SF channel on Discord : weird people...

[unofficial]

Yesterday i post few ideas on discord in the "ideas-only" of the SF channel. Then i tried to explain them in the "ideas-discuss". And this morning i saw my idea about "add a learning feature to SF" was censored !? Yet SF without this feature has no chance !

For example :

On the same hardware, same opening, same settings, SF is clearly weaker than all other engines.


Here we see SF lost tons of games as black after d7c5.


Engines with learning feature need only few tens of games.

[unofficial]

SF kept playing d7c5 :


So he almost lost all his games as black :


Pathetic :

[Peter Grayson]

SF kept playing d7c5 :


So he almost lost all his games as black :


Pathetic :

There are a number of considerations that should be taken into account.

Firstly, that an opening book was being used that took the game to a cut off point that was a disadvantageous position for black. In that case why is the cut off point where the bad move choice is made at very fast time controls and shallow depths?

On my old Q9550 quad + 1Gb hash, in analysis mode Stockfish 130722 finds 13..Nb6 at 26 ply depth and sticks with it. In game mode it takes 4s longer but at the same ply depth of 26 changes to 13..Nb6 and sticks with it. Therefore the question is that after one loss with 13..Nc5, why wasn't the book line changed to 13..Nb6 that would perhaps have avoided all of those losses? The learning demonstration ignores the quickest route to correcting the 13th move that is by changing the book move selection. On my bigger hardware the Stockfish engine finds 13..Nb6 proportionately quicker.

Therefore contrary to the statement Stockfish has no chance without learning feature it is all relative to the time controls and hardware being used. The use of shallow ply depth to demonstrate a characteristic is never a good or convincing starting point!

Secondly, the opening line for Black is a bad choice. In the Sämisch variation of the King's Indian Black does well to completely avoid the demonstration line by playing 8..exd4 when Black has at least some decent fighting chances ...

[Event "Clichy"] [Site "Clichy"] [Date "1998.12.14"] [Round "3"] [White "Murey, Jacob"] [Black "Relange, Eloi"] [Result "1-0"] [ECO "E89"] [WhiteElo "2460"] [BlackElo "2490"] [Annotator "Grayson,Peter"] [PlyCount "87"] [EventDate "1998.05.16"] [EventType "tourn"] [EventRounds "9"] [EventCountry "FRA"] [EventCategory "7"] [SourceTitle "EXT 2000"] [Source "ChessBase"] [SourceDate "1999.11.16"] [SourceVersion "1"] [SourceVersionDate "1999.11.16"] [SourceQuality "1"] 1. d4 Nf6 2. c4 g6 3. Nc3 Bg7 4. e4 d6 5. f3 O-O 6. Be3 e5 7. Nge2 c6 8. Qd2 Nbd7 ({Stockfish 130722:} 8... exd4 9. Bxd4 (9. Nxd4 {Stockfish 130722:} d5 10. cxd5 (10. exd5 {Stockfish 130722:} cxd5 11. O-O-O Nc6 12. Bh6 Bxh6 13. Qxh6 Re8 14. Nc2 d4 15. g4 Bd7 16. Nb5 a6 17. Nbxd4 Qc7 18. Rg1 Nxd4 19. Rxd4 Bc6 20. Qf4 Qxf4+ 21. Rxf4 Nd7 22. Rg3 Kg7 23. Rd4 Ne5 24. f4 Nf3 25. Rd6 Nxh2 26. Bd3 Be4 27. g5 Bxd3 28. Rdxd3 Re2 29. Rge3 Rf2 30. Re4 Rc8 31. b4 Ng4 32. Kb2 h5 33. gxh6+ Kxh6 34. Kb3 Kg7 35. Ne3 Nxe3 36. Rdxe3 Rc6 37. c5 b6 38. Rc3 Rd2 39. Rec4 Re2 {[%eval 26,50]}) 10... cxd5 11. e5 Ne8 12. Bg5 Qd7 13. Nb3 Bxe5 14. Nc5 Bxc3 15. Qxc3 Qd6 16. Rd1 Nc6 17. Kf2 d4 18. Qc1 Bf5 19. Nxb7 Qc7 20. Nc5 Ng7 21. b3 h5 22. Bc4 Rfe8 23. Rhe1 Qxh2 24. Bf4 Qh4+ 25. Kg1 Qf6 26. Bb5 Rxe1+ 27. Rxe1 Rc8 28. Ba6 Rf8 29. Bc4 h4 30. Bg5 Qd6 31. Bxh4 Ne5 32. Bf1 d3 33. Qc3 f6 34. b4 Nh5 35. Bf2 Rc8 36. Qd4 Qxd4 {[%eval 24,50]}) 9... Nbd7 10. Rd1 a6 11. Nf4 b5 12. Be2 Qa5 13. b3 Re8 14. O-O Ne5 15. Ncd5 Qxd2 16. Nxf6+ Bxf6 17. Rxd2 Bg5 18. g3 f5 19. h4 (19. exf5 Bxf5 20. h4 Bh6 {transposes back into line} ) 19... Bh6 20. exf5 Bxf5 21. Kf2 bxc4 22. bxc4 a5 23. Bc3 ({Stockfish 130722: } 23. Be3 Rab8 ({Stockfish 130722:} 23... Nf7 24. Ng2 Bg7 25. g4 Bd7 26. Bd4 Bxd4+ 27. Rxd4 Reb8 28. Rd2 a4 29. Rc1 a3 30. Ne3 Rb2 31. Rcc2 Rab8 32. Nd1 R2b4 33. Rc3 Be6 34. c5 Rb2 35. Nxb2 axb2 36. Bd3 b1=Q 37. Bxb1 Rxb1 38. cxd6 Bd5 39. d7 Rb7 40. Rxd5 cxd5 41. Rc8+ Kg7 42. d8=Q Nxd8 43. Rxd8 Ra7 44. Rxd5 Rxa2+ 45. Kg3 h6 46. Rd7+ Kg8 47. Kf4 Re2 48. Rc7 Re8 49. Rc5 Re7 50. h5 Rf7+ 51. Kg3 g5 52. Re5 Kg7 53. Kf2 Ra7 54. Rb5 Rc7 55. Ke3 Re7+ 56. Kd4 Rf7 { [%eval 17,52]}) 24. g4 Bb1 25. c5 dxc5 26. Rc1 c4 27. Bxc4+ Nxc4 28. Rxc4 Red8 29. Rxd8+ Rxd8 30. Rxc6 Bxf4 31. Bxf4 Bxa2 32. Ra6 Rd7 33. Rxa5 Bd5 34. Be3 Kf7 35. Bh6 Bb7 36. Rc5 Bd5 37. Kg3 Ke8 38. Ra5 Kf7 39. Bc1 Bb7 40. Rc5 Ke6 41. Rb5 Bd5 42. Rb6+ Rd6 43. Rb8 Rd7 44. Bg5 Kf7 45. Rb6 Be6 46. Kf4 Ra7 47. Rb5 Bd7 48. Rb8 Ba4 49. Rb6 h5 50. Rf6+ Kg7 51. Rd6 Rf7+ 52. Ke5 Rxf3 53. gxh5 { [%eval 27,52]}) 23... Nf7 24. h5 a4 25. Rh1 Bxf4 26. gxf4 Kf8 27. Rg1 Re6 28. c5 dxc5 29. Rb2 gxh5 30. Bc4 Rg6 31. Rh1 Nd6 32. Rxh5 Nxc4 {[%eval 38,51]}) 9. d5 cxd5 10. cxd5 a6 11. g4 h5 12. h3 b5 13. Nc1 Nb6 14. b3 ({Stockfish 130722: } 14. a3 Bd7 15. N1a2 Rc8 16. b3 Be8 17. Kd1 Nfd7 18. Rb1 h4 19. Be2 Nc5 20. Ke1 Rc7 21. Rg1 Kh8 22. g5 Qc8 23. Nd1 f5 24. gxf6 Bxf6 25. Nb4 Kh7 26. Nf2 Nbd7 27. Rg4 a5 28. Nc6 a4 29. bxa4 bxa4 30. Bc4 Rb7 31. Rxb7 Qxb7 32. Qb4 Qc7 33. Ke2 Nb3 34. Bxb3 axb3 35. Nd3 g5 36. Rg1 Bh5 37. Qxb3 Qc8 38. Na7 Qe8 39. Nf2 Qf7 40. Ng4 Bxg4 41. hxg4 Bd8 42. Rf1 h3 43. Qb4 Nc5 44. Nc6 h2 45. Bxc5 dxc5 46. Qxc5 Qf4 {[%eval 135,46]}) 14... b4 15. Nd1 a5 16. Nf2 Bb7 17. Bg5 Qe8 18. Bxf6 Bxf6 19. gxh5 Nxd5 20. hxg6 Nf4 21. Qxd6 fxg6 22. Bc4+ Kh8 23. h4 Rd8 24. Qb6 Qe7 25. Ncd3 Ng2+ 26. Ke2 Nf4+ 27. Nxf4 exf4 28. Rad1 Rb8 29. Qd6 Qxd6 30. Rxd6 Rbd8 31. Rb6 Bc8 32. Nd3 Kh7 33. Rb5 Ba6 34. Rxa5 Bxc4 35. bxc4 Bc3 36. h5 g5 37. Rg1 Rd4 38. Rgxg5 Rxc4 39. Ra7+ Kh8 40. Rc5 Rxc5 41. Nxc5 Rg8 42. Ne6 Be5 43. Ra5 Bc3 44. Kd3 1-0

In the competitive chess engine openings book world a book producer is unlikely to get caught a second time with the same loss in a live session. 13 moves falls into the area opening books will cover and the fact there are very few games that have been played with the line to 13.Nc1 confirms 13..Nb6 instead of ..Nc5 is unlikely to have the opportunity of ever being played in a competitive match, possibly because of 8..dxe5! instead of 8..Nbd7?

The merits of position learning would be far better demonstrated if a compelling argument could be supported that a popular opening line was bettered by playing move x rather than move y instead of choosing an obscure, unlikely to be played line that gives one side a tough struggle whichever move is selected! Using bigger hardware or longer time controls to provide deeper in game ply depths would help too!

Peter

[unofficial]

Firstly, that an opening book was being used that took the game to a cut off point that was a disadvantageous position for black.
...why wasn't the book line changed to ...
...Secondly, the opening line for Black is a bad choice...
...In the competitive chess engine openings book world...
...instead of choosing an obscure...

But there is no opening book here ! There is no good or bad opening choice too, engines are only at learning. The goal is to better play openings on tourney and here engines had only to learn this opening (reversed color = same disadvantage) :
[Event "Eman's learning"] [Site "deeds"] [Date "2021.05.24"] [Round "004508"] [White "eman"] [Black "eman"] [Result "*"] [ECO "E89"] [PlyCount "23"] 1. d4 Nf6 2. c4 g6 3. Nc3 Bg7 4. e4 d6 5. f3 O-O 6. Be3 e5 7. d5 c6 8. Nge2 cxd5 9. cxd5 Nbd7 10. Qd2 a6 11. g4 h5 12. h3 *

In that case why is the cut off point where the bad move choice is made at very fast time controls and shallow depths?

knowledge of chess engine trainer : TC 1m+1s was chosen at learning to target best performance of experience data at tourney with TC 3m+2s.
No shallow depth !

On my old Q9550 quad + 1Gb hash, in analysis mode Stockfish 130722 finds...

Sorry but we told about SF15 !

Therefore contrary to the statement Stockfish has no chance without learning feature it is all relative to the time controls and hardware being used. The use of shallow ply depth to demonstrate a characteristic is never a good or convincing starting point!
...The merits of position learning would be far better demonstrated...

As expected, denial of reality LOL

Using bigger hardware or longer time controls to provide deeper in game ply depths would help too!

No evidence SF is unbeatable at longer TC on supercomputer. Besides, i don't think it was the main purpose of SF. So we stay on evidence that without a learning feature, SF has no chance and people have to wait for episodic new nets to hope for better moves.

[kramnik]

There are still people testing engines with opening books...

There are a number of considerations that should be taken into account.

Firstly, that an opening book was being used that took the game to a cut off point that was a disadvantageous position for black. In that case why is the cut off point where the bad move choice is made at very fast time controls and shallow depths?

On my old Q9550 quad + 1Gb hash, in analysis mode Stockfish 130722 finds 13..Nb6 at 26 ply depth and sticks with it. In game mode it takes 4s longer but at the same ply depth of 26 changes to 13..Nb6 and sticks with it. Therefore the question is that after one loss with 13..Nc5, why wasn't the book line changed to 13..Nb6 that would perhaps have avoided all of those losses? The learning demonstration ignores the quickest route to correcting the 13th move that is by changing the book move selection. On my bigger hardware the Stockfish engine finds 13..Nb6 proportionately quicker.

Therefore contrary to the statement Stockfish has no chance without learning feature it is all relative to the time controls and hardware being used. The use of shallow ply depth to demonstrate a characteristic is never a good or convincing starting point!

Secondly, the opening line for Black is a bad choice. In the Sämisch variation of the King's Indian Black does well to completely avoid the demonstration line by playing 8..exd4 when Black has at least some decent fighting chances ...

[Event "Clichy"] [Site "Clichy"] [Date "1998.12.14"] [Round "3"] [White "Murey, Jacob"] [Black "Relange, Eloi"] [Result "1-0"] [ECO "E89"] [WhiteElo "2460"] [BlackElo "2490"] [Annotator "Grayson,Peter"] [PlyCount "87"] [EventDate "1998.05.16"] [EventType "tourn"] [EventRounds "9"] [EventCountry "FRA"] [EventCategory "7"] [SourceTitle "EXT 2000"] [Source "ChessBase"] [SourceDate "1999.11.16"] [SourceVersion "1"] [SourceVersionDate "1999.11.16"] [SourceQuality "1"] 1. d4 Nf6 2. c4 g6 3. Nc3 Bg7 4. e4 d6 5. f3 O-O 6. Be3 e5 7. Nge2 c6 8. Qd2 Nbd7 ({Stockfish 130722:} 8... exd4 9. Bxd4 (9. Nxd4 {Stockfish 130722:} d5 10. cxd5 (10. exd5 {Stockfish 130722:} cxd5 11. O-O-O Nc6 12. Bh6 Bxh6 13. Qxh6 Re8 14. Nc2 d4 15. g4 Bd7 16. Nb5 a6 17. Nbxd4 Qc7 18. Rg1 Nxd4 19. Rxd4 Bc6 20. Qf4 Qxf4+ 21. Rxf4 Nd7 22. Rg3 Kg7 23. Rd4 Ne5 24. f4 Nf3 25. Rd6 Nxh2 26. Bd3 Be4 27. g5 Bxd3 28. Rdxd3 Re2 29. Rge3 Rf2 30. Re4 Rc8 31. b4 Ng4 32. Kb2 h5 33. gxh6+ Kxh6 34. Kb3 Kg7 35. Ne3 Nxe3 36. Rdxe3 Rc6 37. c5 b6 38. Rc3 Rd2 39. Rec4 Re2 {[%eval 26,50]}) 10... cxd5 11. e5 Ne8 12. Bg5 Qd7 13. Nb3 Bxe5 14. Nc5 Bxc3 15. Qxc3 Qd6 16. Rd1 Nc6 17. Kf2 d4 18. Qc1 Bf5 19. Nxb7 Qc7 20. Nc5 Ng7 21. b3 h5 22. Bc4 Rfe8 23. Rhe1 Qxh2 24. Bf4 Qh4+ 25. Kg1 Qf6 26. Bb5 Rxe1+ 27. Rxe1 Rc8 28. Ba6 Rf8 29. Bc4 h4 30. Bg5 Qd6 31. Bxh4 Ne5 32. Bf1 d3 33. Qc3 f6 34. b4 Nh5 35. Bf2 Rc8 36. Qd4 Qxd4 {[%eval 24,50]}) 9... Nbd7 10. Rd1 a6 11. Nf4 b5 12. Be2 Qa5 13. b3 Re8 14. O-O Ne5 15. Ncd5 Qxd2 16. Nxf6+ Bxf6 17. Rxd2 Bg5 18. g3 f5 19. h4 (19. exf5 Bxf5 20. h4 Bh6 {transposes back into line} ) 19... Bh6 20. exf5 Bxf5 21. Kf2 bxc4 22. bxc4 a5 23. Bc3 ({Stockfish 130722: } 23. Be3 Rab8 ({Stockfish 130722:} 23... Nf7 24. Ng2 Bg7 25. g4 Bd7 26. Bd4 Bxd4+ 27. Rxd4 Reb8 28. Rd2 a4 29. Rc1 a3 30. Ne3 Rb2 31. Rcc2 Rab8 32. Nd1 R2b4 33. Rc3 Be6 34. c5 Rb2 35. Nxb2 axb2 36. Bd3 b1=Q 37. Bxb1 Rxb1 38. cxd6 Bd5 39. d7 Rb7 40. Rxd5 cxd5 41. Rc8+ Kg7 42. d8=Q Nxd8 43. Rxd8 Ra7 44. Rxd5 Rxa2+ 45. Kg3 h6 46. Rd7+ Kg8 47. Kf4 Re2 48. Rc7 Re8 49. Rc5 Re7 50. h5 Rf7+ 51. Kg3 g5 52. Re5 Kg7 53. Kf2 Ra7 54. Rb5 Rc7 55. Ke3 Re7+ 56. Kd4 Rf7 { [%eval 17,52]}) 24. g4 Bb1 25. c5 dxc5 26. Rc1 c4 27. Bxc4+ Nxc4 28. Rxc4 Red8 29. Rxd8+ Rxd8 30. Rxc6 Bxf4 31. Bxf4 Bxa2 32. Ra6 Rd7 33. Rxa5 Bd5 34. Be3 Kf7 35. Bh6 Bb7 36. Rc5 Bd5 37. Kg3 Ke8 38. Ra5 Kf7 39. Bc1 Bb7 40. Rc5 Ke6 41. Rb5 Bd5 42. Rb6+ Rd6 43. Rb8 Rd7 44. Bg5 Kf7 45. Rb6 Be6 46. Kf4 Ra7 47. Rb5 Bd7 48. Rb8 Ba4 49. Rb6 h5 50. Rf6+ Kg7 51. Rd6 Rf7+ 52. Ke5 Rxf3 53. gxh5 { [%eval 27,52]}) 23... Nf7 24. h5 a4 25. Rh1 Bxf4 26. gxf4 Kf8 27. Rg1 Re6 28. c5 dxc5 29. Rb2 gxh5 30. Bc4 Rg6 31. Rh1 Nd6 32. Rxh5 Nxc4 {[%eval 38,51]}) 9. d5 cxd5 10. cxd5 a6 11. g4 h5 12. h3 b5 13. Nc1 Nb6 14. b3 ({Stockfish 130722: } 14. a3 Bd7 15. N1a2 Rc8 16. b3 Be8 17. Kd1 Nfd7 18. Rb1 h4 19. Be2 Nc5 20. Ke1 Rc7 21. Rg1 Kh8 22. g5 Qc8 23. Nd1 f5 24. gxf6 Bxf6 25. Nb4 Kh7 26. Nf2 Nbd7 27. Rg4 a5 28. Nc6 a4 29. bxa4 bxa4 30. Bc4 Rb7 31. Rxb7 Qxb7 32. Qb4 Qc7 33. Ke2 Nb3 34. Bxb3 axb3 35. Nd3 g5 36. Rg1 Bh5 37. Qxb3 Qc8 38. Na7 Qe8 39. Nf2 Qf7 40. Ng4 Bxg4 41. hxg4 Bd8 42. Rf1 h3 43. Qb4 Nc5 44. Nc6 h2 45. Bxc5 dxc5 46. Qxc5 Qf4 {[%eval 135,46]}) 14... b4 15. Nd1 a5 16. Nf2 Bb7 17. Bg5 Qe8 18. Bxf6 Bxf6 19. gxh5 Nxd5 20. hxg6 Nf4 21. Qxd6 fxg6 22. Bc4+ Kh8 23. h4 Rd8 24. Qb6 Qe7 25. Ncd3 Ng2+ 26. Ke2 Nf4+ 27. Nxf4 exf4 28. Rad1 Rb8 29. Qd6 Qxd6 30. Rxd6 Rbd8 31. Rb6 Bc8 32. Nd3 Kh7 33. Rb5 Ba6 34. Rxa5 Bxc4 35. bxc4 Bc3 36. h5 g5 37. Rg1 Rd4 38. Rgxg5 Rxc4 39. Ra7+ Kh8 40. Rc5 Rxc5 41. Nxc5 Rg8 42. Ne6 Be5 43. Ra5 Bc3 44. Kd3 1-0

In the competitive chess engine openings book world a book producer is unlikely to get caught a second time with the same loss in a live session. 13 moves falls into the area opening books will cover and the fact there are very few games that have been played with the line to 13.Nc1 confirms 13..Nb6 instead of ..Nc5 is unlikely to have the opportunity of ever being played in a competitive match, possibly because of 8..dxe5! instead of 8..Nbd7?

The merits of position learning would be far better demonstrated if a compelling argument could be supported that a popular opening line was bettered by playing move x rather than move y instead of choosing an obscure, unlikely to be played line that gives one side a tough struggle whichever move is selected! Using bigger hardware or longer time controls to provide deeper in game ply depths would help too!

Peter

[unofficial]

As soon as they see "book" in a screenshot or in a PGN file they believe there is an opening book LOL

Whereas it's just the gui that adds "book" next to the opening moves. Moreover, when the moves come from an opening book, there is no "book" mention nor score nor depth nor time.

For them chess cannot exist without using an opening book to hide the weaknesses of the engines.

Hence their denial of reality when it comes to learning openings, training engines, etc.

[Xalomon]

The rating lists will use learning=off for engines with learning, so implementing this will not appear to make any difference in them.

[unofficial]

They don't even mention the net of each engine.
They don't cover all the eco codes too.
Not very reliable.

[Peter Grayson]

There are still people testing engines with opening books...

You really believe the creation of a learning file is anything more than an engine internal opening book? The learning file and its handling has similarity with its intent but with less features than offered by the Chessbase.ctg opening books tree structure that has been available since at least 1998 as an external and more user friendly method of openings and move management but also if desired, well beyond the opening phase. The .ctg tree format also provides for single positions or positions + extensions entries together with manual, database and variable live result learning that can be focused for any specific engine the end user wishes.

The Stockfish engine score profile confirms how unreliable it is in the early stages of the game with the number of draws being the final outcome of what was suggested to be a potential winning line in the early stages. The engine move choice noise levels increases with ply depth reduction thus the effect varies with hardware resource given to the engine for any given time control. There are also many examples of missed wins and missed draws particularly at fast time controls and low ply depths that would make engine reliant live learning of doubtful value.

The learning file tools provide for pgn file games to be read in and also for merging files thus in nature it offers some .ctg capability for learning that is absent from polyglot books. A good idea overall that may prove better than the simplistic polyglot format including the benefit of full functionality being transferable between different GUIs

Given the presentation and quality of the discussion in the posts seen here it quickly becomes evident as to why the SF Discord took the action they did. It requires no explanation!

Peter

[unofficial]

...SF Discord took the action they did...

As Peter is ashamed to say it clearly, i will do it for the record : SF Discord censored this idea ! Their moderator deleted it. Now we know that Peter sponsors censorship.

@hurnavich I hope this is not the same guy who wanted to access our engines learning room !? Else i better understand why he didn't ask it directly and why nobody sponsored him.

[Peter Grayson]

...SF Discord took the action they did...

As Peter is ashamed to say it clearly, i will do it for the record : SF Discord censored this idea ! Their moderator deleted it. Now we know that Peter sponsors censorship.

@hurnavich I hope this is not the same guy who wanted to access our engines learning room !? Else i better understand why he didn't ask it directly and why nobody sponsored him.

Your comments referring to my being ashamed and that I sponsor censorship border on slander. It is quite clear you have no idea of how to conduct yourself in public debate and you are more familiar with abrasive social media gutter nonsense than intelligent reasoned argument and given those comments it is understandable why the topic here was accessible by password only to protect you from reaction to your abrasive nature. It will probably not be long before you are censured here to.

Peter

[unofficial]

@hurnavich so i think it was the same guy. Unlike Anton101 and ZamChess, this time we avoided to give access to a learning hater just in time.

LS5 @ 15000 60+1 th 7 :

Code: Select all

   # PLAYER                             :  RATING  ERROR   POINTS  PLAYED   (%)      W      D      L  D(%)  OppAvg  OppN
   1 Eman experience                    :       0   ----  31856.5   60000  53.1  19422  24869  15709  41.4     -28     4
   2 brainlearn 16.1 nn-6877cd24400e    :     -17      5   7216.5   15000  48.1   4190   6053   4757  40.4       0     1
   3 JudaS 110422 nn-6877cd24400e       :     -22      5   7142.5   15000  47.6   3918   6449   4633  43.0       0     1
   4 HypnoS 110422 nn-6877cd24400e      :     -34      5   6934.0   15000  46.2   3729   6410   4861  42.7       0     1
   5 stockfish 15.0                     :     -39      4   6850.5   15000  45.7   3872   5957   5171  39.7       0     1

White advantage = -182.48 +/- 1.12
Draw rate (equal opponents) = 50.00 % +/- 0.00

LS6 @ 13000 60+1 th 7 :

Code: Select all

   # PLAYER                             :  RATING  ERROR   POINTS  PLAYED   (%)      W      D     L  D(%)  OppAvg  OppN
   1 Eman experience                    :       0   ----  27486.0   52000  52.9  10987  32998  8015  63.5     -21     4
   2 JudaS 090622 nn-3c0aa92af1da       :      -6      4   6396.0   13000  49.2   2187   8418  2395  64.8       0     1
   3 brainlearn 17.1 nn-3c0aa92af1da    :     -13      4   6264.0   13000  48.2   2102   8324  2574  64.0       0     1
   4 HypnoS 090622 nn-3c0aa92af1da      :     -29      4   5983.0   13000  46.0   1867   8232  2901  63.3       0     1
   5 stockfish 15.0                     :     -36      4   5871.0   13000  45.2   1859   8024  3117  61.7       0     1

White advantage = -76.44 +/- 1.10
Draw rate (equal opponents) = 50.00 % +/- 0.00

We laugh seeing elo on SF regression tests...

[Hurnavich]

Peter has always reasoned with politeness and logical debate in all the time I have known him.
I think you yourself are over reacting and been defencive were I see no need for that
Peter is only putting forward an opinion in a intelligent debate.
No one has to agree or disagree putting forward eg tests and debate can be healthy
I see no insult or attack in the debate from Peters side.

@hurnavich so i think it was the same guy. Unlike Anton101 and ZamChess, this time we avoided to give access to a learning hater just in time.

LS5 @ 15000 60+1 th 7 :

Code: Select all

   # PLAYER                             :  RATING  ERROR   POINTS  PLAYED   (%)      W      D      L  D(%)  OppAvg  OppN
   1 Eman experience                    :       0   ----  31856.5   60000  53.1  19422  24869  15709  41.4     -28     4
   2 brainlearn 16.1 nn-6877cd24400e    :     -17      5   7216.5   15000  48.1   4190   6053   4757  40.4       0     1
   3 JudaS 110422 nn-6877cd24400e       :     -22      5   7142.5   15000  47.6   3918   6449   4633  43.0       0     1
   4 HypnoS 110422 nn-6877cd24400e      :     -34      5   6934.0   15000  46.2   3729   6410   4861  42.7       0     1
   5 stockfish 15.0                     :     -39      4   6850.5   15000  45.7   3872   5957   5171  39.7       0     1

White advantage = -182.48 +/- 1.12
Draw rate (equal opponents) = 50.00 % +/- 0.00

LS6 @ 13000 60+1 th 7 :

Code: Select all

   # PLAYER                             :  RATING  ERROR   POINTS  PLAYED   (%)      W      D     L  D(%)  OppAvg  OppN
   1 Eman experience                    :       0   ----  27486.0   52000  52.9  10987  32998  8015  63.5     -21     4
   2 JudaS 090622 nn-3c0aa92af1da       :      -6      4   6396.0   13000  49.2   2187   8418  2395  64.8       0     1
   3 brainlearn 17.1 nn-3c0aa92af1da    :     -13      4   6264.0   13000  48.2   2102   8324  2574  64.0       0     1
   4 HypnoS 090622 nn-3c0aa92af1da      :     -29      4   5983.0   13000  46.0   1867   8232  2901  63.3       0     1
   5 stockfish 15.0                     :     -36      4   5871.0   13000  45.2   1859   8024  3117  61.7       0     1

White advantage = -76.44 +/- 1.10
Draw rate (equal opponents) = 50.00 % +/- 0.00

We laugh seeing elo on SF regression tests...

[mehmet1921]

Yesterday i post few ideas on discord in the "ideas-only" of the SF channel. Then i tried to explain them in the "ideas-discuss". And this morning i saw my idea about "add a learning feature to SF" was censored !? Yet SF without this feature has no chance !

I'm not surprised that some of your ideas about learning feature were banned at Stockfis Discord Channel. Because I was surprised to see how conservative some of the active people on the Lc0 Discord Channel a few years ago. I can call Lc0 as the most modern chess engine. I changed many settings of Lc0 and introduced it as Lc0 Kayra. Not counting my own tests, Lc0 Kayra had better results in all indepence tests against Stockfish according to Lc0 Master. I made a suggestion to the Lc0 Discord Group. If you do not accept these test results, test Lc0 Kayra against Stockfish and decide based on these test results. They didn' t accept my proposal and Lc0 could not play against Stockfish with Kayra settings at TCEC Super Final and that year Stockfish become the TCEC champion. Lc0 Discord channel was no different from the old American towns where foreigners were not liked at that time for me. Of course at that time there were some people on the Lc0 Discord Channel who supported me.

Unfortunately, there are many people who are closed to innovative ideas at Stockfish and Lc0 Discord Channels. I and many others who are not biased towards innovative ideas know how important your work is. The importance of learning files will certainly be understood much better in the future.

[kramnik]

I didn' mention learning files...

There are still people testing engines with opening books...

You really believe the creation of a learning file is anything more than an engine internal opening book? The learning file and its handling has similarity with its intent but with less features than offered by the Chessbase.ctg opening books tree structure that has been available since at least 1998 as an external and more user friendly method of openings and move management but also if desired, well beyond the opening phase. The .ctg tree format also provides for single positions or positions + extensions entries together with manual, database and variable live result learning that can be focused for any specific engine the end user wishes.

The Stockfish engine score profile confirms how unreliable it is in the early stages of the game with the number of draws being the final outcome of what was suggested to be a potential winning line in the early stages. The engine move choice noise levels increases with ply depth reduction thus the effect varies with hardware resource given to the engine for any given time control. There are also many examples of missed wins and missed draws particularly at fast time controls and low ply depths that would make engine reliant live learning of doubtful value.

The learning file tools provide for pgn file games to be read in and also for merging files thus in nature it offers some .ctg capability for learning that is absent from polyglot books. A good idea overall that may prove better than the simplistic polyglot format including the benefit of full functionality being transferable between different GUIs

Given the presentation and quality of the discussion in the posts seen here it quickly becomes evident as to why the SF Discord took the action they did. It requires no explanation!

Peter