Stockfish 8 SSE (Special Edition)

[peter]

Thank you, Michael

[fantasmadel50]

Many thanks Michael b for providing this engine, and again thanks to T .Poppins for its compilation for windows. It would have some pruguntas 1) The database of almost 62 thousand games, already incorporated in the engine stock 8 see?, If so, I imagine that the consultation on the part of this is auntomatica.? Of the books out there, I obviously choose only one, the cerebellum, or the polyglot.bin. And the two are loaded into the book path.? Thank you very much for your patience and answers, sorry for my ignorance. Muchas gracias Michael b por proveer este motor, y nuevamente gracias a T .Poppins por su compilación para windows. Tendría algunas preguntas 1)La base de datos de casi 62 mil partidas , ya esta incorporada en el motor stock 8 see ?, si es asi , me imagino que la consulta por parte de este es automática.? De los libros que hay , obviamente elijo solo uno, el cerebellum , o el polyglot.bin .? Y los dos se cargan en el book path .? Muchas gracias por su paciencia y respuestas, disculpen mi ignorancia.

[fantasmadel50]

Good to all, Someone would be so kind to explain clearly, as is done to play with the 2 books in the engine stock8 sse, ie the cerebellum plus the polyglot .bin. ? Thank you very much. I use fritz15 gui, arena or aquarium 2016. Buenas a todos , Alguien seria tan amable de explicar claramente, como se hace para jugar con los 2 libros en el motor stock8 sse, o sea el cerebellum mas el polyglot .bin. ? Muchas gracias. Utilizo fritz15 gui, arena o aquarium 2016.

[fern]

Not bad spanish at all ..
and thanks...I already have it in my venerable 32 bits....

Fern

[MikeB]

Many thanks Michael b for providing this engine, and again thanks to T .Poppins for its compilation for windows. It would have some pruguntas 1) The database of almost 62 thousand games, already incorporated in the engine stock 8 see?, If so, I imagine that the consultation on the part of this is auntomatica.? Of the books out there, I obviously choose only one, the cerebellum, or the polyglot.bin. And the two are loaded into the book path.? Thank you very much for your patience and answers, sorry for my ignorance. Muchas gracias Michael b por proveer este motor, y nuevamente gracias a T .Poppins por su compilación para windows. Tendría algunas preguntas 1)La base de datos de casi 62 mil partidas , ya esta incorporada en el motor stock 8 see ?, si es asi , me imagino que la consulta por parte de este es automática.? De los libros que hay , obviamente elijo solo uno, el cerebellum , o el polyglot.bin .? Y los dos se cargan en el book path .? Muchas gracias por su paciencia y respuestas, disculpen mi ignorancia.

You can use mutilple books by using a colon between books under xboard ( not sure about other GUI's):

Code: Select all

-usePolyglotBook true
-polyglotBook "/Users/michaelbyrne/cluster.mfb/e4.bin:/Users/michaelbyrne/cluster.mfb/d4.bin"

[MikeB]

Stockfish 8 SE is the Special Edition of Stockfish as described here:
http://talkchess.com/forum/viewtopic.php?t=61952

So one might be wondering what is the SSE version? Back in the 80's, Pontiac came out with a Bonneville SE Edition model , but then when they came with a newer "top of the line" model Bonneville , it was the "SSE".

SSE incorporates Brainfish - more information here: http://www.zipproth.de ) , essentially, a deeply searched database of chess positions at its disposal. I created an opening based book on WCC games (available here http://talkchess.com/forum/viewtopic.php?t=61999), knowing this would give Stockfish a huge advantage against an opponent that did not have access to the same database - and we can see the results here:

Code: Select all

500 rounds and 500 games completed...
time control: 1500+50
Date: 11/06/16 : 22:29:26
500 game(s) loaded
Rank Name                  Rating   Δ     +    -     #     Σ    Σ%     W    L    D   W%    =%   OppR 
---------------------------------------------------------------------------------------------------------
   1 Stockfish SSE 110616   3160   0.0    9    9   500  337.5  67.5  212   37  251  42.4  50.2  3040 
   2 Komodo 10.2 64-bit     3040 120.8    9    9   500  162.5  32.5   37  212  251   7.4  50.2  3160 
---------------------------------------------------------------------------------------------------------

Not fair to Komodo, and that should be recognized, but it does illustrate what Stockfish what can do when it has has access to a positional database and every line it plays is in the database.

Source and Mac exe can be for here: https://github.com/MichaelB7/Stockfish

You will need to download Cerebellum_light.bin from here: http://www.zipproth.de and please be sure to enter the path AND the file name in the UCI option BookPath.

Just added a book specifically tailored for SSE and to be used in conjunction with Stockfish 8 SSe and Cerebellum Lite ( Brainfish Edition).

https://github.com/MichaelB7/Stockfish/ ... ok.bin.zip

Source: Chess database consisting over 2.5 million games, a 2.3 GB file, mostly from the last 25 years
source: http://theweekinchess.com as well the historical battles of yesteryear
source: http://www.pgnmentor.com/files.html.

The objective was to make a book with the most popular winning lines through the years. First, using pgn-extract, I extracted just the winning lines and separated them between white and black - about 750,000 games with white wins and around 600 games with black wins. Using the polyglot with the white database, made a book with just white wins where the minimum number of games were 5000. Then, alternating back forth, black, , white, performed 3 more runs , each time decreasing the number of minimum number of games required from 5000 to 1000, to 100 and then to 10 for the last pass. Then merge the two white books together, ensuring the first white book created takes priority ( listing it first) , the same with th e black books , finally merging the white and black books together. See the command line scripts below:

Polyglot first pass: polyglot make-book -min-game 5000 -only-white -pgn wwall_t4.pgn -bin wbook1.bin

Polyglot second pass: polyglot make-book -min-game 1000 -only-black -pgn bwall_t4.pgn -bin bbook1.bin

Polyglot third pass: polyglot make-book -min-game 100 -only-white -pgn wwall_t4.pgn -bin wbook2.bin

Polyglot fourth pass: polyglot make-book -min-game 10 -only-black -pgn bwall_t4.pgn -bin bbook2.bin

Polyglot fifth pass: polyglot merge-book -in1 wbook1.bin -in2 wbook2.bin -out wbook.bin

Polyglot fifth pass: polyglot merge-book -in1 wbook1.bin -in2 wbook2.bin -out wbook.bin

Polyglot sixth pass: polyglot merge-book -in1 bbook1.bin -in2 bbook2.bin -out bbook.bin

Polyglot seventh pass: polyglot merge-book -in1 wbook.bin -in2 bbook.bin -out book.bin

Results with relatively two equal engines , Stockfish 8 SSe below was only using the Cerebellum book. However, the intent is that you use this book with Stockfish 8 SSe in CONJUNCTION WITH the Cerebellum book!

Code: Select all

Rank Name                  Rating   Δ     +    -     #     Σ    Σ%     W    L    D   W%    =%   OppR 
---------------------------------------------------------------------------------------------------------
   1 Stockfish SE 110516    3118   0.0   22   22   500  286.5  57.3   98   25  377  19.6  75.4  3082 
   2 Stockfish SSE 110616   3082  36.9   22   22   500  213.5  42.7   25   98  377   5.0  75.4  3118 
---------------------------------------------------------------------------------------------------------

The book is not that big and that's by design - was just trying too guide Stockfish 8 SSe a little bit and let Cerebellum and the engine takeover relatively early.

From the same database, I just made a big book for those that might be interested :

1st pass for white - min 20 wins
1st pass for black - min 10 wins
2nd pass for white - min 5 wins
2nd pass for black - min 3 wins

https://github.com/MichaelB7/Stockfish/ ... ok.bin.zip

bigbook size unzipped: 5,256,624 bytes

[MikeB]

Another update:

source: https://github.com/MichaelB7/Stockfish
macOS exe: https://github.com/MichaelB7/Stockfish/ ... _macOS.zip

One pull request (https://github.com/official-stockfish/S ... h/pull/864) that offered up more accurate "go node xxx", in conjunction with this accuracy, I added a uci parmeter "Ladder" that starts off roughly with a 64 nodes the setting of "1" (mimic go node 64)and it doubles thereafter ( 2 equals "go node 128", etc) up to level 20 which mimics a search of " go node ~33,000,000" - so starting with a "Ladder 1" setting you can work your way up the ladder ;>).

Also added another feature , a UCI option to hide fail-highs/lows. Credit goes to Dann Corbit.

also included the following commits, not necessarily a comprehensive listing:
https://github.com/official-stockfish/S ... 7f8af09edc
https://github.com/official-stockfish/S ... 1389b44e06
https://github.com/official-stockfish/S ... b33c28ff47
https://github.com/official-stockfish/S ... 82096d371d

Example of game vs human with Ladder setting of "2" - my guess around 1400-1500 maybe weaker - not sure, Stockfish was playing with its full knowledge - just limited to a search of about 128 nodes. Was using the bigbook.bin that is also available at the github webpage.
https://github.com/MichaelB7/Stockfish/ ... ok.bin.zip

[pgn][Event "Computer Chess Game"]
[Site "Mac-Pro.local"]
[Date "2016.11.11"]
[Round "-"]
[White "michaelbyrne"]
[Black "Stockfish SSE 111116"]
[Result "1-0"]
[TimeControl "300+5"]
[Annotator "9... +0.96"]

1. d4 Nf6 2. c4 e6 3. Nc3 Bb4 4. Bd2 O-O 5. e3 c5 6. Nf3 d5 7. Bd3 cxd4
{+0.15/2 0.1} 8. exd4 Nc6 {+0.29/2} 9. O-O dxc4 {+0.96/3 0.1} 10. Bxc4 Nxd4
{+0.64/3 0.1} 11. Nxd4 Qxd4 {+1.16/4 0.1} 12. Bb3 Rd8 {+1.36/3} 13. Be3 Qd6
{+0.72/2 0.1} 14. Qf3 Qd3 {+1.20/2 0.1} 15. Rfd1 Bxc3 {-4.31/2 0.1} 16.
Rxd3 Rxd3 {-4.31/3 0.1} 17. bxc3 Rxc3 {-4.24/3 0.1} 18. Rc1 Rxc1+
{-4.03/3 0.1} 19. Bxc1 Kf8 {-5.53/3} 20. Bb2 Nd5 {-4.87/3 0.1} 21. Bxd5
exd5 {-4.68/4 0.1} 22. Qxd5 Be6 {-4.42/3 0.1} 23. Qd6+ Kg8 {-3.25/5 0.1}
24. h3 Bxa2 {-2.57/4 0.1} 25. Qe5 f6 {+4.49/4 0.1} 26. Qc7 Bd5
{-2.20/4 0.1} 27. f3 Re8 {-2.23/3 0.1} 28. Kf2 h6 {-2.41/3 0.1} 29. Qg3 Kh7
{-2.18/3 0.1} 30. Qc7 Kg8 {+0.00/4 0.1} 31. Qc5 Bf7 {-2.92/2 0.1} 32. Qxa7
b5 {-2.87/3 0.1} 33. Qd7 b4 {-2.75/3 0.1} 34. Qb5 b3 {-2.79/3 0.1} 35. g4
Rd8 {-2.55/2 0.1} 36. Qe2 Re8 {+0.45/4 0.1} 37. Qd2 Bc4 {-2.55/3 0.1} 38.
Kg3 Re2 {-2.35/3 0.1} 39. Qc3 Bd5 {-2.36/2 0.1} 40. h4 Rc2 {+0.03/3} 41.
Qd4 Bc6 {-2.63/2 0.1} 42. g5 hxg5 {-2.27/3} 43. hxg5 Kf7 {-3.17/2 0.1} 44.
gxf6 g6 {-3.83/2} 45. Ba3 g5 {-3.41/3 0.1} 46. Qa7+ Kxf6 {-3.71/3 0.1} 47.
Qe7+ Kg6 {-3.71/4 0.1} 48. Qe6+ Kg7 {-7.96/4 0.1} 49. Qe5+ Kg6
{-3.65/3 0.1} 50. Bb2 Kh5 {-2.96/2 0.1} 51. f4 Rg2+ {-3.93/2} 52. Kh3 Rg1
{-3.04/2 0.1} 53. Bd4 Bd7+ {-3.27/2 0.1} 54. Kh2 Rg4 {-3.05/3 0.1} 55. Qh8+
Kg6 {-6.13/4 0.1} 56. Qg7+ Kf5 {-8.03/4 0.1} 57. Qxd7+ Kxf4 {-8.11/3 0.1}
58. Qf7+ Ke4 {-8.11/3} 59. Qe6+ Kf4 {-9.92/3 0.1} 60. Be5+ Ke3
{-128.50/3 0.1} 61. Qxg4 Kd2 {-128.51/2 0.1} 62. Qxg5+ Kc2 {-128.51/3 0.1}
63. Qg2+ Kd3 {-128.51/3 0.1} 64. Qb2 Ke4 {-128.51/4 0.1} 65. Kh3 Kf5
{-128.51/3 0.1} 66. Kg3 Ke4 {-128.51/4 0.1} 67. Kg4 Kd5 {-128.51/3 0.1} 68.
Kf4 Kc4 {-128.51/3 0.1} 69. Ke4 Kb4 {-128.51/3 0.1} 70. Kd4 Ka4
{-128.51/4 0.1} 71. Kc4 Ka5 {-128.51/4 0.1} 72. Qxb3 Ka6 {-128.51/3 0.1}
73. Kc5 Ka7 {-128.51/3} 74. Kc6 Ka6 {-99.98/27} 75. Qb6#
{White mates} 1-0[/pgn]

and like before, it includes all other changes made previously , use of the Cerebellum book (Brainfish), correct handling of multi-pv when probing tb's , with or without the 50 move rule check, faster play when mate is found , "throttle" option, 40 skill levels, and select strength by ELO to name a few.

Will add windows exe's to GitHub if someone can compile them...

[fantasmadel50]

PLEASE someone who knows, compile for windows. Thank you.

[MikeB]

Another update:

source: https://github.com/MichaelB7/Stockfish
macOS exe: https://github.com/MichaelB7/Stockfish/ ... _macOS.zip

Expanded on the "Ladder" option introduced last version, one can now select the Ladder level by rating. Stockfish at it's lowest level - around a 68 node search plays between 1500 and 1600. Now we add the option "ladder by ratings", one can climb the ladder in 25 ELO increments. Just a small enough increase not to get discouraged as you move up to next level. The magic is in timeman.c , each 25 point step up the ladder , increases the search by 18.92% ( 1.1892^4 ~ 2), so after 4 steps up the ladder, Stockfish is now seeing double the nodes, and you my friend are playing ~ 100 ELO higher than what you were playing 4 steps ago.

To see this on the command line:

setoption name Ladder value true
set option name Ladder_Rating value 1550
info depth 3 seldepth 3 multipv 1 score cp 93 nodes 68
..
setoption name Ladder_Rating value 1575
info depth 3 seldepth 3 multipv 1 score cp 65 nodes 76
...
setoption name Ladder_Rating value 1600
info depth 3 seldepth 3 multipv 1 score cp 65 nodes 90
...
setoption name Ladder_Rating value 1625
info depth 4 seldepth 4 multipv 1 score cp 119 nodes 109
...
setoption name Ladder_Rating value 1650
info depth 3 seldepth 3 multipv 1 score cp 47 nodes 128
...

now skipping the intermediary levels
setoption name Ladder_Rating value 1750
info depth 4 seldepth 4 multipv 1 score cp 8 nodes 257
setoption name Ladder_Rating value 1850
info depth 5 seldepth 5 multipv 1 score cp 27 nodes 512
...
now jumping to level 2050
setoption name Ladder_Rating value 2050
info depth 6 seldepth 6 multipv 1 score cp 35 upperbound nodes 2049

Here was a quick match against Sungorus 1.4 - rated about 2312 on the CCRL 40/40 listing with setoption name Ladder_Rating value 2300 (about 11585 nodes per move) Time control was 30 sec 1sec

Code: Select all

Date: 11/13/16 : 20:50:17
322 game(s) loaded
Rank Name                    Rating   Δ     +    -     #     Σ    Σ%     W    L    D   W%    =%   OppR 
---------------------------------------------------------------------------------------------------------
   1 Sungorus 1.4             3109   0.0   17   17   322  169.5  52.6  135  118   69  41.9  21.4  3091 
   2 Stockfish SSE 111316-y   3091  18.3   17   17   322  152.5  47.4  118  135   69  36.6  21.4  3109 
---------------------------------------------------------------------------------------------------------

and of course it maintains all the added features added previously...

Features Added Previously:

One pull request (https://github.com/official-stockfish/S ... h/pull/864) that offered up more accurate "go node xxx", in conjunction with this accuracy, I added a uci parmeter "Ladder" that starts off roughly with a 64 nodes the setting of "1" (mimic go node 64)and it doubles thereafter ( 2 equals "go node 128", etc) up to level 20 which mimics a search of " go node ~33,000,000" - so starting with a "Ladder 1" setting you can work your way up the ladder ;>).

Also added another feature , a UCI option to hide fail-highs/lows. Credit goes to Dann Corbit.

A bigbook.bin that is available at the github webpage.
https://github.com/MichaelB7/Stockfish/ ... ok.bin.zip

Use of the Cerebellum book (Brainfish), correct handling of multi-pv when probing tb's , with or without the 50 move rule check, faster play when mate is found , "throttle" option, 40 skill levels, and select strength by ELO to name a few.

Note: Use the skill level to play at level less than 1550.

Will add windows exe's to GitHub if someone can compile them...

Note: All the ELOs presented here are not meant to be USCF, FIDE or anything else. The ELOs presented are just lines in the sand in which a user can mark his progress. As one climbs up the ladder, they can be confident they are making real progress in their chess playing ability.

[tpoppins]

Windoze bins for 111316:

TinyUpload
Mediafire

Code: Select all

stockfish-111316-SSE-w32-old_tp.exe .... for old Win32 systems
stockfish-111316-SSE-w32-sse_tp.exe .... for Win32 systems with SSE
stockfish-111316-SSE-x64_tp.exe ........ for old x64 systems
stockfish-111316-SSE-x64-popcnt_tp.exe . for 'modern' x64 systems with POPCNT

The w32 compiles have been built and tested on WinXP+SP3, the rest 
on Win7 x64 Pro. No BMI2 compile at this time.

Bench: 5513149