Zugzwang Test Suite ZPTS by Robert Allgeuer

[lech]

Sting Sf 4 can solve no. 14 in 1 minutes and no. 16. in 2 minutes (dual core).
Regarding problems with Stockfish:

"To have or to be?"

[velmarin]

Quite an achievement Sting and Marek,

It would be interesting to try Ivanhoe with Complete Zugzwang function.

Paul henri, if you are interested I can compile.

Jose.

[bob]

I just compile the 120214 version and I tested it twice, as well as with the syzygy. I also do a second test for the version 101113

What is strange is that I do not always get the same result, it seems random for some positions.

Code: Select all

ZPTS,xx	101113	2nd try	120214	2nd try	+syzygy	2nd try
ZPTS.01	Solved	Solved	Solved	Solved	Solved	Solved
ZPTS.02	Solved	Solved	Solved	Solved	Solved	Solved
ZPTS.03	Solved	Solved	Solved	Solved	Solved	Solved
ZPTS.04	Solved	Solved	Solved	Solved	Solved	Solved
ZPTS.05	Solved	Solved	Solved	Solved	Solved	Solved
ZPTS.06	Solved	Solved	Solved	Solved	Solved	Solved
ZPTS.07	Solved	Solved	Solved	Solved	Solved	Solved
ZPTS.08	Solved	Solved	Solved	Solved	Solved	Solved
ZPTS.09	Solved	Solved	Solved	Solved	Solved	Solved
ZPTS.10	Solved	Solved	Solved	Solved	Solved	Solved
ZPTS.11	Solved	> 10s.	> 10s.	> 10s.	Solved	> 10s.
ZPTS.12	Solved	Solved	Solved	Solved	Solved	Solved
ZPTS.13	Solved	Solved	Solved	Solved	Solved	Solved
ZPTS.14	> 10s.	> 10s.	> 10s.	> 10s.	> 10s.	> 10s.
ZPTS.15	Solved	Solved	Solved	Solved	Solved	Solved
ZPTS.16	> 10s.	> 10s.	> 10s.	Solved	Solved	Solved
ZPTS.17	Solved	Solved	Solved	Solved	Solved	Solved
ZPTS.18	Solved	> 10s.	> 10s.	> 10s.	> 10s.	> 10s.
ZPTS.19	> 10s.	> 10s.	> 10s.	> 10s.	> 10s.	> 10s.
ZPTS.20	Solved	Solved	Solved	Solved	Solved	Solved
ZPST.21	> 10s.	> 10s.	> 10s.	> 10s.	> 10s.	> 10s.
ZPST.22	Solved	Solved	Solved	Solved	Solved	Solved
ZPTS.23	Solved	Solved	Solved	Solved	Solved	Solved
ZPTS.24	> 10s.	> 10s.	> 10s.	> 10s.	> 10s.	> 10s.
ZPST.25	Solved	> 10s.	> 10s.	> 10s.	Solved	Solved
ZPST.26	Solved	Solved	> 10s.	> 10s.	> 10s.	> 10s.
ZPTS.27	Solved	Solved	Solved	Solved	Solved	Solved
ZPTS.28	> 10s.	> 10s.	> 10s.	> 10s.	> 10s.	> 10s.
ZPTS.29	Solved	Solved	Solved	Solved	Solved	Solved
ZPST.30	Solved	Solved	Solved	Solved	Solved	Solved
Result:	24/30	 21/30	 20/30	 21/30	 23/30	 22/30
Percent:  80.0%.	70.0%.	66.6%.	70.0%.	76.6%.	73.3%

For the record, there are a couple I don't like. The last one, for example.

Nxg5 is a mate in 5. As is Rxc3. Why is Rxc3 correct and Nxg5 is not? Nxh1 seems to be a mate in 6 as well. I'd think any would be acceptable although the first two seem to be equal.

Others include #26 where Ne3 is a mate in 10, #25 where Qf1 is a mate in 12.

[lech]

Quite an achievement Sting and Marek,

It would be interesting to try Ivanhoe with Complete Zugzwang function.

Paul henri, if you are interested I can compile.

Jose.

If I remember well, all the Ippolit's versions have some problem with this easy zugzwang position.
[d]8/8/8/8/1kB1q3/8/1PRb4/1K6 b - -
To win it needs to lose the tempo move.
An easy continuation (one of more) is: 1...Qh1+ 2.Ka2 Qa8+ 3.Kb1 Qe4!!.
Houdini 1.5a has some problems to solve the above simple and easy zugzwang position (???)

[velmarin]

Sting very good ¡¡¡

Sting SF 4 x64-509MB/ 28 positions Total time: 0:01:41

IvanHoe Zuzgwang x64-509MB/ 26 positions Total time: 0:03:10


IvanHoe Zuzgwand + source.
https://www.dropbox.com/s/f361aayti20mw ... uzgwand.7z

[phenri]

For the record, there are a couple I don't like. The last one, for example.

Nxg5 is a mate in 5. As is Rxc3. Why is Rxc3 correct and Nxg5 is not? Nxh1 seems to be a mate in 6 as well. I'd think any would be acceptable although the first two seem to be equal.

Others include #26 where Ne3 is a mate in 10, #25 where Qf1 is a mate in 12.

#30 you meant Ng5

position26: there are many wining move in mate in 5 but only one in mate in 4. But I agree with you, from a human point of view, this little matter. But anyway the question is not here because Stockfish not find any of these moves! because it prefers to take such a glutton g5

8/3p1p2/5Ppp/K2R2bk/4pPrr/6Pp/4B2P/3N4 w - - 0 1
1. +- (#4): 1.Nc3 d6 2.Bd1 e3 3.Ne2 Rxg3 4.Nxg3#
2. +- (#5): 1.Ne3 d6 2.Rb5 d5 3.Nxd5 e3 4.Ka4 Bxf4 5.Nxf4#
3. +- (#5): 1.Rb5 d6 2.Ne3 d5 3.Nxd5 e3 4.Ka4 Bxf4 5.Nxf4#
4. +- (#5): 1.Ka4 d6 2.Nc3 e3 3.Rb5 d5 4.Nxd5 Bxf4 5.Nxf4#
5. +- (#5): 1.Kb4 d6 2.Nc3 e3 3.Rb5 d5 4.Nxd5 Bxf4 5.Nxf4#
6. +- (#5): 1.Kb5 e3 2.Nc3 d6 3.Bd1 e2 4.Nxe2 Rxg3 5.Nxg3#
7. +- (#5): 1.Kb6 d6 2.Nc3 e3 3.Rb5 d5 4.Nxd5 Bxf4 5.Nxf4#
8. +- (#5): 1.Ka6 d6 2.Nc3 e3 3.Rb5 d5 4.Nxd5 Bxf4 5.Nxf4#

position25: with Qf1 is a mate in 14
I return you the question, why do you prefer mate in 14 or even 12 moves when it is in 3 moves?
Or why do you want a flying Paris-New York with a stop at the North Pole when you can not stop?

[phenri]

The reason is that you use more than 1 core.
The only good test for comparison is with one core unless you want to repeat the same position 10 times and find the number of times that it get it correctly.

I do not find normal to have random results in smp. And return the fault to the CPU is a false excuse.

[bob]

For the record, there are a couple I don't like. The last one, for example.

Nxg5 is a mate in 5. As is Rxc3. Why is Rxc3 correct and Nxg5 is not? Nxh1 seems to be a mate in 6 as well. I'd think any would be acceptable although the first two seem to be equal.

Others include #26 where Ne3 is a mate in 10, #25 where Qf1 is a mate in 12.

#30 you meant Ng5

position26: there are many wining move in mate in 5 but only one in mate in 4. But I agree with you, from a human point of view, this little matter. But anyway the question is not here because Stockfish not find any of these moves! because it prefers to take such a glutton g5

8/3p1p2/5Ppp/K2R2bk/4pPrr/6Pp/4B2P/3N4 w - - 0 1
1. +- (#4): 1.Nc3 d6 2.Bd1 e3 3.Ne2 Rxg3 4.Nxg3#
2. +- (#5): 1.Ne3 d6 2.Rb5 d5 3.Nxd5 e3 4.Ka4 Bxf4 5.Nxf4#
3. +- (#5): 1.Rb5 d6 2.Ne3 d5 3.Nxd5 e3 4.Ka4 Bxf4 5.Nxf4#
4. +- (#5): 1.Ka4 d6 2.Nc3 e3 3.Rb5 d5 4.Nxd5 Bxf4 5.Nxf4#
5. +- (#5): 1.Kb4 d6 2.Nc3 e3 3.Rb5 d5 4.Nxd5 Bxf4 5.Nxf4#
6. +- (#5): 1.Kb5 e3 2.Nc3 d6 3.Bd1 e2 4.Nxe2 Rxg3 5.Nxg3#
7. +- (#5): 1.Kb6 d6 2.Nc3 e3 3.Rb5 d5 4.Nxd5 Bxf4 5.Nxf4#
8. +- (#5): 1.Ka6 d6 2.Nc3 e3 3.Rb5 d5 4.Nxd5 Bxf4 5.Nxf4#

position25: with Qf1 is a mate in 14
I return you the question, why do you prefer mate in 14 or even 12 moves when it is in 3 moves?
Or why do you want a flying Paris-New York with a stop at the North Pole when you can not stop?

Simple. I prefer a human-like approach here. It is quite common for a program to find a longer mate first, particularly when it is produced by a long sequence of checks. A shorter mate with a couple of non-checking moves is harder to find and takes more iterations. After a couple of iterations with a mate score, I don't see any reason to keep searching to see if there is a shorter mate. I have code in Crafty that says "if you find a mate in N this move, the next search has to continue until a mate in N-1 is found to make sure there is progress.

So I really don't like a longer mate over a shorter one, but at times the longer mate is found first, not because of zugzwang, or anything else, other than the longer mate has more forcing moves (checks) which extend the search significantly deeper than a sequence of quiet moves...

As far as Qf1 goes, Crafty says mate in 10 after a few seconds. And then finds Qf3 as a mate in 8. Good zugzwang test positions should have one move that wins, if one can deal with the zugzwang, where all others lose, or else one move that draws where all others lose... Then the zugzwang detection becomes critical since it changes the game outcome if it is correct...

[bob]

The reason is that you use more than 1 core.
The only good test for comparison is with one core unless you want to repeat the same position 10 times and find the number of times that it get it correctly.

I do not find normal to have random results in smp. And return the fault to the CPU is a false excuse.

Unfortunately, he is correct. Parallel search is most definitely a non-deterministic algorithm, and does regularly produce different results. The time to reach a specific ply is almost always wildly variable, and sometimes event the score or best move changes.

[phenri]

I took my GPS in car so he guides me from point A to point B. If it makes me go through xyz points, I throw it out the window.