Seer 2.0.0

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

connor_mcmonigle
Posts: 544
Joined: Sun Sep 06, 2020 4:40 am
Full name: Connor McMonigle

Re: Seer 2.0.0

Post by connor_mcmonigle »

Rebel wrote: Sat Dec 04, 2021 12:15 am Seer 2.4.0 for the GRL.

Games : 1400

Elo pool : 3348

http://rebel13.nl/a/grl.htm
Thanks for testing Ed! I do feel Seer still performs a bit worse at cyclical TCs as compared to other engines so I'll be curious to see your results.
User avatar
Rebel
Posts: 7312
Joined: Thu Aug 18, 2011 12:04 pm
Full name: Ed Schröder

Re: Seer 2.0.0

Post by Rebel »

So far it looks Seer is heading for a top-10 place :wink:
90% of coding is debugging, the other 10% is writing bugs.
User avatar
CMCanavessi
Posts: 1142
Joined: Thu Dec 28, 2017 4:06 pm
Location: Argentina

Re: Seer 2.0.0

Post by CMCanavessi »

Awesome release! 100 elo + ponder is great. FRC next? :mrgreen: :roll:
Follow my tournament and some Leela gauntlets live at http://twitch.tv/ccls
User avatar
Rebel
Posts: 7312
Joined: Thu Aug 18, 2011 12:04 pm
Full name: Ed Schröder

Re: Seer 2.0.0

Post by Rebel »

Estimated elo gain for Seer 2.4.0 : +90

Code: Select all

Gambit Rating List
Running      : Gauntlet Seer 2.4.0
Time Control : Time control : 40/120
Games        : 1400

Results from file gauntlet-seer.pgn:

No. Name            Win Draw Loss Unf.  Score Games       %
-----------------------------------------------------------
  1 Seer 2.4.0     +478 =653 -269   *0  804.5  1400   57.5%
  2 Stockfish 8     +54  =85  -61   *0   96.5   200   48.2%
  3 Ethereal 12.75  +44  =96  -60   *0   92.0   200   46.0%
  4 Komodo 11       +51  =79  -70   *0   90.5   200   45.2%
  5 Igel 3.0.5      +36 =104  -60   *0   88.0   200   44.0%
  6 Pedone 3.1      +32 =107  -61   *0   85.5   200   42.8%
  7 Berserk 5       +37  =84  -79   *0   79.0   200   39.5%
  8 Nemorino 6.00   +15  =98  -87   *0   64.0   200   32.0%

Total Games:    1400
White Wins:      362 (25.9%)
Black Wins:      385 (27.5%)
Draws:           653 (46.6%)
Unfinished:        0 (0.0%)

Estimated elo gain for Seer_2.4.0
Elo pool : 3348
Seer 2.3.0 : 3304.0
Seer_2.4.0 : 3394.8
Difference : 90.8
Provisional Rating List - TOP 10

Code: Select all

     PLAYER                 :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)     W     D     L  D(%)
   1 Stockfish 14.1         :  3709.5   10.7  2368.0    3000    79      59  1768  1200    32    40
   2 Komodo-Dragon 2.5      :  3646.9   18.4  2332.5    3400    69     100  1492  1681   227    49
   3 Lc0 v28                :  3581.3   14.5   591.5    1000    59     100   335   513   152    51
   4 Koivisto 7.9           :  3501.3   15.9  1156.0    2200    53      75   579  1154   467    52
   5 SlowChess 2.8          :  3485.0   10.5  1246.0    2400    52      59   599  1294   507    54
   6 Ethereal 13.25-NNUE    :  3454.0    6.6  2717.0    5600    49      58  1305  2824  1471    50
   7 Berserk 7              :  3452.8   13.7  1139.0    2400    47      73   461  1356   583    57
   8 RubiChess 2.2          :  3407.1   10.7  2472.0    5800    43      78  1057  2830  1913    49
   9 Seer 2.4.0             :  3400.6   17.2   804.5    1400    57      53   478   653   269    47
  10 Igel 3.0.5             :  3355.1    4.5  2289.5    6100    38      55   846  2887  2367    47
Congrats Connor with place 9 :!:
90% of coding is debugging, the other 10% is writing bugs.
User avatar
Gabor Szots
Posts: 1442
Joined: Sat Jul 21, 2018 7:43 am
Location: Budapest, Hungary
Full name: Gabor Szots

Re: Seer 2.0.0

Post by Gabor Szots »

connor_mcmonigle wrote: Fri Dec 03, 2021 10:44 pm
Gabor Szots wrote: Fri Dec 03, 2021 10:36 pm
connor_mcmonigle wrote: Fri Dec 03, 2021 8:47 pm I've released a new version of Seer here: https://github.com/connormcmonigle/seer ... tag/v2.4.0. It should be substantially stronger than the previous release and adds support for Syzygy EGTB probing as well as pondering. More information can be found in the release description.
Hi Connor,

In the makefile the net file name is different from the latest net unless I have overlooked something.
Hello Gabor,
The network hash in the makefile/network listed in the README is correct. It's also the most recent network as listed on the "seer-training" release page, though it's not listed as "latest" as I'm using the network hash as the release name which results in GitHub assuming an incorrect ordering. In general, the release page for the "seer-training" repo should be ignored.
OK, I don't know what wget is but explicitly giving the nnue address from the readme I could download the appropriate net and compilation was successful.
Gabor Szots
CCRL testing group
connor_mcmonigle
Posts: 544
Joined: Sun Sep 06, 2020 4:40 am
Full name: Connor McMonigle

Re: Seer 2.0.0

Post by connor_mcmonigle »

Rebel wrote: Sat Dec 04, 2021 8:18 am Estimated elo gain for Seer 2.4.0 : +90

Code: Select all

Gambit Rating List
Running      : Gauntlet Seer 2.4.0
Time Control : Time control : 40/120
Games        : 1400

Results from file gauntlet-seer.pgn:

No. Name            Win Draw Loss Unf.  Score Games       %
-----------------------------------------------------------
  1 Seer 2.4.0     +478 =653 -269   *0  804.5  1400   57.5%
  2 Stockfish 8     +54  =85  -61   *0   96.5   200   48.2%
  3 Ethereal 12.75  +44  =96  -60   *0   92.0   200   46.0%
  4 Komodo 11       +51  =79  -70   *0   90.5   200   45.2%
  5 Igel 3.0.5      +36 =104  -60   *0   88.0   200   44.0%
  6 Pedone 3.1      +32 =107  -61   *0   85.5   200   42.8%
  7 Berserk 5       +37  =84  -79   *0   79.0   200   39.5%
  8 Nemorino 6.00   +15  =98  -87   *0   64.0   200   32.0%

Total Games:    1400
White Wins:      362 (25.9%)
Black Wins:      385 (27.5%)
Draws:           653 (46.6%)
Unfinished:        0 (0.0%)

Estimated elo gain for Seer_2.4.0
Elo pool : 3348
Seer 2.3.0 : 3304.0
Seer_2.4.0 : 3394.8
Difference : 90.8
Provisional Rating List - TOP 10

Code: Select all

     PLAYER                 :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)     W     D     L  D(%)
   1 Stockfish 14.1         :  3709.5   10.7  2368.0    3000    79      59  1768  1200    32    40
   2 Komodo-Dragon 2.5      :  3646.9   18.4  2332.5    3400    69     100  1492  1681   227    49
   3 Lc0 v28                :  3581.3   14.5   591.5    1000    59     100   335   513   152    51
   4 Koivisto 7.9           :  3501.3   15.9  1156.0    2200    53      75   579  1154   467    52
   5 SlowChess 2.8          :  3485.0   10.5  1246.0    2400    52      59   599  1294   507    54
   6 Ethereal 13.25-NNUE    :  3454.0    6.6  2717.0    5600    49      58  1305  2824  1471    50
   7 Berserk 7              :  3452.8   13.7  1139.0    2400    47      73   461  1356   583    57
   8 RubiChess 2.2          :  3407.1   10.7  2472.0    5800    43      78  1057  2830  1913    49
   9 Seer 2.4.0             :  3400.6   17.2   804.5    1400    57      53   478   653   269    47
  10 Igel 3.0.5             :  3355.1    4.5  2289.5    6100    38      55   846  2887  2367    47
Congrats Connor with place 9 :!:
Thanks for testing Ed! This is a fantastic result :)
connor_mcmonigle
Posts: 544
Joined: Sun Sep 06, 2020 4:40 am
Full name: Connor McMonigle

Re: Seer 2.0.0

Post by connor_mcmonigle »

Gabor Szots wrote: Sat Dec 04, 2021 8:22 am ...

OK, I don't know what wget is but explicitly giving the nnue address from the readme I could download the appropriate net and compilation was successful.
Great, I'm glad you were able to figure it out. I should probably update the README to improve the clarity. Needless to say, please be sure to compare NPS with the release binaries as Seer's performance is quite sensitive to the compiler/associated configuration.
connor_mcmonigle
Posts: 544
Joined: Sun Sep 06, 2020 4:40 am
Full name: Connor McMonigle

Re: Seer 2.0.0

Post by connor_mcmonigle »

CMCanavessi wrote: Sat Dec 04, 2021 3:04 am Awesome release! 100 elo + ponder is great. FRC next? :mrgreen: :roll:
Thanks! The next release will likely feature FRC support, though no guarantees...
User avatar
Gabor Szots
Posts: 1442
Joined: Sat Jul 21, 2018 7:43 am
Location: Budapest, Hungary
Full name: Gabor Szots

Re: Seer 2.0.0

Post by Gabor Szots »

connor_mcmonigle wrote: Sat Dec 04, 2021 9:14 amNeedless to say, please be sure to compare NPS with the release binaries as Seer's performance is quite sensitive to the compiler/associated configuration.
I have already done so. My compile is only slightly faster than the one you provided.
Gabor Szots
CCRL testing group
connor_mcmonigle
Posts: 544
Joined: Sun Sep 06, 2020 4:40 am
Full name: Connor McMonigle

Re: Seer 2.0.0

Post by connor_mcmonigle »

Gabor Szots wrote: Sat Dec 04, 2021 9:41 am
connor_mcmonigle wrote: Sat Dec 04, 2021 9:14 amNeedless to say, please be sure to compare NPS with the release binaries as Seer's performance is quite sensitive to the compiler/associated configuration.
I have already done so. My compile is only slightly faster than the one you provided.
Awesome, thanks!