zurichess geneva released

[brtzsnr]

Hello, everyone!

I'm happy to release a new version of my engine, zurichess zurichess.


The source code for this release can be browsed from https://bitbucket.org/zurichess/zuriche ... ase.geneva
The release binaries can be downloaded at: https://bitbucket.org/zurichess/zurichess/downloads
Linux AMD64 https://bitbucket.org/zurichess/zuriche ... inux-amd64
Windows AMD64 https://bitbucket.org/zurichess/zuriche ... -amd64.exe


Since last release
* Implement fifty-move draw rule. Add HasLegalMoves and InsufficientMaterial methods.
* Improve move ordering: add killer phase; remove sorting.
* Improve time control: add more time when the move is predicted.
* Add basic futility pruning.
* Switch tuning to using TensorFlow framework. txt is now deprecated.
* Evaluate rooks on open and half-open files.
* Improve mobility calculation.
* Tweak null-move conditions: allow double null-moves.
* Usual code clean ups, speed ups and bug fixes.

geneva is about 100 ELOs stronger than fribourg (which is 2441 on CCRL 40/40), depending on the time control.

Code: Select all

40/60+1
Rank Name                          ELO   Games   Score   Draws
   1 geneva                         56    1380     58%     14%
   2 stockfish5-13                 -17     460     48%     13%
   3 fribourg                      -49     459     43%     23%
   4 stockfish5-12                -103     461     36%      7%

Code: Select all

40/120+2
Rank Name                          ELO   Games   Score   Draws
   1 donna-3.1                     156     231     71%     21%
   2 stockfish5-14                 -17     229     48%     11%
   3 geneva                        -66     460     41%     16%

This is the same version that played in HGM's tourney last week. As usual the source code can be found at https://bitbucket.org/zurichess/zurichess/overview . The latest development Linux AMD64 binary can downloaded from drone.io
https://drone.io/bitbucket.org/zuriches ... /zurichess

Please test the new version if you maintain a rating list.

Regards,
Alexandru

[supersharp77]

Thank You Alexandru...........nice engine......keep up the good work!!
AR

[Graham Banks]

Thanks Alexandru.

[MikeB]

Hello, everyone!

I'm happy to release a new version of my engine, zurichess zurichess.


The source code for this release can be browsed from https://bitbucket.org/zurichess/zuriche ... ase.geneva
The release binaries can be downloaded at: https://bitbucket.org/zurichess/zurichess/downloads
Linux AMD64 https://bitbucket.org/zurichess/zuriche ... inux-amd64
Windows AMD64 https://bitbucket.org/zurichess/zuriche ... -amd64.exe


Since last release
* Implement fifty-move draw rule. Add HasLegalMoves and InsufficientMaterial methods.
* Improve move ordering: add killer phase; remove sorting.
* Improve time control: add more time when the move is predicted.
* Add basic futility pruning.
* Switch tuning to using TensorFlow framework. txt is now deprecated.
* Evaluate rooks on open and half-open files.
* Improve mobility calculation.
* Tweak null-move conditions: allow double null-moves.
* Usual code clean ups, speed ups and bug fixes.

geneva is about 100 ELOs stronger than fribourg (which is 2441 on CCRL 40/40), depending on the time control.

Code: Select all

40/60+1
Rank Name                          ELO   Games   Score   Draws
   1 geneva                         56    1380     58%     14%
   2 stockfish5-13                 -17     460     48%     13%
   3 fribourg                      -49     459     43%     23%
   4 stockfish5-12                -103     461     36%      7%

Code: Select all

40/120+2
Rank Name                          ELO   Games   Score   Draws
   1 donna-3.1                     156     231     71%     21%
   2 stockfish5-14                 -17     229     48%     11%
   3 geneva                        -66     460     41%     16%

This is the same version that played in HGM's tourney last week. As usual the source code can be found at https://bitbucket.org/zurichess/zurichess/overview . The latest development Linux AMD64 binary can downloaded from drone.io
https://drone.io/bitbucket.org/zuriches ... /zurichess

Please test the new version if you maintain a rating list.

Regards,
Alexandru

Thank you Alexandru!

For those who have a Mac.

https://www.dropbox.com/s/jykgqd4bpbjsb ... x.zip?dl=1


go depth 15 (suedo bench)

Code: Select all

Mac-Pro:gows michaelbyrne$ cd /users/michaelbyrne/documents/sources/gows/bin
Mac-Pro:bin michaelbyrne$ ./zurichess
zurichess geneva https://bitbucket.org/zurichess/zurichess
build with go1.5 at 2015-11-29 16:16:21, running on amd64
uci   
id name zurichess geneva
id author Alexandru Moșoi

option name UCI_AnalyseMode type check default false
option name Hash type spin default 64 min 1 max 65536
option name Ponder type check default true
uciok
go depth 15
info depth 0 seldepth 0 score cp 0 nodes 2 time 0 nps 18907 pv
info depth 1 seldepth 1 score cp 47 nodes 43 time 0 nps 140891 pv e2e3
info depth 2 seldepth 2 score cp 0 nodes 158 time 0 nps 239273 pv e2e3 e7e6
info depth 3 seldepth 3 score cp 36 nodes 560 time 1 nps 347992 pv e2e3 e7e6 b1c3
info depth 4 seldepth 4 score cp 4 nodes 4082 time 6 nps 583250 pv e2e4 d7d5 b1c3 e7e6 e4d5 e6d5
info depth 5 seldepth 6 score cp 11 nodes 6776 time 11 nps 594871 pv e2e4 d7d5 b1c3 d5e4 c3e4
info depth 6 seldepth 6 score cp 6 nodes 18404 time 27 nps 670164 pv e2e4 d7d5 e4d5 d8d5 b1c3 d5d6
info depth 7 seldepth 8 score cp 22 nodes 68336 time 76 nps 891583 pv e2e3 e7e6 b1c3 d7d5 g1f3 c8d7 a2a3
info depth 8 seldepth 8 score cp 0 nodes 203358 time 172 nps 1177398 pv e2e3 e7e6 b1c3 d7d5 a2a3 c8d7 g1f3 c7c5
info depth 9 seldepth 10 score cp 24 nodes 547129 time 396 nps 1379632 pv e2e3 b8c6 d2d4 g8f6 b1c3 e7e6 g1f3 a7a6 a2a3
info depth 10 seldepth 11 score cp 2 nodes 1520490 time 1059 nps 1434518 pv g1f3 g8f6 b1c3 d7d5 e2e3 b8c6 h2h3 e7e6 a2a3 a7a6
info depth 11 seldepth 12 score cp 17 nodes 2776784 time 1945 nps 1427554 pv g1f3 c7c5 e2e3 b8c6 f1e2 d7d5 e1g1 e7e5 d2d4 c5d4 e3d4
info depth 12 seldepth 14 score cp 3 nodes 7106446 time 4854 nps 1463813 pv d2d4 g8f6 b1c3 d7d5 c1f4 a7a6 e2e3 b8c6 g1f3 e7e6 f3e5 c8d7 e5d7 d8d7
info depth 13 seldepth 14 score cp 13 nodes 13091513 time 8820 nps 1484201 pv d2d4 g8f6 b1c3 d7d5 e2e3 c7c6 g1f3 e7e6 a2a3 f8d6 f1e2 e8g8 e1g1
info depth 14 seldepth 16 score cp 13 nodes 42954231 time 28234 nps 1521348 pv e2e4 c7c5 g1f3 e7e6 b1c3 b8c6 f1e2 a7a6 e1g1 b7b5 a2a3 f8d6 d2d4 c5d4 f3d4
info depth 15 seldepth 16 score cp 21 nodes 59355739 time 38935 nps 1524450 pv e2e4 c7c5 g1f3 e7e6 b1c3 b8c6 f1b5 g8f6 d2d3 h7h6 e1g1 f8d6 b5c6 b7c6 c1d2
bestmove e2e4 ponder c7c5

[brtzsnr]

Thanks, Mike!

I can provide a Mac binary in the future, if there is sufficient interest. However, when zurichess joined HGM's monthly contest from a Mac it behaved strangely wrt time management.

[MikeB]

Thanks, Mike!

I can provide a Mac binary in the future, if there is sufficient interest. However, when zurichess joined HGM's monthly contest from a Mac it behaved strangely wrt time management.

Let me talk to the programmer about that - oh that's you!

Actually time for Mac can be different, depending on time how is gathered in the first place. Had to make patch before for somebody else's engine - have not tested time on your engine , but let me take a look. Was it playing too fast or too slow?

[brtzsnr]

Actually time for Mac can be different, depending on time how is gathered in the first place. Had to make patch before for somebody else's engine - have not tested time on your engine , but let me take a look. Was it playing too fast or too slow?

Too slow. Some moves especially in the beginning would take a lot of time, >50% of the total time. I haven't ruled out the GUI or the network (at that time I was on a hotel's wifi).

[MikeB]

Too slow. Some moves especially in the beginning would take a lot of time, >50% of the total time. I haven't ruled out the GUI or the network (at that time I was on a hotel's wifi).

You might want revisit time - I noticed "go movetime 5000" ( means a 5 second search in uci speak) was not working properly. I printed out the variable and 5000 ms is mutilplied by 1 million - that doesn't seem right. But the whole run only took 1 second

build with go1.5 at 2015-11-29 18:47:51, running on amd64
go movetime 5000
time 5000000000
time 5000000000
info depth 0 seldepth 0 score cp 0 nodes 2 time 0 nps 23614 pv
info depth 1 seldepth 1 score cp 47 nodes 43 time 0 nps 204367 pv e2e3
info depth 2 seldepth 2 score cp 0 nodes 158 time 0 nps 360964 pv e2e3 e7e6
info depth 3 seldepth 3 score cp 36 nodes 560 time 1 nps 406843 pv e2e3 e7e6 b1c3
info depth 4 seldepth 4 score cp 4 nodes 4082 time 6 nps 585910 pv e2e4 d7d5 b1c3 e7e6 e4d5 e6d5
info depth 5 seldepth 6 score cp 11 nodes 6776 time 10 nps 620291 pv e2e4 d7d5 b1c3 d5e4 c3e4
info depth 6 seldepth 6 score cp 6 nodes 18404 time 26 nps 701331 pv e2e4 d7d5 e4d5 d8d5 b1c3 d5d6
info depth 7 seldepth 8 score cp 22 nodes 68336 time 72 nps 944116 pv e2e3 e7e6 b1c3 d7d5 g1f3 c8d7 a2a3
info depth 8 seldepth 8 score cp 0 nodes 203358 time 161 nps 1256076 pv e2e3 e7e6 b1c3 d7d5 a2a3 c8d7 g1f3 c7c5
info depth 9 seldepth 10 score cp 24 nodes 547129 time 364 nps 1501051 pv e2e3 b8c6 d2d4 g8f6 b1c3 e7e6 g1f3 a7a6 a2a3
info depth 10 seldepth 11 score cp 2 nodes 1520490 time 929 nps 1636213 pv g1f3 g8f6 b1c3 d7d5 e2e3 b8c6 h2h3 e7e6 a2a3 a7a6
bestmove g1f3 ponder g8f6

[brtzsnr]

That's how time works in golang. The underlying type, time.Duration, represents number of nanoseconds. See documentation https://golang.org/pkg/time/#Duration

See also this example

http://play.golang.org/p/9bWBQgz-kz

which outputs

5s 5000000000


The fact that `go movetime 5000` takes only a sec is a consequence of poor branching factor estimation.

[Sylwy]

Thank you, Alexandru !