KID(Sämisch Variation) better as Black than average opening?

[Laskos]

Stockfish wouldn't go by itself into this KID opening:

1. d4 Nf6
2. c4 g6
3. Nc3 Bg7
4. e4 d6
5. f3

[d]rnbqk2r/ppp1ppbp/3p1np1/8/2PPP3/2N2P2/PP4PP/R1BQKBNR b KQkq - 0 1[/d]

The evaluation of Stockfish 8 is below par as Black:

Analysis by Stockfish 8 64 BMI2:

5...0-0 6.Be3 a6 7.h4 e5 8.dxe5 dxe5 9.Nge2 Be6 10.Qxd8 Rxd8 11.Nd5 Bxd5 12.0-0-0 c6 13.Nc3 Nbd7 14.cxd5 cxd5 15.Nxd5 Nxd5 16.Rxd5 Nf8 17.Kb1 b5 18.Rxd8 Rxd8 19.a4 bxa4 20.Bxa6 Ne6 21.Bb5 Bf8 22.Bxa4 Nf4 23.g3 Rd3 24.Bxf4 exf4
+/= (0.35) Depth: 36/50 00:17:54 5487MN
(Kai, 12.03.2017)

From this opening I built a 5-mover opening book (EPD file) by a method which I will describe later. This "Sämisch" EPD file of 300+ openings can be found here:
http://s000.tinyupload.com/?file_id=612 ... 1070438816
From these openings, I self-played Stockfish dev. for 1000 games at 10''+0.1''. The results from White POV are:

Code: Select all

Games         1-0   =-=   0-1   White perf.
 1000         235   574   191     52.2%

The PGN can be found here:
http://s000.tinyupload.com/?file_id=844 ... 8794331058
White performance for KID (Sämisch Variation) is 52.2%. It already strikes as not that high compared to 0.35 eval of Stockfish.

Then, to see if I am not doing something wrong with the opening book building, I took 8moves_GM.pgn opening suite of GM games to move 8, and built the same kind of 5-mover book from those 8-move positions. This "book" represents some sort of compilation of typical GM opening positions. Then, the result in Stockfish dev. self-play 1000 games from this representative sample from White POV was:

Code: Select all

Games         1-0   =-=   0-1   White perf.
 1000         309   518   173     56.8%

White performance is here 56.8%, identical to what I get directly form 8moves_GM.pgn book. The book building doesn't distort the result and it is much higher for White here than for that KID.

So, KID (Sämisch Variation) seems to be better for Black than an average GM opening according to Stockfish fast self-play, although Stockfish doesn't like it and would not go into it unforced (by its eval).

[carldaman]

SF does not like to go into the KID on its own, despite being able to play it very well from the Black side, (beyond move 4, at least). It is encumbered by a hole in its eval, not liking to fianchetto the KB for some reason. The SF developers need to fix this flaw in its eval.

CL

[jdart]

The Sämisch used to be quite popular but I have not seen it much lately. I think White now prefers other systems against the KID.

But one thing to watch out for is that this is an opening with a 50-year+ history. If you use a big PGN database you will have a lot of lines in it that are now discarded, and that will skew the results a bit.

After 6. .. a6, the Noomen book likes Qd2. A main line is:

1. d4 Nf6 2. c4 g6 3. Nc3 Bg7 4. e4 d6 5. f3 O-O 6. Be3 a6 7. Qd2 Nc6 8. Nge2 Rb8 9. Rc1

which seems to be fine for White.

--Jon

[Laskos]

SF does not like to go into the KID on its own, despite being able to play it very well from the Black side, (beyond move 4, at least). It is encumbered by a hole in its eval, not liking to fianchetto the KB for some reason. The SF developers need to fix this flaw in its eval.

CL

That's exactly why I fooled around with KID, not only Stockfish, many engines won't go into KID. Since at least 15 years ago, engines dislike King's Indian Defense and they like to grab space advantage. I was surprised that although Stockfish gives bad eval to almost all KID positions, in self-play they perform not that bad. Probably in self-play long term plans might unveil more pronouncedly than in 30-40 ply analysis from initial KID position.

[Laskos]

The Sämisch used to be quite popular but I have not seen it much lately. I think White now prefers other systems against the KID.

But one thing to watch out for is that this is an opening with a 50-year+ history. If you use a big PGN database you will have a lot of lines in it that are now discarded, and that will skew the results a bit.

After 6. .. a6, the Noomen book likes Qd2. A main line is:

1. d4 Nf6 2. c4 g6 3. Nc3 Bg7 4. e4 d6 5. f3 O-O 6. Be3 a6 7. Qd2 Nc6 8. Nge2 Rb8 9. Rc1

which seems to be fine for White.

--Jon

Yes, the databases can be sometimes skewed, especially with sometime ago popular openings. That Noomen line in Sämisch is fine for White, with 58.1% performance in 1000 self-games from a built book, but I am not sure Black will go into it. I have done the same for several KID popular openings, 1000 Stockfish self-games to see White performance. It should be compared to 56.8 performance from general 8-move GM openings.

KID Classical 57.4% 1.d4 Nf6 2.c4 g6 3.Nc3 Bg7 4.e4 d6 5.Nf3 0-0 6.Be2 e5
[d]rnbq1rk1/ppp2pbp/3p1np1/4p3/2PPP3/2N2N2/PP2BPPP/R1BQK2R w KQ - 0 6
KID Main Line 62.8% 1.d4 Nf6 2.c4 g6 3.Nc3 Bg7 4.e4 d6 5.Nf3 0-0 6.Be2 e5 7.0-0 Nc6 8.d5 Ne7
[d]r1bq1rk1/ppp1npbp/3p1np1/3Pp3/2P1P3/2N2N2/PP2BPPP/R1BQ1RK1 w - - 0 8
KID Petrosian System 58.7% 1.d4 Nf6 2.c4 g6 3.Nc3 Bg7 4.e4 d6 5.Nf3 0-0 6.Be2 e5 7.d5
[d]rnbq1rk1/ppp2pbp/3p1np1/3Pp3/2P1P3/2N2N2/PP2BPPP/R1BQK2R b KQ - 0 7
KID Gligoric System 55.8% 1.d4 Nf6 2.c4 g6 3.Nc3 Bg7 4.e4 d6 5.Nf3 0-0 6.Be2 e5 7.Be3
[d]rnbq1rk1/ppp2pbp/3p1np1/4p3/2PPP3/2N1BN2/PP2BPPP/R2QK2R b KQ - 0 7
KID Classical Na6 60.0% 1.d4 Nf6 2.c4 g6 3.Nc3 Bg7 4.e4 d6 5.Nf3 0-0 6.Be2 e5 7.0-0 Na6 8.Be3 Ng4 9.Bg5 Qe8!
[d]r1b1qrk1/ppp2pbp/n2p2p1/4p1B1/2PPP1n1/2N2N2/PP2BPPP/R2Q1RK1 w - - 0 10

KID Sämisch 52.2% 1.d4 Nf6 2.c4 g6 3.Nc3 Bg7 4.e4 d6 5.f3
[d]rnbqk2r/ppp1ppbp/3p1np1/8/2PPP3/2N2P2/PP4PP/R1BQKBNR b KQkq - 0 5
KID Sämisch (Noomen) 58.1% 1.d4 Nf6 2.c4 g6 3.Nc3 Bg7 4.e4 d6 5.f3 O-O 6.Be3 a6 7.Qd2 Nc6 8.Nge2 Rb8 9.Rc1
[d]1rbq1rk1/1pp1ppbp/p1np1np1/8/2PPP3/2N1BP2/PP1QN1PP/2R1KB1R b K - 0 9

KID Averbach 55.6% 1.d4 Nf6 2.c4 g6 3.Nc3 Bg7 4.e4 d6 5.Be2 0-0 6.Bg5
[d]rnbq1rk1/ppp1ppbp/3p1np1/6B1/2PPP3/2N5/PP2BPPP/R2QK1NR b KQ - 0 6

The exaggerated eval of Stockfish averaging 0.40 or so for these openings is not confirmed in self-games. Aside "Main Line", they seem playable as black by Stockfish.

[Graham Banks]

.........So, KID (Sämisch Variation) seems to be better for Black than an average GM opening according to Stockfish fast self-play, although Stockfish doesn't like it and would not go into it unforced (by its eval).

I avoided f3 in my handmade book. I have the following lines:

1. d4 Nf6 2. c4 g6 3. Nf3 Bg7 4. g3 d5 5. cxd5 Nxd5 6. Bg2 Nb6 7. Nc3 *
1. d4 Nf6 2. c4 g6 3. Nf3 Bg7 4. e3 c5 5. d5 O-O 6. Be2 e6 7. Nc3 exd5 8. cxd5 d6 9. Nd2 Re8 10. O-O *
1. d4 Nf6 2. c4 g6 3. Nf3 Bg7 4. Nc3 d5 5. Qb3 dxc4 6. Qxc4 O-O 7. Bf4 Bf5 *
1. d4 Nf6 2. c4 g6 3. Nf3 Bg7 4. Nc3 d5 5. h4 O-O 6. cxd5 Nxd5 7. h5 c5 8. hxg6 hxg6 9. Bh6 Bxh6 10. Rxh6 Nxc3 11. bxc3 cxd4 12. Qd2 Kg7 *
1. d4 Nf6 2. c4 g6 3. Nf3 Bg7 4. Nc3 d5 5. Qa4+ Bd7 6. Qb3 Bc6 *
1. d4 Nf6 2. c4 g6 3. Nf3 Bg7 4. Nc3 d5 5. Qa4+ Bd7 6. Qb3 dxc4 7. Qxc4 O-O 8. Bf4 Bf5 *
1. d4 Nf6 2. c4 g6 3. Nf3 Bg7 4. Nc3 d5 5. Qa4+ Bd7 6. Qb3 dxc4 7. Qxc4 O-O 8. Bf4 Na6 9. e4 c5 10. e5 Be6 *
1. d4 Nf6 2. c4 g6 3. Nf3 Bg7 4. Nc3 d5 5. Qa4+ Bd7 6. Qb3 dxc4 7. Qxc4 O-O 8. Bf4 c6 *
1. d4 Nf6 2. c4 g6 3. Nf3 Bg7 4. Nc3 d5 5. Qa4+ Bd7 6. Qb3 dxc4 7. Qxc4 a6 8. Bf4 Nc6 *
1. d4 Nf6 2. c4 g6 3. Nf3 Bg7 4. Nc3 d5 5. cxd5 Nxd5 6. Bg5 c5 *
1. d4 Nf6 2. c4 g6 3. Nf3 Bg7 4. Nc3 d5 5. cxd5 Nxd5 6. Qa4+ Nc6 7. e4 Nb6 8. Qc2 e5 9. dxe5 O-O *
1. d4 Nf6 2. c4 g6 3. Nf3 Bg7 4. Nc3 d5 5. cxd5 Nxd5 6. Qa4+ Nc6 7. e4 Nb6 8. Qc2 Bxd4 *
1. d4 Nf6 2. c4 g6 3. Nf3 Bg7 4. Nc3 d5 5. cxd5 Nxd5 6. Qa4+ Bd7 *
1. d4 Nf6 2. c4 g6 3. Nf3 Bg7 4. Nc3 d5 5. cxd5 Nxd5 6. e4 Nxc3 7. bxc3 c5 8. h3 *
1. d4 Nf6 2. c4 g6 3. Nf3 Bg7 4. Nc3 d5 5. cxd5 Nxd5 6. e4 Nxc3 7. bxc3 c5 8. Be3 Bg4 *
1. d4 Nf6 2. c4 g6 3. Nf3 Bg7 4. Nc3 d5 5. cxd5 Nxd5 6. e4 Nxc3 7. bxc3 c5 8. Rb1 O-O 9. Be2 Qc7 *

[jdart]

In a real game, outside of engine matches that use a limited book, you are going to have to assume you are playing a booked-up opponent. So evals in the early opening don't matter. The real test is how do you do against the opponent's strongest lines, and what your eval looks like when you exit book.

I find it helpful to look at what the correspondence players play, for example with the ChessBase Correspondence Database. KID is still reasonably popular there, with a variety of systems being played. So I'd agree it's playable even if the opponent is well prepared.

I think you didn't mention

1. d4 Nf6 2. c4 g6 3. Nc3 Bg7 4. e4 d6 5. Nf3 O-O 6. Be2 e5 7. O-O Nc6 8. d5 Ne7 9. b4 Nh5

(E97), which is one of the better white tries.

However, my engine's book avoids KID with Black at the default book selectivity settings, although it may possibly transpose into it. I have just seen it come out of book too many times with a disadvantage. If it does play it, it especially avoids the Classical by playing 7. .. Nbd7 or 7. .. exd4.

--Jon

[Henrik Dinesen]

SF does not like to go into the KID on its own, despite being able to play it very well from the Black side, (beyond move 4, at least). It is encumbered by a hole in its eval, not liking to fianchetto the KB for some reason. The SF developers need to fix this flaw in its eval.

CL

That's exactly why I fooled around with KID, not only Stockfish, many engines won't go into KID. Since at least 15 years ago, engines dislike King's Indian Defense and they like to grab space advantage. I was surprised that although Stockfish gives bad eval to almost all KID positions, in self-play they perform not that bad. Probably in self-play long term plans might unveil more pronouncedly than in 30-40 ply analysis from initial KID position.

I'm not a programmer, but I suspect that in many cases king-safety issues kicks in before the strenght of fianchettoed bishops/long diagonals. It's been long since I last looked in it, but I recall how only one version of HIARCS came up with such "ideas". Other engines, when presented with with position could in some cases akknowledge the power of the bishop-diagonal, but most often they didn't see the "truth" before the knight infront of it was gone. Maybe Catalan is a bit easier for the engines since there's many varitations where the knight moves behind the pawns. Same can maybe be said a about Dutch variations where White plays the knight to f4 via h3.

Anyway, it's all just guesswork from my side, and my impressions is old.

[Dann Corbit]

Greasy Kid Stuff:

http://rybkaforum.net/cgi-bin/rybkaforu ... pid=572459

[lkaufman]

Stockfish wouldn't go by itself into this KID opening:

1. d4 Nf6
2. c4 g6
3. Nc3 Bg7
4. e4 d6
5. f3

[d]rnbqk2r/ppp1ppbp/3p1np1/8/2PPP3/2N2P2/PP4PP/R1BQKBNR b KQkq - 0 1[/d]

The evaluation of Stockfish 8 is below par as Black:

Analysis by Stockfish 8 64 BMI2:

5...0-0 6.Be3 a6 7.h4 e5 8.dxe5 dxe5 9.Nge2 Be6 10.Qxd8 Rxd8 11.Nd5 Bxd5 12.0-0-0 c6 13.Nc3 Nbd7 14.cxd5 cxd5 15.Nxd5 Nxd5 16.Rxd5 Nf8 17.Kb1 b5 18.Rxd8 Rxd8 19.a4 bxa4 20.Bxa6 Ne6 21.Bb5 Bf8 22.Bxa4 Nf4 23.g3 Rd3 24.Bxf4 exf4
+/= (0.35) Depth: 36/50 00:17:54 5487MN
(Kai, 12.03.2017)

From this opening I built a 5-mover opening book (EPD file) by a method which I will describe later. This "Sämisch" EPD file of 300+ openings can be found here:
http://s000.tinyupload.com/?file_id=612 ... 1070438816
From these openings, I self-played Stockfish dev. for 1000 games at 10''+0.1''. The results from White POV are:

Code: Select all

Games         1-0   =-=   0-1   White perf.
 1000         235   574   191     52.2%

The PGN can be found here:
http://s000.tinyupload.com/?file_id=844 ... 8794331058
White performance for KID (Sämisch Variation) is 52.2%. It already strikes as not that high compared to 0.35 eval of Stockfish.

Then, to see if I am not doing something wrong with the opening book building, I took 8moves_GM.pgn opening suite of GM games to move 8, and built the same kind of 5-mover book from those 8-move positions. This "book" represents some sort of compilation of typical GM opening positions. Then, the result in Stockfish dev. self-play 1000 games from this representative sample from White POV was:

Code: Select all

Games         1-0   =-=   0-1   White perf.
 1000         309   518   173     56.8%

White performance is here 56.8%, identical to what I get directly form 8moves_GM.pgn book. The book building doesn't distort the result and it is much higher for White here than for that KID.

So, KID (Sämisch Variation) seems to be better for Black than an average GM opening according to Stockfish fast self-play, although Stockfish doesn't like it and would not go into it unforced (by its eval).

I think that your statistics (but not Stockfish evals) correlate pretty well with GM opinion of the lines you tested. If White was required to play the Saemisch against KI, then Ki would be fairly popular. The critical line is 5...0-0 6.Be3 c5, a totally sound gambit where White struggles to prove an edge by declining. The main reason we see the KI Saemisch now is that White often plays 3.f3 to avoid the Gruenfeld, at the price of being forced to play Saemisch vs. KI. I wrote an entire book for New in Chess about this 3.f3 move. Short conclusion: the normal KI favors White, the normal Gruenfeld gives Black virtual equality, while 3.f3 gives White just a small edge against both Gruenfeld and KI.
Komodo 10.4 will have a new feature called "variety" which will allow you to get different games from a given position without having to count on MP or timing variations. Perhaps you will find this useful for similar experiments.