.Be4 Kf7
5.Bd5+ Ke7 6.Bg8 Kf8 7.Ba2 Kg7 8.Kb7
Kg6 9.Be6 Kg7 10.Kc7 Kg6 11.Kd6 Kg7
12.Ke5 Kh8 13.Ba2 Kg7 14.Kf5 Kh8 15.Bd5
Kg7 16.Kg5 Kh8 17.Ne5 Kg7 18.Ng6 Kh7
19.Kf6 Kh6 [<]
312809654 38: 139.9 +5.87 1.Ka7 Kd8 2.Bc6 Ke7 3.Ng4 Ke6 4.Be4 Kf7
5.Bd5+ Ke7 6.Bg8 Kf8 7.Ba2 Kg7 8.Kb7
Kg6 9.Be6 Kg7 10.Kc7 Kg6 11.Kd6 Kg7
12.Ke5 Kh8 13.Ba2 Kg7 14.Kf5 Kh8 15.Bd5
Kg7 16.Kg5 Kh8 17.Ne5 Kg7 18.Ng6 Kh7
19.Kf6 Kh6 [<]
314212353 39 140.7 :-) 1.Ka7
363450897 39 165.3 +Mat_60 1.Ka7 Kd8 2.Ba4 Ke7 3.Kb6 Kd6 4.Kb5 Kd5
5.Bb3+ Kd4 6.Nf3+ Ke4 7.Ng5+ Ke5 8.Nf7+
Kf5 9.Kc5 Kg6 10.Kc6 Kf6 11.Kd6 Kg7
12.Ke7 Kh7 13.Kf6 Kg8 14.Bc2 Kf8 [TT]
369154219 39 168.6 +Mat_59 1.Ng4 Kd8 2.Bg6 Kc7 3.Be4 [TT]
375501297 39 172.3 +Mat_57 1.Bb5 Kc7 2.Ka7 Kd6 3.Kb6 Kd5 4.Ng4 Kd4
5.Kc6 [TT]
379021642 39: 174.1 +Mat_57 1.Bb5 Kc7 2.Ka7 Kd6 3.Kb6 Kd5 4.Ng4 Kd4
5.Kc6 [TT]
382716394 40 176.6 +Mat_54 1.Bb5 Kc7 2.Ka7 Kd6 3.Kb6 Kd5 4.Bc6+
Kd4 5.Nf3+ Kd3 6.Kc5 Ke3 7.Kc4 Kf4
8.Kd4 Kf5 9.Bd5 Kf6 10.Ke4 Kg6 11.Ke5
Kg7 12.Kf5 Kh8 13.Ng5 Kg7 14.Nf7 Kh7
[TT]
388808840 40 180.2 +Mat_52 1.Ka7 Kc7 2.Ng4 Kd8 3.Bg6 Kc7 4.Be4 Kd7
5.Kb6 Kd6 6.Kb5 Ke6 7.Kc6 Ke7 8.Bd5 Kf8
9.Kd7 Kg7 10.Be4 Kg8 11.Ne5 Kf8 12.Bd5
Kg7 13.Ke7 Kh8 14.Kf6 Kh7 15.Nf7 Kg8
16.Kg6 Kf8 [TT]
390087515 40: 181.0 +Mat_52 1.Ka7 Kc7 2.Ng4 Kd8 3.Bg6 Kc7 4.Be4 Kd7
5.Kb6 Kd6 6.Kb5 Ke6 7.Kc6 Ke7 8.Bd5 Kf8
9.Kd7 Kg7 10.Be4 Kg8 11.Ne5 Kf8 12.Bd5
Kg7 13.Ke7 Kh8 14.Kf6 Kh7 15.Nf7 Kg8
16.Kg6 Kf8 [TT]
395941140 41 184.7 +Mat_43 1.Ka7 Kc7 2.Nf3 Kd8 3.Bg6 Kc7 4.Be4 Kd7
5.Kb6 Kd6 6.Kb5 [TT]
397805105 41: 185.7 +Mat_43 1.Ka7 Kc7 2.Nf3 Kd8 3.Bg6 Kc7 4.Be4 Kd7
5.Kb6 Kd6 6.Kb5 [TT]
418827078 42 195.7 +Mat_43 1.Ka7 Kd8 2.Bb5 Ke7 3.Kb6 Kd6 4.Nf3 Kd5
5.Bd3 Kd6 6.Be4 Ke6 [TT]
424883519 42: 199.1 +Mat_43 1.Ka7 Kd8 2.Bb5 Ke7 3.Kb6 Kd6 4.Nf3 Kd5
5.Bd3 Kd6 6.Be4 Ke6 [TT]
427047975 43 200.4 +Mat_43 1.Ka7 Kd8 2.Bb5 Ke7 3.Kb6 Kd6 4.Nf3 Kd5
5.Bd3 Kd6 6.Be4 Ke6 [TT]
432164525 43: 203.1 +Mat_43 1.Ka7 Kd8 2.Bb5 Ke7 3.Kb6 Kd6 4.Nf3 Kd5
5.Bd3 Kd6 6.Be4 Ke6 [TT]
434234911 44 204.4 +Mat_42 1.Ka7 Kd8 2.Bb5 Ke7 3.Kb6 Kd6 4.Nf3 Kd5
5.Bd3 Kd6 6.Nd4 Kd5 7.Nc6 Ke6 8.Kc5 Kf6
9.Kd6 Kf7 10.Ke5 Kg7 11.Ke6 Kh8 12.Kf6
Kg8 13.Ne5 Kh8 14.Bc2 Kg8 15.Nf7 Kf8
16.Bh7 Ke8 [TT]
437005254 44: 205.9 +Mat_42 1.Ka7 Kd8 2.Bb5 Ke7 3.Kb6 Kd6 4.Nf3 Kd5
5.Bd3 Kd6 6.Nd4 Kd5 7.Nc6 Ke6 8.Kc5 Kf6
9.Kd6 Kf7 10.Ke5 Kg7 11.Ke6 Kh8 12.Kf6
Kg8 13.Ne5 Kh8 14.Bc2 Kg8 15.Nf7 Kf8
16.Bh7 Ke8 [TT]
438966042 45 207.1 +Mat_42 1.Ka7 Kd8 2.Bb5 Ke7 3.Kb6 Kd6 4.Nf3 Kd5
5.Bd3 Kd6 6.Nd4 Kd5 7.Nc6 Ke6 8.Kc5 Kf6
9.Kd6 Kf7 10.Ke5 Kg7 11.Ke6 Kh8 12.Kf6
Kg8 13.Nd8 Kh8 14.Bc2 Kg8 15.Nf7 Kf8
[TT]
442367932 45: 209.1 +Mat_42 1.Ka7 Kd8 2.Bb5 Ke7 3.Kb6 Kd6 4.Nf3 Kd5
5.Bd3 Kd6 6.Nd4 Kd5 7.Nc6 Ke6 8.Kc5 Kf6
9.Kd6 Kf7 10.Ke5 Kg7 11.Ke6 Kh8 12.Kf6
Kg8 13.Nd8 Kh8 14.Bc2 Kg8 15.Nf7 Kf8
[TT]
459039438 46 217.9 +Mat_39 1.Ka7 Kd8 2.Bc6 Ke7 3.Ng4 Kf7 4.Be4 Ke6
5.Kb6 Kd6 6.Kb5 Ke7 7.Kc6 Ke6 8.Bg6 Ke7
9.Kd5 Kf8 10.Ke6 Kg7 11.Bb1 Kh8 12.Kf6
Kg8 13.Nh6+ Kf8 14.Nf7 Kg8 15.Be4 Kf8
[TT]
461793386 46: 219.5 +Mat_39 1.Ka7 Kd8 2.Bc6 Ke7 3.Ng4 Kf7 4.Be4 Ke6
5.Kb6 Kd6 6.Kb5 Ke7 7.Kc6 Ke6 8.Bg6 Ke7
9.Kd5 Kf8 10.Ke6 Kg7 11.Bb1 Kh8 12.Kf6
Kg8 13.Nh6+ Kf8 14.Nf7 Kg8 15.Be4 Kf8
[TT]
463579775 47 220.6 +Mat_39 1.Ka7 Kd8 2.Bc6 Ke7 3.Ng4 Kf7 4.Be4 Ke6
5.Kb6 Kd6 6.Kb5 Ke7 7.Kc6 Ke6 8.Bg6 Ke7
9.Kd5 Kf8 10.Ke6 Kg7 11.Bb1 Kh8 12.Kf6
Kg8 13.Nh6+ Kf8 14.Nf7 Kg8 15.Be4 Kf8
[TT]
466585707 47: 222.2 +Mat_39 1.Ka7 Kd8 2.Bc6 Ke7 3.Ng4 Kf7 4.Be4 Ke6
5.Kb6 Kd6 6.Kb5 Ke7 7.Kc6 Ke6 8.Bg6 Ke7
9.Kd5 Kf8 10.Ke6 Kg7 11.Bb1 Kh8 12.Kf6
Kg8 13.Nh6+ Kf8 14.Nf7 Kg8 15.Be4 Kf8
[TT]
468734770 48 223.5 +Mat_39 1.Ka7 Kd8 2.Bc6 Ke7 3.Ng4 Kf7 4.Be4 Ke6
5.Kb6 Kd6 6.Kb5 Ke7 7.Kc6 Ke6 8.Bg6 Ke7
9.Kd5 Kf8 10.Ke6 Kg7 11.Bc2 Kh8 12.Ne5
Kg7 13.Nf7 Kg8 14.Kf6 Kf8 [TT]
471910346 48: 225.3 +Mat_39 1.Ka7 Kd8 2.Bc6 Ke7 3.Ng4 Kf7 4.Be4 Ke6
5.Kb6 Kd6 6.Kb5 Ke7 7.Kc6 Ke6 8.Bg6 Ke7
9.Kd5 Kf8 10.Ke6 Kg7 11.Bc2 Kh8 12.Ne5
Kg7 13.Nf7 Kg8 14.Kf6 Kf8 [TT]
475290818 49 227.3 +Mat_39 1.Ka7 Kd8 2.Bc6 Ke7 3.Ng4 Kf7 4.Be4 Ke6
5.Kb6 Kd6 6.Kb5 Kd7 7.Kc5 Ke6 8.Kc6 Ke7
9.Kd5 Kf7 10.Kd6 Kg8 11.Ke6 Kg7 [TT]
483987333 49: 232.1 +Mat_39 1.Ka7 Kd8 2.Bc6 Ke7 3.Ng4 Kf7 4.Be4 Ke6
5.Kb6 Kd6 6.Kb5 Kd7 7.Kc5 Ke6 8.Kc6 Ke7
9.Kd5 Kf7 10.Kd6 Kg8 11.Ke6 Kg7 [TT]
572608341 50 273.6 +Mat_36 1.Ka7 Kd8 2.Bb5 Ke7 3.Kb6 Kd6 4.Nf3 Kd5
5.Bd3 Kd6 6.Bc4 Ke7 7.Kc5 Kf6 8.Bd3 Kg7
9.Ne5 Kg8 10.Kd5 Kh8 11.Kd6 Kg7 12.Ke6
Kh8 [TT]
579117277 50: 277.3 +Mat_36 1.Ka7 Kd8 2.Bb5 Ke7 3.Kb6 Kd6 4.Nf3 Kd5
5.Bd3 Kd6 6.Bc4 Ke7 7.Kc5 Kf6 8.Bd3 Kg7
9.Ne5 Kg8 10.Kd5 Kh8 11.Kd6 Kg7 12.Ke6
Kh8 [TT]
607593285 51 293.1 +Mat_34 1.Ka7 Kd8 2.Ba4 Ke7 3.Kb6 Kd6 4.Nf3 Kd5
5.Bc2 Kc4 6.Kc6 Kc3 7.Bb1 Kb2 8.Bh7 Kc3
9.Kc5 Kb3 10.Nd4+ Kc3 11.Bc2 Kb2 12.Kb4
Ka2 13.Kc3 Ka1 14.Bb3 Kb1 15.Nc2 Kc1
16.Ba2 Kd1 17.Nd4 Ke1 18.Kd3 Kf2 19.Ne2
Kg2 20.Be6 Kf3 21.Bh3 Kf2 22.Bg4 Ke1
23.Ke3 Kd1 24.Bf5 Ke1 25.Bc2 Kf1 [TT]
615000191 51: 297.2 +Mat_34 1.Ka7 Kd8 2.Ba4 Ke7 3.Kb6 Kd6 4.Nf3 Kd5
5.Bc2 Kc4 6.Kc6 Kc3 7.Bb1 Kb2 8.Bh7 Kc3
9.Kc5 Kb3 10.Nd4+ Kc3 11.Bc2 Kb2 12.Kb4
Ka2 13.Kc3 Ka1 14.Bb3 Kb1 15.Nc2 Kc1
16.Ba2 Kd1 17.Nd4 Ke1 18.Kd3 Kf2 19.Ne2
Kg2 20.Be6 Kf3 21.Bh3 Kf2 22.Bg4 Ke1
23.Ke3 Kd1 24.Bf5 Ke1 25.Bc2 Kf1 [TT]
617222187 52 298.7 +Mat_34 1.Ka7 Kd8 2.Ba4 Ke7 3.Kb6 Kd6 4.Nf3 Kd5
5.Bc2 Kd6 6.Be4 Ke7 7.Kc5 Ke6 8.Kc6 Kf7
[TT]
628746746 52: 304.8 +Mat_34 1.Ka7 Kd8 2.Ba4 Ke7 3.Kb6 Kd6 4.Nf3 Kd5
5.Bc2 Kd6 6.Be4 Ke7 7.Kc5 Ke6 8.Kc6 Kf7
[TT]
663898666 53 319.9 +Mat_33 1.Ka7 Kd8 2.Ba4 Ke7 3.Kb6 Kd6 4.Nf3 Kd5
5.Bc2 Kd6 6.Be4 Ke7 7.Kc5 Ke6 8.Kc6 Kf7
9.Kd7 Kf6 10.Kd6 Kg7 11.Ke7 Kh8 12.Ne5
Kg7 13.Nf7 Kg8 14.Kf6 Kf8 15.Bh7 Ke8
16.Ne5 Kd8 17.Ke6 Kc7 18.Nd7 Kb7 19.Bd3
Kc6 20.Bc4 Kc7 21.Bb5 Kd8 22.Kd6 Ke8
23.Bc4 Kd8 24.Bf7 Kc8 [TT]Code: Select all
849809506 64: 413.0 +Mat_33 1.Ka7 Kd8 2.Bg6 Ke7 3.Kb6 Kd6 4.Nf3 Kd5
5.Kc7 Kc5 6.Be4 Kb4 7.Kb6 Kc4 8.Kc6 Kb3
9.Kb5 Kc3 10.Kc5 Kb2 11.Kb4 Ka1 12.Nd4
Ka2 13.Kc3 Ka1 14.Bc2 Ka2 15.Nb3 Ka3
16.Bb1 Ka4 [TT]-
michiguel
- Posts: 6401
- Joined: Thu Mar 09, 2006 8:30 pm
- Location: Chicago, Illinois, USA
Re: Best was to Recognize Know Endgames?
That is very impressive.syzygy wrote:I have a bonus for getting the losing king closer to the right corner, and a penalty proportional to the distance between the two kings.hgm wrote:The "sepcial knowledge" here being that it would know which corner is the deadly one?syzygy wrote:Without special knowledge (and tablebases off) my engine takes 2 minutes to find a mate in 29 which then quickly goes down to mate in 21.
With special knowledge a mate in 25 is found in less than 8 seconds and mate in 21 in 14 seconds.
On this position my engine without tablebases and without special heuristics starts to report impossibly large mate values after 25 seconds (due to TT interactions). Mate in 45 after 42 seconds, mate in 33 after 94 seconds.Don wrote:Here is a KBNvsK position that requires 33 moves, which is the longest this should take with perfect play:
[d]K1k1B3/8/8/8/8/8/7N/8 w - - 0 1
I'm going to run this overnight - it should take a lot longer than the last one took.
Don
With special heuristics for KBNK it takes 31 seconds to get a mate in 45 and 61 seconds to get to mate in 33. No impossibly large mate values.
Miguel
Just for fun the two PVs I get:
1.Ka7 Kd8 2.Ba4 Kc7 3.Nf3 Kd6 4.Kb6 Kd5 5.Bc2 Kd6 6.Bb3 Kd7 7.Kc5 Ke7 8.Kd4 Kf6 9.Ke4 Kg6 10.Ke5 Kg7 11.Kf5 Kh6 12.Kf6 Kh7 13.Ne5 Kh8 14.Bc4 Kh7 15.Ng6 Kh6 16.Bg8 Kh5 17.Ne5 Kh4 18.Kf5 Kg3 19.Ng4 Kg2 20.Bc4 Kf3 21.Bb5 Kg3 22.Bc6 Kh3 23.Kf4 Kh4 24.Be8 Kh3 25.Ne3 Kh4 26.Ng2+ Kh3 27.Kf3 Kh2 28.Kf2 Kh3 29.Bd7+ Kh2 30.Ne3 Kh1 31.Bf5 Kh2 32.Nf1+ Kh1 33.Be4#
1.Ka7 Kd8 2.Bb5 Kc7 3.Nf3 Kd6 4.Kb6 Kd5 5.Bd3 Kd6 6.Bc4 Kd7 7.Kc5 Ke7 8.Kc6 Kf6 9.Kd6 Kf5 10.Bd3+ Kf6 11.Ne5 Kg7 12.Ke7 Kg8 13.Kf6 Kh8 14.Bc4 Kh7 15.Ng6 Kh6 16.Bg8 Kh5 17.Ne5 Kh4 18.Kf5 Kg3 19.Ng4 Kg2 20.Bc4 Kf3 21.Bd3 Kg3 22.Be2 Kh4 23.Nf2 Kg3 24.Nd3 Kh3 25.Kf4 Kh4 26.Ne1 Kh3 27.Bg4+ Kh2 28.Kf3 Kg1 29.Bh3 Kh2 30.Kg4 Kg1 31.Kg3 Kh1 32.Bg2+ Kg1 33.Nf3#