New MGV-77 testset from Michael Gurevich

[Eelco de Groot]

Dear forum members,

Michael Gurevich completed the second part of his tough positions testset taken from from human GM practice, as testpositions for the computer.

Regarding the first part of the testset I wrote the following:

All the positions are carefully tested. I think they were all published separately in the CSS-forum and only the ones best suited for a test were chosen. In the CSS-forum the positions were tested for at least twenty minutes CPU time and often for several hours or sometimes even days, by a lot of different users so there is ample comparison material at long timecontrols to be found doing a search in the CSS-forum, this makes it possible to check all the PV-output of your engine, not just the solutions themselves. I think that makes these tests more valuable because it can give a balanced view and comparison at all timecontrols and of the analysis behaviour during the whole search.

Download of the second part of the MGV test is possible here: http://www.computerschach.de/index.php? ... info&id=82

First part of the MGV-testsuite:
http://www.computerschach.de/index.php? ... info&id=79

Michael writes in German:

Die neue, erweiterte Testsuite von Dr. Michael Gurevich (Deutschland) "Meister der Gegenwart und Vergangenheit", (MGV77-Test) mit 77 Positionen steht nun auf der CSS-Online-Download-Seite zum freien Runterladen: http://www.computerschach.de/index.php? ... info&id=82

Der Download enthält folgende Dateien:
1. Eine Datei mit allen 77 Teststellungen in epd-, pgn- und cbv-Formaten.
2. Eine Textdatei mit den Testbedingungen.
3. Eine Aufgabenauswahl in den PGN- und CBV-Formaten nach jeweils K-, P- und E-Spiel.
4. Eine PGN-Datei mit der ausführlichen Variantenbeschreibung der neuen Aufgaben 41-77
Der gesamte Test enthält:
- 29 Königsangriff-Aufgaben (K);
- 29 Aufgaben aus dem Bereich Positionsspiel (P);
- 19 Aufgaben aus der Endspielphase der Partie (E).

Die meisten Positionen haben einen erhöhten Schwierigkeitsgrad uns sind zum ersten Mal in einem Stellungstest veröffentlicht.
Der erste Teil war bekanntlich dem modernen menschlichen Schach gewidmet:
http://www.computerschach.de/index.php? ... c=fileinfo
Der zweite Teil des MGV-Tests stellt eine Sammlung der Kunstwerke alter Meister des 20.Jahrhunderts dar, darin gibt es unter anderem solche großen Namen wie Rubinstein, Keres, Geller, Reshevsky, Bronstein, Portisch, Polugaevsky, Timman, Kortschnoj.
Neben der breiten öffentlichen Besprechung im CSS-Forum und der speziell geführten Langzeitanalysen jeder Position 15-20 Stunden lang, wurde jede Aufgabe von unserem Team sorgfältig mit den Varianten geprüft, um die Eindeutigkeit und Korrektheit der menschlichen Kombinationen zu beweisen.
Findet eine Engine den einzigen besten Zug in einer in der Regel komplizierten, anspruchsvollen Position, so demonstriert sie ihr hohes Schachverständnis. Eine Lösung aus einem falschen Grund ist bei der Engine-Analyse im MGV-Test unwahrscheinlich:
a) Die instabilen Aufgaben wurden in unserer Auswahl anhand der monatelangen Untersuchungen aussortiert;
b) Die empfohlene Testdauer 5-10 min reicht meistens aus, um zufällige Lösungen zu vermeiden.

Viel Spaß wünscht

Michael Gurevich


Freely translated:

The new enlarged testset from Dr. Michael Gurevich, "Masters of the present and the past", MGV-77, with 37 additional positions is now available in the CSS-Online-Download-repository.

The download consists of the following files:

• A file with all the 77 testpositions in epd-, pgn- und cbv-format.
• A textfile with the testconditions.
• Sorted files with testpositions in the categories (K)King-attack, (P) positional play and (E) endgames, in pgn- and cbv-format.
• A pgn-file with detailed description of analysed lines for the new testpositions 41-77.

In all, this testsuite has:

• 29 King attack positions;
• 29 Positions characterized by position play;
• 19 Endgame psitions.

Most positions are of an increased level of difficulty and are published for the first time as part of a test-suite.

Where the first part of the MGV-test was, as you may recall, dedicated to modern chess endeavours, the second part of the MGV-test represents a collection of masterpieces by the old masters of the twentieth century, with renowned names like Rubinstein, Keres, Geller, Reshevsky, Bronstein, Portisch, Polugaevsky, Timman and Korchnoi.

In addition to the public analysis in the CSS- forum and additional overnight analysis of 10 to 20 hours, the side-variations of every position were tested by the CSS team to guarantee the correctness of the human combinations.

Does an engine find the single best move in an, as a rule complex and taxing, testposition, then it shows that it possesses effective chess knowledge. Because of the careful selection, finding the right move for the wrong reasons is unlikely in this MGV-testsuite:

1. Any unstable testpositions were sorted out by timeconsuming tests;
2. The recommended testing time of 5-10 minutes is usually sufficient to rule out accidental solutions.

Hoping that you will enjoy the test,

Michael

[Jouni]

I already tested 4 engines in P2400 with 5 min limit. As always Fritz is the best:

Fritz 10-128MB Rybka 2.3.2a 32-bit -128MB bright-0.2c-128MB Toga II 1.4 -128MB
RUBINSTEIN - Johner 0:02:30 0:00:54 0:05:00 0:05:00
RICHTER - Abramavicius 0:00:38 0:05:00 0:00:12 0:02:25
KERES - Euwe 0:05:00 0:05:00 0:05:00 0:05:00
KERES - Kotov 0:00:01 0:05:00 0:05:00 0:05:00
Tal - KERES 0:01:28 0:05:00 0:02:21 0:02:00
GELLER - Lombardy 0:00:17 0:05:00 0:05:00 0:00:10
GELLER - Vasiukov 0:04:03 0:00:21 0:05:00 0:03:25
Geller - UHLMANN 0:05:00 0:05:00 0:05:00 0:05:00
BRONSTEIN - Blau 0:05:00 0:04:36 0:01:32 0:00:20
BRONSTEIN - Ljubojevic 0:00:12 0:05:00 0:02:15 0:05:00
Shamkovich - NEZHMETDINOV 0:00:57 0:00:40 0:02:35 0:00:46
Evans - RESHEVSKY 0:05:00 0:05:00 0:05:00 0:05:00
Petrosian,T - PORTISCH 0:00:56 0:01:32 0:05:00 0:05:00
PORTISCH - Radulov 0:02:39 0:05:00 0:05:00 0:05:00
Gheorghiu - PORTISCH 0:00:17 0:05:00 0:05:00 0:05:00
Mecking - PORTISCH 0:00:01 0:05:00 0:04:46 0:00:13
DIEZ del CORRAL - Portisch 0:05:00 0:05:00 0:05:00 0:05:00
POLUGAEVSKY - Tal 0:05:00 0:01:43 0:05:00 0:04:12
POLUGAEVSKY - Torre 0:00:13 0:00:12 0:05:00 0:00:18
KUZMIN - Sveshnikov 0:05:00 0:00:01 0:01:24 0:05:00
Westerinen - KORTSCHNOJ 0:05:00 0:00:03 0:00:13 0:01:53
KORTSCHNOJ - Huebner 0:04:05 0:05:00 0:00:20 0:05:00
KORTSCHNOJ - Hodgson 0:05:00 0:03:12 0:05:00 0:05:00
KORTSCHNOJ - Svidler 0:04:31 0:03:16 0:05:00 0:01:05
Van der WIEL - Ribli 0:00:17 0:05:00 0:00:06 0:00:35
TIMMAN - Huebner 0:00:12 0:00:14 0:00:27 0:05:00
TIMMAN - Ljubojevic 0:05:00 0:00:01 0:05:00 0:05:00
TIMMAN - Jussupow 0:05:00 0:05:00 0:01:48 0:05:00
NUNN - Lalic 0:05:00 0:05:00 0:05:00 0:05:00
NUNN - Smeets 0:05:00 0:05:00 0:05:00 0:05:00
GUREVICH,M - Nikolic 0:05:00 0:00:06 0:05:00 0:05:00
GUREVICH,M - Anand 0:05:00 0:00:01 0:00:08 0:05:00
BLATNY - Eisterer 0:00:37 0:05:00 0:03:54 0:05:00
ONISCHUK - Vescovi 0:00:19 0:02:05 0:05:00 0:05:00
NAVARA - Uhlmann 0:05:00 0:05:00 0:05:00 0:05:00
Psakhis - CHEKHOV 0:01:57 0:00:23 0:05:00 0:02:58
KAMSKY - Short 0:05:00 0:05:00 0:05:00 0:05:00
Total time: 1:11:10 1:19:20 1:37:01 1:40:19
Total solved: 20 17 14 13

Sorry for formatting - don't know how to fix...
Jouni

[Mike S.]

I already tested 4 engines in P2400 with 5 min limit.

Did you adjust "Extra ply" to maximum (5)?

The default 1 is not safe. Theoretically, an engine could decide for the correct move at a very small depth for whatever (wrong, or incomplete) reason, but deviate from it 2 plies later. Extra ply 5 finishes at least the current and the next 4 plies, as long as it is within the maximum time frame.

I know that the positions were selected carefully in a way that such "unstable" solutions seem unlikely, but IMO that is not always possible or 100% reliable. Optimal would be to go to the maximum time always and find the latest stable decision time (if kept up to max. time), but Fritz' automatic test suite feature doesn't offer that.

[Jouni]

Of course, even if sometimes Fritz GUI still seems to accepts wrong solutions. Ep=99 is safer.

I forgot to add, that I tested new 37 positions only.

Jouni