ProDeo (A Crazy Idea)

[mclane]

Here an interesting game that shows an interesting „problem“ where MMV made a mistake with a natural move

[pgn]
[Event "Match Nr. 57"]
[Site "Zürich"]
[Date "2020.??.??"]
[Round "17"]
[White "Scorpio 68000 sel. 4"]
[Black "Mephisto MM V HG 550"]
[Result "1-0"]
[ECO "C11"]
[WhiteElo "2080"]
[BlackElo "1995"]
[Annotator "KUT"]
[PlyCount "74"]
[EventDate "2020.??.??"]

1. e4 e6 {Match 120'/40 | Bediener: Rolf Bühler | 17. Partie: Weiss (Scorpio)
gewinnt | Zwischenstand: 9,5 : 7,5 oder 55% für Scorpio 68000 | Vergleich mit
SFC s4 = 12 : 5 oder 70% für SFC s4 | Vergleich mit SFC normal = 7,5 : 9,5
oder 44% für SFC normal | ERÖFFNUNG: In der Französischen Verteidigung
(Steinitz-Variante) wählte Schwarz (MMV) ein kritisches Abspiel mit dem
bekannten Springeropfer 10...Sxb4 gegen drei Bauern. | MITTELSPIEL: Mit 16...
c4 begeht der MM V jedoch bereits einen entscheidenden Fehler. Und im
Gegensatz zum SFC normal und SFC s4 findet der Novag Scorpio 68000 die
gewinnbringende Fortsetzung durch Rückgabe der Figur in der richtigen
Reihenfolge. Der dann folgende Königsangriff ist tödlich und führt zu einer
hübschen Kurzpartie für das Programm von David Kittinger.} 2. d4 d5 3. Nc3
Nf6 4. e5 Nfd7 5. f4 c5 6. Nf3 Nc6 7. Be3 Qb6 8. Na4 Qa5+ 9. c3 cxd4 10. b4
Nxb4 11. cxb4 Bxb4+ 12. Bd2 Bxd2+ 13. Nxd2 b6 {***ENDE BUCH***} 14. Bd3 Nc5 15.
Nxc5 bxc5 16. O-O {***ENDE BUCH***} c4 $2 {[#] Ein nur auf den ersten Blick
gut aussehender Vorstoss, der jedoch die Partie verliert. Der Mephisto MM V
bewertet diese Stellung mit 0.08 positiv, bei ST 8.01 habe ich nach 20 Min
abgebrochen.} (16... O-O 17. Qc2 Qc3 18. Bxh7+ Kh8 19. Rab1 Qxc2 20. Bxc2 c4
21. Ba4 Ba6 22. Nf3 d3 23. Kf2 c3 24. Bb5 Bxb5 25. Rxb5 Rac8 26. Ke3 d2 27. a3
a6 28. Rb6 Rc4 29. Rb4 Rxb4 30. axb4 Rc8 31. Kd3 Rc4 32. h3 {1/2-1/2 (32)
Keuter,K (2494)-Baranowski,T (2419) ICCF email 2014}) 17. Bxc4 $1 {Im
Gegensatz zu seinem Vorgänger Super Forte C 6 MHz normal und sel. 4 findet
der Scorpio den richtigen Weg, sprich Gewinnzug und gewinnt damit auch,
während die beiden anderen Partien mit Remis endeten. Der offensichtliche
Vorteil des Textzuges liegt darin, dass auf c4 schliesslich der Springer
auftaucht mit Angriff auf die Dame und der nachfolgenden Möglichkeit, dem
Gegner mit Sd6+ die Rochade zu verteufeln. Interessant indessen ist, dass der
Novag Scorpio 68000 den Gewinnzug mit einer negativen Bewertung von -00.54 bei
ST d08 ausspielt. Schockierend einmal mehr die Tatsache, wenn man (eigentlich
nicht zulässige) Vergleiche mit den modernen Engines anstellt, die allesamt
nach 16...dxc4 sofort bzw. im Sekundenbereich die weisse Stellung als gewonnen
betrachten.} (17. Qc1 $2 Ba6 18. Re1 O-O 19. Bf1 d3 20. Nb1 Rab8 21. Kh1 Rb7
22. Rd1 Rfb8 23. Qe3 Kh8 24. Qd4 Qb6 25. Qxb6 Rxb6 26. Nc3 $2 d4 27. Ne4 Rb2
28. Nc5 Bb5 29. Kg1 Rc8 30. a4 Bc6 31. Nxd3 cxd3 32. Rxd3 Rb4 33. Rc1 a6 34.
Rd2 Bb7 35. Rxc8+ Bxc8 36. Rc2 Bb7 37. Rc7 Kg8 38. a5 h5 39. Bd3 Rb2 40. g3 Rd2
$2 41. Rxb7 Rxd3 42. Ra7 Rd1+ 43. Kg2 Rd2+ 44. Kf3 Rxh2 45. Rxa6 d3 46. Ke3 Rg2
47. Ra8+ Kh7 48. a6 d2 49. Rd8 Rxg3+ 50. Kxd2 Ra3 51. Rd7 Rxa6 52. Rxf7 h4 53.
Ke3 Ra3+ 54. Kf2 Ra2+ 55. Kf3 h3 56. Rd7 h2 57. Rd1 Kg6 58. Re1 Kf5 59. Rf1 Ra4
60. Kg3 Rd4 61. Rh1 Rxf4 62. Rxh2 Kxe5 63. Rh8 Ke4 64. Rg8 Rf7 65. Re8 e5 66.
Kg4 Kd4 67. Rd8+ Ke3 68. Re8 e4 69. Ra8 Re7 70. Ra3+ Ke2 71. Ra2+ Kf1 72. Ra1+
Kf2 73. Ra2+ Ke1 74. Ra1+ Kd2 75. Ra2+ Kc3 76. Ra3+ Kd4 77. Ra4+ Kd5 78. Ra5+
Kc4 79. Ra4+ Kb5 80. Ra1 Kc5 81. Ra5+ Kd6 82. Ra6+ Ke5 83. Ra3 Rf7 84. Ra5+ Kd4
85. Ra4+ Kd3 86. Ra3+ Kd2 87. Ra2+ Kc3 88. Ra3+ Kc4 89. Ra4+ Kd5 90. Ra5+ Kc6
91. Re5 Rf1 92. Rxe4 Kd5 93. Re8 Kd6 94. Kg5 Rg1+ 95. Kf5 g6+ 96. Kf6 g5 97.
Kf5 g4 98. Kf4 Kd5 99. Rd8+ Kc4 100. Rd7 Kc3 101. Rd8 g3 102. Kg4 Kb4 103. Rc8
Ka5 104. Kf3 Kb5 105. Rc3 Kb4 106. Rc2 Rf1+ 107. Kxg3 Rf7 {½-½ (107) Super
Forte C 6 MHz sel. 4 (2022)-Mephisto MM V HG 550 (1982) Zurich 2019}) (17. Nxc4
$2 dxc4 18. Be4 Rb8 19. Qxd4 Qb6 20. Rfd1 Bd7 21. Rac1 Bb5 22. Qxb6 Rxb6 23.
Rd4 O-O 24. a4 Bxa4 25. Rcxc4 Bb3 26. Rc7 a5 27. Ra7 a4 28. Kf2 g6 29. Ke3 Kg7
30. g3 h6 31. h4 h5 32. Kf3 Kh6 33. Ke3 Kh7 34. Rdd7 Kg7 35. Rd4 Kh6 36. Kf2
Kh7 37. Ke3 Kg8 38. Rdd7 Kg7 39. Rd4 Rc8 40. Rdd7 Rf8 {½-½ (40) Super Forte
C 6 MHz (2022)-Mephisto MM V HG 550 (1982) Zurich 2019}) 17... dxc4 {Bis zu
den getesteten 23 Min bewertet der MM V bei ST 9 dieses Schlagen mit -0.30, d.
h. das Programm hat noch nicht den geringsten Durchblick.} 18. Nxc4 {Nicht
mehr bekannt ist, wie lange der Scorpio 68000 für diese Antwort brauchte. Ein
Test unter CB-Emu ergab ein -00.39 bis ST d07 und eine positive Bewertung von
00.37 bei ST d08 erst nach 7 Min 4 Sek.} Qb4 {Noch immer schläft der MM V,
wie folgenden Bewertungen zeigen: Nach 3 MIn nur -0.38 bei ST 7 / nach 4 Min
28 Sek nur -0.27 bei ST 8 / nach 35 Min nur -0.59 bei ST 9.05 mit erwarteter
Variante ...Dc7, Sd6 Kf8, Tac1.} 19. Nd6+ {Die Bewertung von Scorpio 68000 ist
auf 00.45 gestiegen.} Kf8 {Selbst hier ist sich der MM V der trostlosen Lage
nicht bewusst, denn nur -0.91 bei ST 8.00 nach 7 Min 55 Sek in Erwartung von
Df3.} 20. Qh5 {Das ist schon absolut tödlich, wobei die positive Bewertung
des Scorpio 68000 mit bescheidenen 1.28 bescheinigt, dass der vollständige
Durchblick noch nicht vorhanden ist. Erst nach 8 Min 32 Sek springt die
Bewertung bei ST d08 auf +2.03.} Qb7 {Endlich sieht der MM V ein, dass es
keine Verteidigung mehr gibt und opfert die Dame. Nach steigenden
Zwischenbewertungen von -0.03, -0.38, -0.69 bis ST 7.00 erfolgt bei ST 7.01
nach 3 MIn 20 Sek mit einer Bewertung von -5.27 das böse Erwachen.} (20... g6
{wird beantwortet mit} 21. Qh4 Qb6 22. Qf6 d3+ 23. Kh1 $18 {mit der
erzwungenen Folge des Damenopfers} Qxd6 24. Qxh8+ Ke7 25. exd6+ Kxd6 26. Qf8+
Kd7 27. Rac1 Bb7 28. Qxf7+ Kd8 29. Rc7 Bxg2+ 30. Kxg2 d2 31. Qd7#) 21. Nxb7 $18
Bxb7 {Man könnte abbrechen, denn mit Dame gegen L+B hat Weiss leichtes Spiel.}
22. Rfc1 (22. f5 {führt noch rascher zum Matt}) 22... Be4 23. Rc7 {droht Matt
auf f7} Bg6 24. Qd1 (24. Qf3 {bringt ein schnelleres Matt}) 24... Kg8 25. Qxd4
{Der Bediener lässt nur noch deshalb weiterlaufen, weil er sehen will, wie
lange der MM V noch braucht, um die mehr als hoffnungslose Partie aufzugeben.}
h6 26. Rxa7 Rxa7 27. Qxa7 Kh7 28. Qc7 Ra8 29. a4 Kg8 30. a5 Be4 31. a6 Rf8 32.
a7 g6 33. Rd1 h5 34. Rd8 Rxd8 35. Qxd8+ Kg7 36. Kf2 Bb7 37. a8=Q Bxa8 {Schwarz
gibt (endlich) auf} 1-0[/pgn]


The critical mistake is 16...c4 ?
that gives away the game.

C4 looks so natural.
But the black position has a problem.
Black did not have castled.
And the square d6 could be a good position for a white knight.

You have to see that these dedicated chess computers did only 4-7 searches.
And often this depth was splitted between , let’s say 3 brute force with a selective peak of 4 etc.

Todays engines come 30 searches deep in the first second.
In those early days where the cpu was only 5 MHz 6502 the computer needed minutes to reach into depth 7.

[mclane]

Have the position now on my Monte Carlo IV.

In ply 8 it sees c4 bxc4 Dxc Nxc4 qd5 Nd6+ But Evaluates this with 0.12 plus.
It’s now computing arround 2h in ply 8.
And not evaluating the end position heavily enough,

[maksimKorzh]

Seeing that ProDeo is rated "only" 2722, have you never thought about just starting a new engine from scratch?

umm, "Rustic" hasn't been properly rated yet, but you're already lecturing a former world champion, that's hilarious
2700 is a low rating for you perhaps? you barely wrote 1 engine yet you already tell other people to do a rewrite?
maybe you should make a youtube video about that? oh wait, that's the other guy.

where does this BS with rewriting come from anyway? if you rewrite, you need to catch up with your last version first, which will take some time (been there, done that).
you don't magically gain 500 elo out of thin air along the way like Sceviour of course (assuming native=>native).

Any problems with youtube videos?
I don't see any 2500+ chess engine authors making video tutorials. Probably puny mortals aren't worthy to touch strong chess engines. It feels like discrimination - like if you're too dumb to understand our 30+ years development source code in super complicated C++ then you just stay there at 1600 elo and don't talk much.

I don't want to offend anybody and I respect top engines and especially old engines like crafty (probably the only engine that tries to help others with it's source code) but the problem is that there's a great gap between 1600-2100 and 2700+ engines. Not in terms that there are no engines between, they are according to CCRL, but in terms of understanding of all of those tiny little details.

Just to give an example: I swear I was trying to read Andrew Grant's paper on tuning eval but after reading first page I've realized that I need at least 5 more years to get at least the concepts and another 5 years to implement that when in 10 years there won't be handcrafted eval at all.

There are lots of cool high level articles for PROs like deepmind's papers on alpha go but what those documents say to less skilled people? Literally the say like: 'you, idiots, are too stupid to understand it'

Hence my youtube work and it already starts giving results - newcomers starting their own engines, they try to follow my tutorials in other languages (kotlin/python) by rewriting my C code in their languages. Others learn to code via following chess programming tutorials.

Do you think this is bad?
Or not worth doing it?

I wish I could to learn from someone more skilled than I but noone makes easy to follow video tutorials. Bluefever was the only guy and thanks to him so many people got involved into chess programming.

I'm just extending his work and serve the same purpose - remove descrimination by brain skills and bring chess programming to everyone.

[maksimKorzh]

umm, "Rustic" hasn't been properly rated yet, but you're already lecturing a former world champion, that's hilarious
2700 is a low rating for you perhaps? you barely wrote 1 engine yet you already tell other people to do a rewrite?

OMG. Why are you always screaming "Where's you're engine?!" and "Where's you're rating?" if someone asks a genuine question about another engine? (My engine is right there on Github, and currently, its rating is somewhere around 1600... for now.)

I have nothing but respect for what Ed (and others that started at around that time) did in chess computer programming, but you can't have missed that there are NEW engines that start at 1400 Elo, and after less than two years of development, exceed 2700 or even 2800 Elo on the CCRL list, beyond the engines that have been around quite some time. This suggests that it is very hard to still improve engines that are based on an older code base, and easier to improve newer engines.

Therefore I asked if some of the authors of engines that were once top of the line, aren't considering in writing something new from scratch. Yes, it may take a year, but then you have a new code base that can last another 30 years.

maybe you should make a youtube video about that? oh wait, that's the other guy.

That's just unnecessary. Maksim may not be the god of chess programming (and I'm not either), but he's doing some good work in getting the basics explained for people who can program and want to try writing a chess engine, but don't know where to start.

where does this BS with rewriting come from anyway? if you rewrite, you need to catch up with your last version first, which will take some time (been there, done that).
you don't magically gain 500 elo out of thin air along the way like Sceviour of course (assuming native=>native).

Where the rewrite BS is coming from? Some engines have a code-base that goes back 30 years. Some are even older. They may contain code that was optimized perfectly for the CPU's of the day, but which doesn't run that great nowadays. Older code may prevent the author from using newer compilers, and/or prevent them from implementing newer techniques in the engine. It can be worthwhile to write a new engine, or even to port the engine to a newer version of the language it is written in.

Fabien Letouzy wrote Fruit in the beginning to mid-2000's. That program reached an Elo-rating of 2695. 10 years later he released Senpai, and that program reached a rating of 3100. That's a 400 point advantage for a newer engine.

Totally agree and support you position, Marcel.
And thanks for explaining what am I doing on youtube.