Scorpio 2.8.7 MCTS+NN windows version

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

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Daniel Shawul
Posts: 3810
Joined: Tue Mar 14, 2006 10:34 am
Location: Ethiopia
Contact:

Re: Scorpio 2.8.7 MCTS+NN windows version

Post by Daniel Shawul » Sun Sep 23, 2018 4:33 pm

Scally wrote:
Sun Sep 23, 2018 3:36 pm
Hi Daniel,

Is it possible to compile a version for the 32 bit CPU armv7l Raspberry Pi? I managed to compile LC0 using openblas and can get up to 100 nps on 3 cores, but I’m having problems compiling your engine.

Thanks,

Al.
Scorpio itself can be compiled for arm 32/64 bit no problem -- I have actually done that before for android.
To use the NN you would need to compile egbbdll using tensorflow or tensorflow-lite using bazel.
I have seen people do that before but I have not tried it myself.

User avatar
Scally
Posts: 99
Joined: Thu Sep 28, 2017 7:34 pm
Location: Bermondsey, London
Full name: Alan Cooper
Contact:

Re: Scorpio 2.8.7 MCTS+NN windows version

Post by Scally » Sun Sep 23, 2018 6:32 pm

Daniel Shawul wrote:
Sun Sep 23, 2018 4:33 pm
Scally wrote:
Sun Sep 23, 2018 3:36 pm
Hi Daniel,

Is it possible to compile a version for the 32 bit CPU armv7l Raspberry Pi? I managed to compile LC0 using openblas and can get up to 100 nps on 3 cores, but I’m having problems compiling your engine.

Thanks,

Al.
Scorpio itself can be compiled for arm 32/64 bit no problem -- I have actually done that before for android.
To use the NN you would need to compile egbbdll using tensorflow or tensorflow-lite using bazel.
I have seen people do that before but I have not tried it myself.
Thanks Daniel.

Al.
Alan Cooper
My Chess Computers

joechess
Posts: 24
Joined: Sun Oct 30, 2016 10:02 pm

Re: Scorpio 2.8.7 MCTS+NN windows version

Post by joechess » Sun Sep 23, 2018 11:03 pm

I have uploaded both now. The default supports sse2 intructions and is the one that should work for you.
Those with modern cpus can use the one in avx/ sub-folder.
Let me know if it works for you now.
working well now, many thanks

BrendanJNorman
Posts: 1661
Joined: Sun Feb 07, 2016 11:43 pm
Full name: Brendan J Norman

Re: Scorpio 2.8.7 MCTS+NN windows version

Post by BrendanJNorman » Mon Sep 24, 2018 6:55 am

I just played a 3 + 2 blitz game against version 2.8.8, using 2x32 nets with the sole change from default parameters being my changing treeht to 5000.

I gave Scorpio a book made from the decisive games of 2700+ players, because in the first game he played 1.d4 d5 2.Qd3 against me, making it impractical for training without a book.

I am using Windows 10 on a Macbook 2015 model via BootCamp, so the hardware is relatively dated.

Would appreciate any advice on strengthening it up a bit.


Joerg Oster
Posts: 693
Joined: Fri Mar 10, 2006 3:29 pm
Location: Germany

Re: Scorpio 2.8.7 MCTS+NN windows version

Post by Joerg Oster » Mon Sep 24, 2018 9:46 am

Daniel,
just for your information, the Linux version crashes immediately (memory access fault) if I don't set "use_nn" to 0 in the ini file.
./scorpio go quit
feature done=0
ht 4194304 X 16 = 64.0 MB
eht 524288 X 8 = 8.0 MB
pht 32768 X 24 = 0.8 MB
treeht 26843136 X 40 = 1024.0 MB
processors [3]
EgbbProbe not Loaded!
loading_time = 0s
[st = 11114ms, mt = 29250ms , hply = 0 , moves_left 10]
Speicherzugriffsfehler
I'm still in the process of putting everything together step by step (no nets, no egbb's atm)
and wanted to check if Scorpio is working correctly on my box.

Output with 3 Threads, MCTS, 1024 MB Hashtree size

Code: Select all

./scorpio go quit
feature done=0
ht 4194304 X 16 = 64.0 MB
eht 524288 X 8 = 8.0 MB
pht 32768 X 24 = 0.8 MB
treeht 26843136 X 40 = 1024.0 MB
processors [3]
EgbbProbe not Loaded!
loading_time = 0s
[st = 11114ms, mt = 29250ms , hply = 0 , moves_left 10]
63 38 111 39028  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4
64 39 223 70554  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 d7-d6 d2-d4 Nc6xd4 Nf3xd4 e5xd4 Qd1xd4 Bc8-d7
65 38 335 93586  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 d7-d6 Bf1-c4 Ng8-f6 Ke1-g1 Bc8-e6 Bc4xe6 f7xe6
66 38 447 109927  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4 Bf8-b4 f2-f3 Ke8-g8 Bc1-e3 d7-d5 Nd4xc6 b7xc6 e4-e5 Bb4xc3 b2xc3
67 38 558 124308  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4 Bf8-b4 f2-f3 Ke8-g8 Bf1-b5 a7-a6 Bb5-a4 Bb4xc3 b2xc3 Nc6xd4 c3xd4
68 38 670 139875  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4 Bf8-b4 f2-f3 Ke8-g8 Bf1-b5 a7-a6 Bb5-a4 Nc6xd4 Qd1xd4 c7-c5 Qd4-d2 b7-b5 Ba4-b3 c5-c4 e4-e5 Rf8-e8 Ke1-g1 Re8xe5 Qd2-d4 Bb4-c5 Qd4xc5 Re5xc5
69 38 781 152683  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4 Bf8-b4 f2-f3 Ke8-g8 Bf1-b5 a7-a6 Bb5-a4 Nc6xd4 Qd1xd4 c7-c5 Qd4-d3 b7-b5 e4-e5 Rf8-e8 Ke1-g1 Re8xe5 Bc1-f4 c5-c4 Qd3-d2 Bb4-c5 Kg1-h1 Re5-e6 b2-b4 Bc5xb4 Ra1-b1 Bb4xc3 Qd2xc3 b5xa4 Qc3xc4
70 38 893 165201  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4 Bf8-b4 f2-f3 Ke8-g8 Bf1-b5 a7-a6 Bb5-a4 Nc6xd4 Qd1xd4 c7-c5 Qd4-d3 b7-b5 e4-e5 Rf8-e8 Ke1-g1 Re8xe5 Ba4-b3 d7-d5 a2-a3 c5-c4 Qd3-d4 Bb4xc3 b2xc3 Re5-e2 Bb3-a2 Bc8-f5 Bc1-g5 Qd8-d6 Bg5xf6 g7xf6 f3-f4 Ra8-e8
71 38 1005 178914  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4 Bf8-b4 f2-f3 Ke8-g8 Bf1-b5 a7-a6 Bb5-a4 Nc6xd4 Qd1xd4 c7-c5 Qd4-d1 d7-d5 e4-e5 d5-d4 Ke1-g1 d4xc3 Qd1xd8 Rf8xd8 e5xf6 g7xf6 a2-a3 Bb4-a5 Ba4-b3 Bc8-e6 Bb3xe6 f7xe6 Ra1-b1 b7-b5 b2-b3 Rd8-d5 Rf1-e1 e6-e5 a3-a4 b5-b4 Bc1-e3 Ra8-d8 Rb1-a1 Kg8-g7 Be3-f2

#  1     40  142659 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4 Bf8-b4 f2-f3 Ke8-g8 Bf1-b5 a7-a6 Bb5-a4 Nc6xd4 Qd1xd4 c7-c5 Qd4-d1 d7-d5 e4-e5 d5-d4 Ke1-g1 d4xc3 Qd1xd8 Rf8xd8 e5xf6 g7xf6 a2-a3 Bb4-a5 Ba4-b3 Bc8-e6 Bb3xe6 f7xe6 Ra1-b1 b7-b5 Bc1-e3 c3xb2 Rb1xb2 c5-c4 Rb2-b1 Ba5-c3 Rf1-d1 f6-f5 Be3-b6 Rd8-d5 Rd1xd5 e6xd5 Rb1-d1 Ra8-b8 Bb6-d4 Bc3xd4 Rd1xd4 Rb8-d8 c2-c3 Kg8-f7 Kg1-f2 Kf7-f6 Rd4-d1 d5-d4 Rd1xd4 Rd8xd4 c3xd4
#  2     33   28312 d2-d4 d7-d5 Ng1-f3 Ng8-f6 e2-e3 e7-e6 Bf1-d3 Bf8-e7 Ke1-g1 Ke8-g8 Nb1-c3 Nb8-c6 b2-b3 b7-b6 Bc1-b2 Bc8-b7 a2-a3 Qd8-d7 h2-h3 a7-a6 Qd1-d2 h7-h6 Rf1-e1 Rf8-e8 e3-e4 d5xe4 Nc3xe4 Nf6xe4 Bd3xe4 Be7-f6 c2-c3 a6-a5 Ra1-d1 a5-a4 b3-b4 Kg8-f8 Nf3-e5 Nc6xe5 Be4xb7 Ne5-c4 Qd2-e2 Nc4xb2 Qe2xb2 Ra8-d8 Qb2-c2 e6-e5 d4xe5 Re8xe5 Bb7-f3 Re5xe1 Rd1xe1 Qd7-d3 Qc2xd3 Rd8xd3 Bf3-c6 Rd3-d8 Bc6xa4 Bf6xc3 Re1-e3 Bc3-f6 Ba4-b3
#  3     20     694 Nb1-c3 d7-d5 d2-d4 Ng8-f6
#  4     33   15387 Ng1-f3 d7-d5 Nb1-c3 Ng8-f6 e2-e3 e7-e6 Bf1-e2 Bf8-d6 d2-d4 Ke8-g8 Ke1-g1 Nb8-c6 a2-a3 a7-a6 Bc1-d2 h7-h6 h2-h3 Bc8-d7 Rf1-e1 Qd8-e7 Ra1-c1 Ra8-d8 Be2-d3 e6-e5 e3-e4 d5xe4 Nc3xe4 Nc6xd4 Ne4xd6 Nd4xf3 Qd1xf3 Qe7xd6 Rc1-d1 Bd7-c6 Qf3-g3 e5-e4 Qg3xd6 Rd8xd6 Bd3-c4 Rd6-d4 Bc4-b3 Rf8-d8 Re1-e2 Bc6-b5 c2-c4 Bb5xc4 Bb3xc4 Rd4xc4
#  5     13     393 e2-e3 d7-d5 d2-d4 e7-e6 Ng1-f3
#  6     15     451 d2-d3 e7-e5 e2-e4 d7-d6 Ng1-f3
#  7     -5     148 g2-g3 d7-d5 d2-d4 Ng8-f6
#  8     -5     147 b2-b3 d7-d5 d2-d4 Ng8-f6
#  9     -7     137 f2-f3 d7-d5 e2-e4 e7-e6
# 10     -7     137 h2-h3 d7-d5 d2-d4 Ng8-f6
# 11     -7     134 c2-c3 d7-d5 d2-d4 Ng8-f6
# 12     -7     136 a2-a3 d7-d5 d2-d4 Ng8-f6
# 13    -14     104 Nb1-a3 d7-d5 Ng1-f3 e7-e6
# 14    -16     101 Ng1-h3 e7-e5 e2-e4 Nb8-c6
# 15     -6     145 f2-f4 d7-d5 e2-e3 e7-e6 d2-d4
# 16     -4     151 c2-c4 e7-e5 d2-d3 d7-d6 e2-e4
# 17    -16     100 h2-h4 d7-d5 d2-d4
# 18    -16     101 a2-a4 d7-d5 d2-d4
# 19    -22      85 g2-g4 d7-d5 e2-e3
# 20    -22      84 b2-b4 e7-e5 Ng1-f3 d7-d6

nodes = 36156127 <92% qnodes> time = 11117ms nps = 3252327 eps = 2392311 nneps = 0
Tree: nodes = 6122792 depth = 62 pps = 17053 visits = 189587 
      qsearch_calls = 1983453 search_calls = 0
move e2e4
Bye Bye
Jörg Oster

Daniel Shawul
Posts: 3810
Joined: Tue Mar 14, 2006 10:34 am
Location: Ethiopia
Contact:

Re: Scorpio 2.8.7 MCTS+NN windows version

Post by Daniel Shawul » Mon Sep 24, 2018 12:47 pm

BrendanJNorman wrote:
Mon Sep 24, 2018 6:55 am
I just played a 3 + 2 blitz game against version 2.8.8, using 2x32 nets with the sole change from default parameters being my changing treeht to 5000.

I gave Scorpio a book made from the decisive games of 2700+ players, because in the first game he played 1.d4 d5 2.Qd3 against me, making it impractical for training without a book.

I am using Windows 10 on a Macbook 2015 model via BootCamp, so the hardware is relatively dated.

Would appreciate any advice on strengthening it up a bit.

Hi Brendan,

I have mentioned this but try backup_type = 0 (minimax intstead of averaging). That is much better tactically.

Daniel

Daniel Shawul
Posts: 3810
Joined: Tue Mar 14, 2006 10:34 am
Location: Ethiopia
Contact:

Re: Scorpio 2.8.7 MCTS+NN windows version

Post by Daniel Shawul » Mon Sep 24, 2018 12:54 pm

Joerg Oster wrote:
Mon Sep 24, 2018 9:46 am
Daniel,
just for your information, the Linux version crashes immediately (memory access fault) if I don't set "use_nn" to 0 in the ini file.
./scorpio go quit
feature done=0
ht 4194304 X 16 = 64.0 MB
eht 524288 X 8 = 8.0 MB
pht 32768 X 24 = 0.8 MB
treeht 26843136 X 40 = 1024.0 MB
processors [3]
EgbbProbe not Loaded!
loading_time = 0s
[st = 11114ms, mt = 29250ms , hply = 0 , moves_left 10]
Speicherzugriffsfehler
I'm still in the process of putting everything together step by step (no nets, no egbb's atm)
and wanted to check if Scorpio is working correctly on my box.

Output with 3 Threads, MCTS, 1024 MB Hashtree size

Code: Select all

./scorpio go quit
feature done=0
ht 4194304 X 16 = 64.0 MB
eht 524288 X 8 = 8.0 MB
pht 32768 X 24 = 0.8 MB
treeht 26843136 X 40 = 1024.0 MB
processors [3]
EgbbProbe not Loaded!
loading_time = 0s
[st = 11114ms, mt = 29250ms , hply = 0 , moves_left 10]
63 38 111 39028  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4
64 39 223 70554  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 d7-d6 d2-d4 Nc6xd4 Nf3xd4 e5xd4 Qd1xd4 Bc8-d7
65 38 335 93586  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 d7-d6 Bf1-c4 Ng8-f6 Ke1-g1 Bc8-e6 Bc4xe6 f7xe6
66 38 447 109927  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4 Bf8-b4 f2-f3 Ke8-g8 Bc1-e3 d7-d5 Nd4xc6 b7xc6 e4-e5 Bb4xc3 b2xc3
67 38 558 124308  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4 Bf8-b4 f2-f3 Ke8-g8 Bf1-b5 a7-a6 Bb5-a4 Bb4xc3 b2xc3 Nc6xd4 c3xd4
68 38 670 139875  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4 Bf8-b4 f2-f3 Ke8-g8 Bf1-b5 a7-a6 Bb5-a4 Nc6xd4 Qd1xd4 c7-c5 Qd4-d2 b7-b5 Ba4-b3 c5-c4 e4-e5 Rf8-e8 Ke1-g1 Re8xe5 Qd2-d4 Bb4-c5 Qd4xc5 Re5xc5
69 38 781 152683  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4 Bf8-b4 f2-f3 Ke8-g8 Bf1-b5 a7-a6 Bb5-a4 Nc6xd4 Qd1xd4 c7-c5 Qd4-d3 b7-b5 e4-e5 Rf8-e8 Ke1-g1 Re8xe5 Bc1-f4 c5-c4 Qd3-d2 Bb4-c5 Kg1-h1 Re5-e6 b2-b4 Bc5xb4 Ra1-b1 Bb4xc3 Qd2xc3 b5xa4 Qc3xc4
70 38 893 165201  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4 Bf8-b4 f2-f3 Ke8-g8 Bf1-b5 a7-a6 Bb5-a4 Nc6xd4 Qd1xd4 c7-c5 Qd4-d3 b7-b5 e4-e5 Rf8-e8 Ke1-g1 Re8xe5 Ba4-b3 d7-d5 a2-a3 c5-c4 Qd3-d4 Bb4xc3 b2xc3 Re5-e2 Bb3-a2 Bc8-f5 Bc1-g5 Qd8-d6 Bg5xf6 g7xf6 f3-f4 Ra8-e8
71 38 1005 178914  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4 Bf8-b4 f2-f3 Ke8-g8 Bf1-b5 a7-a6 Bb5-a4 Nc6xd4 Qd1xd4 c7-c5 Qd4-d1 d7-d5 e4-e5 d5-d4 Ke1-g1 d4xc3 Qd1xd8 Rf8xd8 e5xf6 g7xf6 a2-a3 Bb4-a5 Ba4-b3 Bc8-e6 Bb3xe6 f7xe6 Ra1-b1 b7-b5 b2-b3 Rd8-d5 Rf1-e1 e6-e5 a3-a4 b5-b4 Bc1-e3 Ra8-d8 Rb1-a1 Kg8-g7 Be3-f2

#  1     40  142659 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4 Bf8-b4 f2-f3 Ke8-g8 Bf1-b5 a7-a6 Bb5-a4 Nc6xd4 Qd1xd4 c7-c5 Qd4-d1 d7-d5 e4-e5 d5-d4 Ke1-g1 d4xc3 Qd1xd8 Rf8xd8 e5xf6 g7xf6 a2-a3 Bb4-a5 Ba4-b3 Bc8-e6 Bb3xe6 f7xe6 Ra1-b1 b7-b5 Bc1-e3 c3xb2 Rb1xb2 c5-c4 Rb2-b1 Ba5-c3 Rf1-d1 f6-f5 Be3-b6 Rd8-d5 Rd1xd5 e6xd5 Rb1-d1 Ra8-b8 Bb6-d4 Bc3xd4 Rd1xd4 Rb8-d8 c2-c3 Kg8-f7 Kg1-f2 Kf7-f6 Rd4-d1 d5-d4 Rd1xd4 Rd8xd4 c3xd4
#  2     33   28312 d2-d4 d7-d5 Ng1-f3 Ng8-f6 e2-e3 e7-e6 Bf1-d3 Bf8-e7 Ke1-g1 Ke8-g8 Nb1-c3 Nb8-c6 b2-b3 b7-b6 Bc1-b2 Bc8-b7 a2-a3 Qd8-d7 h2-h3 a7-a6 Qd1-d2 h7-h6 Rf1-e1 Rf8-e8 e3-e4 d5xe4 Nc3xe4 Nf6xe4 Bd3xe4 Be7-f6 c2-c3 a6-a5 Ra1-d1 a5-a4 b3-b4 Kg8-f8 Nf3-e5 Nc6xe5 Be4xb7 Ne5-c4 Qd2-e2 Nc4xb2 Qe2xb2 Ra8-d8 Qb2-c2 e6-e5 d4xe5 Re8xe5 Bb7-f3 Re5xe1 Rd1xe1 Qd7-d3 Qc2xd3 Rd8xd3 Bf3-c6 Rd3-d8 Bc6xa4 Bf6xc3 Re1-e3 Bc3-f6 Ba4-b3
#  3     20     694 Nb1-c3 d7-d5 d2-d4 Ng8-f6
#  4     33   15387 Ng1-f3 d7-d5 Nb1-c3 Ng8-f6 e2-e3 e7-e6 Bf1-e2 Bf8-d6 d2-d4 Ke8-g8 Ke1-g1 Nb8-c6 a2-a3 a7-a6 Bc1-d2 h7-h6 h2-h3 Bc8-d7 Rf1-e1 Qd8-e7 Ra1-c1 Ra8-d8 Be2-d3 e6-e5 e3-e4 d5xe4 Nc3xe4 Nc6xd4 Ne4xd6 Nd4xf3 Qd1xf3 Qe7xd6 Rc1-d1 Bd7-c6 Qf3-g3 e5-e4 Qg3xd6 Rd8xd6 Bd3-c4 Rd6-d4 Bc4-b3 Rf8-d8 Re1-e2 Bc6-b5 c2-c4 Bb5xc4 Bb3xc4 Rd4xc4
#  5     13     393 e2-e3 d7-d5 d2-d4 e7-e6 Ng1-f3
#  6     15     451 d2-d3 e7-e5 e2-e4 d7-d6 Ng1-f3
#  7     -5     148 g2-g3 d7-d5 d2-d4 Ng8-f6
#  8     -5     147 b2-b3 d7-d5 d2-d4 Ng8-f6
#  9     -7     137 f2-f3 d7-d5 e2-e4 e7-e6
# 10     -7     137 h2-h3 d7-d5 d2-d4 Ng8-f6
# 11     -7     134 c2-c3 d7-d5 d2-d4 Ng8-f6
# 12     -7     136 a2-a3 d7-d5 d2-d4 Ng8-f6
# 13    -14     104 Nb1-a3 d7-d5 Ng1-f3 e7-e6
# 14    -16     101 Ng1-h3 e7-e5 e2-e4 Nb8-c6
# 15     -6     145 f2-f4 d7-d5 e2-e3 e7-e6 d2-d4
# 16     -4     151 c2-c4 e7-e5 d2-d3 d7-d6 e2-e4
# 17    -16     100 h2-h4 d7-d5 d2-d4
# 18    -16     101 a2-a4 d7-d5 d2-d4
# 19    -22      85 g2-g4 d7-d5 e2-e3
# 20    -22      84 b2-b4 e7-e5 Ng1-f3 d7-d6

nodes = 36156127 <92% qnodes> time = 11117ms nps = 3252327 eps = 2392311 nneps = 0
Tree: nodes = 6122792 depth = 62 pps = 17053 visits = 189587 
      qsearch_calls = 1983453 search_calls = 0
move e2e4
Bye Bye
Finally someone who uses linux...where have you been ? :)
Note that I haven't provided egbbdll binaries compiled for linux so you have to compile it yourself.
The easiest method there is to use the tensorflow_cc approach I described in the readme after changing USE_TF=1 (not 2).
Then you do a make in the egbbdll src and hopefully you get egbbso.so which you rename as egbbso64.so and put it somewhere
to be used by Scorpio.

The second approach is to put the egbbdll src inside tensorflow (I put it in tensorflow/cc/egbbll) after cloning the tensorflow repo and
build it with bazel like

bazel build --config=opt --config=monolithic //tensorflow/cc/egbbdll:egbbdll.so

Both methods work put i prefer the tensorflow_cc approach for linux.

Daniel

BrendanJNorman
Posts: 1661
Joined: Sun Feb 07, 2016 11:43 pm
Full name: Brendan J Norman

Re: Scorpio 2.8.7 MCTS+NN windows version

Post by BrendanJNorman » Mon Sep 24, 2018 12:54 pm

Daniel Shawul wrote:
Mon Sep 24, 2018 12:47 pm
BrendanJNorman wrote:
Mon Sep 24, 2018 6:55 am
I just played a 3 + 2 blitz game against version 2.8.8, using 2x32 nets with the sole change from default parameters being my changing treeht to 5000.

I gave Scorpio a book made from the decisive games of 2700+ players, because in the first game he played 1.d4 d5 2.Qd3 against me, making it impractical for training without a book.

I am using Windows 10 on a Macbook 2015 model via BootCamp, so the hardware is relatively dated.

Would appreciate any advice on strengthening it up a bit.

Hi Brendan,

I have mentioned this but try backup_type = 0 (minimax intstead of averaging). That is much better tactically.

Daniel
Thanks Daniel, I must have completely missed when you said it.

As a non-programmer, much of this language is meaningless to me, which doesn't help.

But I'll give backup=0 a try.

Thanks again.

Daniel Shawul
Posts: 3810
Joined: Tue Mar 14, 2006 10:34 am
Location: Ethiopia
Contact:

Re: Scorpio 2.8.7 MCTS+NN windows version

Post by Daniel Shawul » Mon Sep 24, 2018 3:49 pm

BrendanJNorman wrote:
Mon Sep 24, 2018 12:54 pm
Daniel Shawul wrote:
Mon Sep 24, 2018 12:47 pm
BrendanJNorman wrote:
Mon Sep 24, 2018 6:55 am
I just played a 3 + 2 blitz game against version 2.8.8, using 2x32 nets with the sole change from default parameters being my changing treeht to 5000.

I gave Scorpio a book made from the decisive games of 2700+ players, because in the first game he played 1.d4 d5 2.Qd3 against me, making it impractical for training without a book.

I am using Windows 10 on a Macbook 2015 model via BootCamp, so the hardware is relatively dated.

Would appreciate any advice on strengthening it up a bit.

Hi Brendan,

I have mentioned this but try backup_type = 0 (minimax intstead of averaging). That is much better tactically.

Daniel
Thanks Daniel, I must have completely missed when you said it.

As a non-programmer, much of this language is meaningless to me, which doesn't help.

But I'll give backup=0 a try.

Thanks again.
Not your fault, I said that in the other scorpio-nn thread not this one. Also use mt=16 that should double your nps compared to mt=1

Code: Select all

backup_type 0
mt 16

Joerg Oster
Posts: 693
Joined: Fri Mar 10, 2006 3:29 pm
Location: Germany

Re: Scorpio 2.8.7 MCTS+NN windows version

Post by Joerg Oster » Mon Sep 24, 2018 4:50 pm

Daniel Shawul wrote:
Mon Sep 24, 2018 12:54 pm
Joerg Oster wrote:
Mon Sep 24, 2018 9:46 am
Daniel,
just for your information, the Linux version crashes immediately (memory access fault) if I don't set "use_nn" to 0 in the ini file.
./scorpio go quit
feature done=0
ht 4194304 X 16 = 64.0 MB
eht 524288 X 8 = 8.0 MB
pht 32768 X 24 = 0.8 MB
treeht 26843136 X 40 = 1024.0 MB
processors [3]
EgbbProbe not Loaded!
loading_time = 0s
[st = 11114ms, mt = 29250ms , hply = 0 , moves_left 10]
Speicherzugriffsfehler
I'm still in the process of putting everything together step by step (no nets, no egbb's atm)
and wanted to check if Scorpio is working correctly on my box.

Output with 3 Threads, MCTS, 1024 MB Hashtree size

Code: Select all

./scorpio go quit
feature done=0
ht 4194304 X 16 = 64.0 MB
eht 524288 X 8 = 8.0 MB
pht 32768 X 24 = 0.8 MB
treeht 26843136 X 40 = 1024.0 MB
processors [3]
EgbbProbe not Loaded!
loading_time = 0s
[st = 11114ms, mt = 29250ms , hply = 0 , moves_left 10]
63 38 111 39028  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4
64 39 223 70554  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 d7-d6 d2-d4 Nc6xd4 Nf3xd4 e5xd4 Qd1xd4 Bc8-d7
65 38 335 93586  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 d7-d6 Bf1-c4 Ng8-f6 Ke1-g1 Bc8-e6 Bc4xe6 f7xe6
66 38 447 109927  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4 Bf8-b4 f2-f3 Ke8-g8 Bc1-e3 d7-d5 Nd4xc6 b7xc6 e4-e5 Bb4xc3 b2xc3
67 38 558 124308  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4 Bf8-b4 f2-f3 Ke8-g8 Bf1-b5 a7-a6 Bb5-a4 Bb4xc3 b2xc3 Nc6xd4 c3xd4
68 38 670 139875  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4 Bf8-b4 f2-f3 Ke8-g8 Bf1-b5 a7-a6 Bb5-a4 Nc6xd4 Qd1xd4 c7-c5 Qd4-d2 b7-b5 Ba4-b3 c5-c4 e4-e5 Rf8-e8 Ke1-g1 Re8xe5 Qd2-d4 Bb4-c5 Qd4xc5 Re5xc5
69 38 781 152683  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4 Bf8-b4 f2-f3 Ke8-g8 Bf1-b5 a7-a6 Bb5-a4 Nc6xd4 Qd1xd4 c7-c5 Qd4-d3 b7-b5 e4-e5 Rf8-e8 Ke1-g1 Re8xe5 Bc1-f4 c5-c4 Qd3-d2 Bb4-c5 Kg1-h1 Re5-e6 b2-b4 Bc5xb4 Ra1-b1 Bb4xc3 Qd2xc3 b5xa4 Qc3xc4
70 38 893 165201  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4 Bf8-b4 f2-f3 Ke8-g8 Bf1-b5 a7-a6 Bb5-a4 Nc6xd4 Qd1xd4 c7-c5 Qd4-d3 b7-b5 e4-e5 Rf8-e8 Ke1-g1 Re8xe5 Ba4-b3 d7-d5 a2-a3 c5-c4 Qd3-d4 Bb4xc3 b2xc3 Re5-e2 Bb3-a2 Bc8-f5 Bc1-g5 Qd8-d6 Bg5xf6 g7xf6 f3-f4 Ra8-e8
71 38 1005 178914  e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4 Bf8-b4 f2-f3 Ke8-g8 Bf1-b5 a7-a6 Bb5-a4 Nc6xd4 Qd1xd4 c7-c5 Qd4-d1 d7-d5 e4-e5 d5-d4 Ke1-g1 d4xc3 Qd1xd8 Rf8xd8 e5xf6 g7xf6 a2-a3 Bb4-a5 Ba4-b3 Bc8-e6 Bb3xe6 f7xe6 Ra1-b1 b7-b5 b2-b3 Rd8-d5 Rf1-e1 e6-e5 a3-a4 b5-b4 Bc1-e3 Ra8-d8 Rb1-a1 Kg8-g7 Be3-f2

#  1     40  142659 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Nf3xd4 Bf8-b4 f2-f3 Ke8-g8 Bf1-b5 a7-a6 Bb5-a4 Nc6xd4 Qd1xd4 c7-c5 Qd4-d1 d7-d5 e4-e5 d5-d4 Ke1-g1 d4xc3 Qd1xd8 Rf8xd8 e5xf6 g7xf6 a2-a3 Bb4-a5 Ba4-b3 Bc8-e6 Bb3xe6 f7xe6 Ra1-b1 b7-b5 Bc1-e3 c3xb2 Rb1xb2 c5-c4 Rb2-b1 Ba5-c3 Rf1-d1 f6-f5 Be3-b6 Rd8-d5 Rd1xd5 e6xd5 Rb1-d1 Ra8-b8 Bb6-d4 Bc3xd4 Rd1xd4 Rb8-d8 c2-c3 Kg8-f7 Kg1-f2 Kf7-f6 Rd4-d1 d5-d4 Rd1xd4 Rd8xd4 c3xd4
#  2     33   28312 d2-d4 d7-d5 Ng1-f3 Ng8-f6 e2-e3 e7-e6 Bf1-d3 Bf8-e7 Ke1-g1 Ke8-g8 Nb1-c3 Nb8-c6 b2-b3 b7-b6 Bc1-b2 Bc8-b7 a2-a3 Qd8-d7 h2-h3 a7-a6 Qd1-d2 h7-h6 Rf1-e1 Rf8-e8 e3-e4 d5xe4 Nc3xe4 Nf6xe4 Bd3xe4 Be7-f6 c2-c3 a6-a5 Ra1-d1 a5-a4 b3-b4 Kg8-f8 Nf3-e5 Nc6xe5 Be4xb7 Ne5-c4 Qd2-e2 Nc4xb2 Qe2xb2 Ra8-d8 Qb2-c2 e6-e5 d4xe5 Re8xe5 Bb7-f3 Re5xe1 Rd1xe1 Qd7-d3 Qc2xd3 Rd8xd3 Bf3-c6 Rd3-d8 Bc6xa4 Bf6xc3 Re1-e3 Bc3-f6 Ba4-b3
#  3     20     694 Nb1-c3 d7-d5 d2-d4 Ng8-f6
#  4     33   15387 Ng1-f3 d7-d5 Nb1-c3 Ng8-f6 e2-e3 e7-e6 Bf1-e2 Bf8-d6 d2-d4 Ke8-g8 Ke1-g1 Nb8-c6 a2-a3 a7-a6 Bc1-d2 h7-h6 h2-h3 Bc8-d7 Rf1-e1 Qd8-e7 Ra1-c1 Ra8-d8 Be2-d3 e6-e5 e3-e4 d5xe4 Nc3xe4 Nc6xd4 Ne4xd6 Nd4xf3 Qd1xf3 Qe7xd6 Rc1-d1 Bd7-c6 Qf3-g3 e5-e4 Qg3xd6 Rd8xd6 Bd3-c4 Rd6-d4 Bc4-b3 Rf8-d8 Re1-e2 Bc6-b5 c2-c4 Bb5xc4 Bb3xc4 Rd4xc4
#  5     13     393 e2-e3 d7-d5 d2-d4 e7-e6 Ng1-f3
#  6     15     451 d2-d3 e7-e5 e2-e4 d7-d6 Ng1-f3
#  7     -5     148 g2-g3 d7-d5 d2-d4 Ng8-f6
#  8     -5     147 b2-b3 d7-d5 d2-d4 Ng8-f6
#  9     -7     137 f2-f3 d7-d5 e2-e4 e7-e6
# 10     -7     137 h2-h3 d7-d5 d2-d4 Ng8-f6
# 11     -7     134 c2-c3 d7-d5 d2-d4 Ng8-f6
# 12     -7     136 a2-a3 d7-d5 d2-d4 Ng8-f6
# 13    -14     104 Nb1-a3 d7-d5 Ng1-f3 e7-e6
# 14    -16     101 Ng1-h3 e7-e5 e2-e4 Nb8-c6
# 15     -6     145 f2-f4 d7-d5 e2-e3 e7-e6 d2-d4
# 16     -4     151 c2-c4 e7-e5 d2-d3 d7-d6 e2-e4
# 17    -16     100 h2-h4 d7-d5 d2-d4
# 18    -16     101 a2-a4 d7-d5 d2-d4
# 19    -22      85 g2-g4 d7-d5 e2-e3
# 20    -22      84 b2-b4 e7-e5 Ng1-f3 d7-d6

nodes = 36156127 <92% qnodes> time = 11117ms nps = 3252327 eps = 2392311 nneps = 0
Tree: nodes = 6122792 depth = 62 pps = 17053 visits = 189587 
      qsearch_calls = 1983453 search_calls = 0
move e2e4
Bye Bye
Finally someone who uses linux...where have you been ? :)
Note that I haven't provided egbbdll binaries compiled for linux so you have to compile it yourself.
The easiest method there is to use the tensorflow_cc approach I described in the readme after changing USE_TF=1 (not 2).
Then you do a make in the egbbdll src and hopefully you get egbbso.so which you rename as egbbso64.so and put it somewhere
to be used by Scorpio.

The second approach is to put the egbbdll src inside tensorflow (I put it in tensorflow/cc/egbbll) after cloning the tensorflow repo and
build it with bazel like

bazel build --config=opt --config=monolithic //tensorflow/cc/egbbdll:egbbdll.so

Both methods work put i prefer the tensorflow_cc approach for linux.

Daniel
I also prefer the 1st approach.

I'm currently in the process of building and installing tensorflow_cc.
There are many warnings during the process, though.

OK, seems to have finished correctly.

Unfortunately, building egbbdll fails.
It looks like there are many header files included inside the tensorflow_cc header files
like "error_codes.pb.h", which aren't present after building and installation.
It seems I don't have any header file with "pb.h" in its name.

Maybe something went wrong during the build process?

Btw., where can I download the egbb files currently?
The links on your website seems to be outdated.
Jörg Oster

Post Reply