Slipper: no progress

[Henk]

Gradients should be correct now. But still problem of how many layers cells to use, hyperparameters. optimization method, input representation, which type of relu, batch normalization, etcetera.

After each change effecting number of weights you can remove weight file and start all over again.

Also speed problem. One game costing half an hour on my computer.

Saturday I noticed all values of cells in hidden layer were zero. So I am now using leaky relu's.


[pgn]
1. d2-d4 { 13.700 0.00 } { }
1. a7-a6 { 15.861 -0.13 } { }
2. f2-f3 { 22.855 0.00 } { }
2. a6-a5 { 17.059 0.00 } { }
3. h2-h4 { 21.653 0.00 } { }
3. a5-a4 { 19.402 0.00 } { }
4. g2-g3 { 21.934 0.00 } { }
4. a4-a3 { 17.431 0.00 } { }
5. Ng1-h3 { 37.044 0.00 } { }
5. a3-b2 { 32.227 0.00 } { }
6. Rh1-g1 { 41.415 0.00 } { }
6. Ra8-a2 { 31.256 0.00 } { }
7. Ra1-a2 { 46.023 0.00 } { }
7. b2-c1Q { 26.441 0.00 } { }
8. Ra2-a5 { 35.026 0.00 } { }
8. Qc1-d1 { 34.898 0.00 } { }
9. Ke1-d1 { 35.921 0.00 } { }
9. b7-b6 { 29.338 0.00 } { }
10. Nb1-c3 { 34.484 0.00 } { }
10. b6-a5 { 35.737 0.00 } { }
11. Nc3-b5 { 33.347 0.00 } { }
11. a5-a4 { 41.422 0.00 } { }
12. Nb5-a7 { 42.026 0.00 } { }
12. a4-a3 { 36.543 0.00 } { }
13. Na7-c8 { 38.650 0.00 } { }
13. Qd8-c8 { 33.889 0.00 } { }
14. Kd1-e1 { 39.848 0.00 } { }
14. a3-a2 { 37.774 0.00 } { }
15. Bf1-g2 { 36.218 0.00 } { }
15. a2-a1Q { 47.938 0.00 } { }
16. Ke1-d2 { 44.191 0.00 } { }
16. Qa1-g1 { 39.514 0.00 } { }
17. Kd2-d3 { 41.248 0.00 } { }
17. Qg1-g2 { 39.039 0.00 } { }
18. Kd3-d2 { 27.750 0.00 } { }
18. Qg2-h3 { 44.184 0.00 } { }
19. Kd2-d1 { 24.626 0.00 } { }
19. Qh3-h4 { 36.833 0.00 } { }
20. c2-c3 { 21.977 0.00 } { }
20. Qh4-d4 { 34.087 0.00 } { }
21. c3-d4 { 23.183 0.00 } { }
21. c7-c6 { 35.118 0.00 } { }
22. Kd1-e1 { 20.420 0.00 } { }
22. c6-c5 { 38.460 0.00 } { }
23. d4-c5 { 19.879 0.00 } { }
23. Qc8-c5 { 33.959 0.00 } { }
24. Ke1-f1 { 17.486 0.00 } { }
24. Qc5-c1 { 31.023 0.00 } { }
25. Kf1-g2 { 14.940 0.00 } { }
25. Qc1-a1 { 29.813 0.00 } { }
26. Kg2-h3 { 13.085 0.00 } { }
26. Qa1-b1 { 39.374 0.00 } { }
27. Kh3-g4 { 13.277 0.00 } { }
27. Qb1-a1 { 42.342 0.00 } { }
28. Kg4-h3 { 13.761 0.00 } { }
28. Qa1-b1 { 33.322 0.00 } { }
29. Kh3-g4 { 13.705 0.00 } { }
29. Qb1-a1 { 26.646 0.00 } { }
30. Kg4-h3 { 13.564 0.00 } { }
[/pgn]

[Henk]

[pgn]
[FEN "rnbqkbnr/1ppppppp/8/p7/3P4/7P/PPP1PPP1/RNBQKBNR w KQkq - 0 1"]
1. g2-g4 { 11.645 0.00 } { }
1. f7-f6 { 11.689 0.00 } { }
2. g4-g5 { 11.741 0.00 } { }
2. a5-a4 { 11.793 0.00 } { }
3. g5-f6 { 11.843 0.00 } { }
3. Ra8-a7 { 11.893 0.00 } { }
4. f6-g7 { 11.931 0.00 } { }
4. c7-c5 { 11.969 0.00 } { }
5. d4-c5 { 12.003 0.00 } { }
5. b7-b5 { 12.040 0.00 } { }
6. c5-b6 { 12.083 0.00 } { }
6. h7-h6 { 12.126 0.00 } { }
7. Bc1-h6 { 12.167 0.00 } { }
7. Qd8-b6 { 12.209 0.00 } { }
8. g7-f8R { 12.244 0.00 } { }
[/pgn]

[Henk]

[pgn]
1. f2-f4 { 33.499 0.00 } { }
1. d7-d5 { 44.616 0.00 } { }
2. f4-f5 { 27.831 0.00 } { }
2. Nb8-c6 { 45.451 0.00 } { }
3. g2-g4 { 37.553 0.00 } { }
3. Nc6-d4 { 50.880 0.00 } { }
4. h2-h4 { 39.337 0.00 } { }
4. c7-c6 { 26.598 -0.14 } { }
5. e2-e4 { 38.336 0.05 } { }
5. Qd8-c7 { 26.603 -0.97 } { }
6. e4-d5 { 2.692 -0.98 } { }
6. Qc7-g3 { 2.090 -0.99 } { }
[/pgn]

[tpoppins]

Slipper AKA Skipper -- Putting the "I have no idea what I'm doing" concept on a Solid Scientific Basis.

[zullil]

Slipper AKA Skipper -- Putting the "I have no idea what I'm doing" concept on a Solid Scientific Basis.

Don't be too harsh. Whatever it is, it seems to play legal moves. At least for now, until he tries to implement some new feature.

[Rodolfo Leoni]

Slipper AKA Skipper -- Putting the "I have no idea what I'm doing" concept on a Solid Scientific Basis.

Don't be too harsh. Whatever it is, it seems to play legal moves. At least for now, until he tries to implement some new feature.

If it learns that a checkmate is painful it's a huge progress IMO.

[tpoppins]

Don't be too harsh.

Wouldn't dream of it, not to good ol' Henk. Besides possessing a quirky, wry sense of humor he also has a keen sense of proportion. You'd never see him claiming to be world's best programmer, being stronger than Stockfish or that French Defence is busted and 1.d4 is a weak move. Delicate trolling is a subtle art, and Talkchess would be a duller place if (God forbid) S(k/l)ipper and its author suddenly disappeared.

[tpoppins]

I had a closer look at the last Slipper game and suddenly had a moment of inspiration
[pgn]
1. f2-f4!? { 33.499 0.00 } {A good move. According to the world's leading opening-expert Prof. Flowerov, 1.c4 is White's only serious try for an opening advantage. Skipper's brilliant idea to mirror that move on the kingside conclusively reveals that there may be an alternative. Besides laying an indirect claim to the center, this pawn advance portends the steamrolling pawn storm to follow.}
1. d7-d5?! { 44.616 0.00 } {Aware that his options have been severely limited (e.g. 1...e5?? 2.fxe5 and White has an overwhelming pawn preponderance in the center), Black nevertheless proceeds in the "classical" fashion. 1...Nf6 or, better yet, 1...Nc6 were the only plausible attempts to counter White's onslaught.}
2. f4-f5!! { 27.831 0.00 } {An extremely strong move that you won't find in any opening reference and few GMs would choose over the board even with computer aid. The main strength of this move is that it is weak! That is to say, it's weak only when viewed through the clouded lens of the routine classical dogma; however, a deep closer look affords a glimpse of strategy beyond human understanding. Of course, any beginner can observe the direct benefits of this apparently headlong advance: the severe cramping of Black's kingside, the lockout of the black light-square bishop etc.; however, the true power of this move is not revealed until several moves later. And isn't there a parallel between it and Prof. Flowerov's revolutionary attempt at refuting the Dutch Defence with 2.d5? Brilliant minds -- be they protein or silicone -- think alike!}
2. Nb8-c6! { 45.451 0.00 } {A good move. Black finally thinks of attending to his development -- albeit it must be admitted he's a move late -- and wisely avoids falling in with White's plans with 2...Bxf5, as that would merely open new lines for White's attack and after 3.g4! would allow White to achieve the all-powerful fianchetto of his light-square bishop (normally taking two moves) without loss of time. Needless to say, a pawn or two are of no consequence in a strategy of scope as broad as White demonstrates in this game.}
3. g2-g4! { 37.553 0.00 } {A good move -- the aforementioned pawn storm begins! It's amazing how making seemingly natural moves and no obvious mistakes Black soon finds himself completely at the mercy of his opponent.}
3. Nc6-d4 { 50.880 0.00 } {This knight leap would merit an exclamation mark if it were not for the fact that Black has no better move at his disposal. The strong position of the centralized knight appears to offer Black a glimmer of hope, however bitter disappointment awaits him...}
4. h2-h4! { 39.337 0.00 } {A good move! With the center being static White is completely correct to ignore Black's demonstration there and proceeds to build up his kingside roller.}
4. c7-c6? { 26.598 -0.14 } {A good prophylactic move made for the wrong reasons. As we will presently see, this innocent pawn advance is the camouflage for Black's cunning plan that is naturally doomed at its inception due to the sheer superiority of White's position. 4...Nf6, developing at least one kingside piece, was the last chance to offer some resistance before getting pushed off the board completely. }
5. e2-e4! { 38.336 0.05 } {A good move! This thematic pawn break completely unhinges Black's position in the place he expected the least -- the center! Here's the ultimate proof of the age-old-yet-forgotten axiom: thou shalt not embark upon operations in the center 'ere thine position on the flanks is secured definitely.}
5. Qd8-c7? { 26.603 -0.97 } {Blissfully blind, Black goes on with his plan. 5...Qc7 move does nothing to halt the white kingside steamroller and less still to aid Black's development. At this point, however, it is hard to suggest a rational alternative. For example, the natural 5...Nf6 would be help as much as heat pad would help a corpse, as after 6.e5! and 7.e6! White would have a winning attack. }
6. e4-d5! { 2.692 -0.98 } {A good move! Incomprehensible as it might look it is imbued with such strength as to cause the game to end in a few moves.}
6. Qc7-g3 { 2.090 -0.99 } {Desperation! Realizing that his game is lost in any case, Black sends out his queen on a one-way journey deep into the enemy camp. At any rate, he deserves credit for deciding to end the game with a flourish rather than allowing himself to get squeezed to death... }
[/pgn]

Help! I cannot continue to demonstrate the game's brilliant finale as Chessbase won't let me enter any more moves! Is this a known bug?

[carldaman]

I had a closer look at the last Slipper game and suddenly had a moment of inspiration
[pgn]
1. f2-f4!? { 33.499 0.00 } {A good move. According to the world's leading opening-expert Prof. Flowerov, 1.c4 is White's only serious try for an opening advantage. Skipper's brilliant idea to mirror that move on the kingside conclusively reveals that there may be an alternative. Besides laying an indirect claim to the center, this pawn advance portends the steamrolling pawn storm to follow.}
1. d7-d5?! { 44.616 0.00 } {Aware that his options have been severely limited (e.g. 1...e5?? 2.fxe5 and White has an overwhelming pawn preponderance in the center), Black nevertheless proceeds in the "classical" fashion. 1...Nf6 or, better yet, 1...Nc6 were the only plausible attempts to counter White's onslaught.}
2. f4-f5!! { 27.831 0.00 } {An extremely strong move that you won't find in any opening reference and few GMs would choose over the board even with computer aid. The main strength of this move is that it is weak! That is to say, it's weak only when viewed through the clouded lens of the routine classical dogma; however, a deep closer look affords a glimpse of strategy beyond human understanding. Of course, any beginner can observe the direct benefits of this apparently headlong advance: the severe cramping of Black's kingside, the lockout of the black light-square bishop etc.; however, the true power of this move is not revealed until several moves later. And isn't there a parallel between it and Prof. Flowerov's revolutionary attempt at refuting the Dutch Defence with 2.d5? Brilliant minds -- be they protein or silicone -- think alike!}
2. Nb8-c6! { 45.451 0.00 } {A good move. Black finally thinks of attending to his development -- albeit it must be admitted he's a move late -- and wisely avoids falling in with White's plans with 2...Bxf5, as that would merely open new lines for White's attack and after 3.g4! would allow White to achieve the all-powerful fianchetto of his light-square bishop (normally taking two moves) without loss of time. Needless to say, a pawn or two are of no consequence in a strategy of scope as broad as White demonstrates in this game.}
3. g2-g4! { 37.553 0.00 } {A good move -- the aforementioned pawn storm begins! It's amazing how making seemingly natural moves and no obvious mistakes Black soon finds himself completely at the mercy of his opponent.}
3. Nc6-d4 { 50.880 0.00 } {This knight leap would merit an exclamation mark if it were not for the fact that Black has no better move at his disposal. The strong position of the centralized knight appears to offer Black a glimmer of hope, however bitter disappointment awaits him...}
4. h2-h4! { 39.337 0.00 } {A good move! With the center being static White is completely correct to ignore Black's demonstration there and proceeds to build up his kingside roller.}
4. c7-c6? { 26.598 -0.14 } {A good prophylactic move made for the wrong reasons. As we will presently see, this innocent pawn advance is the camouflage for Black's cunning plan that is naturally doomed at its inception due to the sheer superiority of White's position. 4...Nf6, developing at least one kingside piece, was the last chance to offer some resistance before getting pushed off the board completely. }
5. e2-e4! { 38.336 0.05 } {A good move! This thematic pawn break completely unhinges Black's position in the place he expected the least -- the center! Here's the ultimate proof of the age-old-yet-forgotten axiom: thou shalt not embark upon operations in the center 'ere thine position on the flanks is secured definitely.}
5. Qd8-c7? { 26.603 -0.97 } {Blissfully blind, Black goes on with his plan. 5...Qc7 move does nothing to halt the white kingside steamroller and less still to aid Black's development. At this point, however, it is hard to suggest a rational alternative. For example, the natural 5...Nf6 would be help as much as heat pad would help a corpse, as after 6.e5! and 7.e6! White would have a winning attack. }
6. e4-d5! { 2.692 -0.98 } {A good move! Incomprehensible as it might look it is imbued with such strength as to cause the game to end in a few moves.}
6. Qc7-g3 { 2.090 -0.99 } {Desperation! Realizing that his game is lost in any case, Black sends out his queen on a one-way journey deep into the enemy camp. At any rate, he deserves credit for deciding to end the game with a flourish rather than allowing himself to get squeezed to death... }
[/pgn]

Help! I cannot continue to demonstrate the game's brilliant finale as Chessbase won't let me enter any more moves! Is this a known bug?

ROFLOL - very funny indeed

[Ovyron]

Help! I cannot continue to demonstrate the game's brilliant finale as Chessbase won't let me enter any more moves! Is this a known bug?

Yes! The Chessbase GUI doesn't yet support Stanley Random Chess!