Help with Scorpio 3.0.8

[Carlos777]

Hello,

I am trying to make Scorpio 3.0.8 MCTS NN (no GPU card, just CPU) play against an engine under Winboard, but it seems to crash and lose on time after book moves.

This is my ini config:

Code: Select all

###################################################
# log on/off   - turn logging on and off
# resign       - resign value in centipawns
# contempt     - +ve value discourages drawishness
###################################################
log                 off
resign              700
contempt            0
##################################################################
# montecarlo - Use montecarlo tree search (MCTS) if set to 1
# treeht - Maximum size of tree to store in memory given in MB.
#          Note that this memory is not allocated at start up; it only
#          specifies the maximum limit the tree could grow to.
#          Note: Don't think you are making scorpio weak by not setting
#          this memory to high value, infact treeht=0 is the strongest 
#          MCTS setting.
# frac_freeze_tree    [0 - 100] - freeze tree after this percentage of time is spent
# frac_alphabeta      [0 - 100] - rollouts alpha-beta search percentage when using AB-rollouts + MCTS with shared tree
# frac_abrollouts     [0 - 100] - rollouts alpha-beta search percentage when using AB-rollouts + AB with shared tree
# frac_abprior        [0 - 100] - standard alpha-beta search percentage when using AB + MCTS, and AB used as prior
# alphabeta_man_c     [0 - 32]  - switch to regular alpha-beta search with this many pieces on board
# mcts_strategy_depth [0 - 64]  - Lookahead depth for MCTS rollouts
# alphabeta_depth     [0 - 64]  - Fixed search depth for standard alphabeta searches at the leaves
# evaluate_depth      [-4 - 64]  - Fixed search depth for evaluation of new children during expansion 
#                                (-4=eval,-3=see,-2=recap qsearch,-1=no-checks qsearch 0=qsearch,x=search)
# backup_type         [0 - 7]   - minmax = 0, averaging = 1, mixed = 2, for mixing score with 
#                                 previous ID use 3,4,5 resp. 6 is classic MCTS, 7 is mixing by visit count
# reuse_tree                    - reuse MCTS tree between moves
# virtual_loss                  - virtual loss for parallel search
# visit_threshold               - minimum number of visits for minimax backup
# policy_temp                   - Temperature for move policy
# cpuct_init                    - Starting cpuct value
# cpuct_base                    - Denominator of cpuct forumula
# fpu_is_loss                   - FPU is loss (1=true, 0=false)
# fpu_red                       - Reduction factor for FPU
##################################################################
montecarlo          1
treeht              12800
frac_freeze_tree    100
frac_alphabeta      0
frac_abrollouts     20
frac_abprior        30
alphabeta_man_c     20
mcts_strategy_depth 30
alphabeta_depth     16
evaluate_depth      0
backup_type         6
reuse_tree          1
virtual_loss        1
visit_threshold     3200
policy_temp         235
cpuct_init          125
cpuct_base          19652
fpu_is_loss         0
fpu_red             33
#############################################################################
# Parallel search setup
#   mt            - number of processors
#   smp_depth     - split depth for smp
#   cluster_depth - split depth for cluster search
#   message_poll_nodes - number nodes to search before checking for message
#############################################################################
mt                  1
smp_depth           8
cluster_depth       12
message_poll_nodes  20
smp_type            ABDADA
cluster_type        YBW 
##################################################################
# Hashtable sizes in mb.
#      ht - main hash table shared by ALL threads
#     eht - evalualtion hash table allocated PER thread
#     pht - pawn hash table allocated PER thread
# Usually 1m or 2mb is enough for pawn hash table, but the others
# can be set as high as possible
##################################################################
ht                  128
eht                 4
pht                 1
##########################################################################
# EGBB
# egbb_path              - path to dll/so file for probing endgame egbbs 
# egbb_files_path        - path to endgame egbbs, if not specified it takes
#                          value of egbb_path
# egbb_cache_size        - cache size for egbbs in MBs
# egbb_ply_limit_percent - percent of plies from the root to egbb probing
# egbb_depth_limit       - depth from the leaves to egbb probing
# egbb_load_type         - egbb load type with the following values
#                    0 = none are loaded
#                    1 = all 3/4 men are loaded
#                    2 = Not implemented yet
#                    3 = all 5 men are loaded
##########################################################################
egbb_path                /home/daniel/egbb/
egbb_files_path          /home/daniel/egbb/
egbb_cache_size          32
egbb_load_type           3
egbb_depth_limit         6
#######################################
# NN
# use_nn                 - turn off/on neural network
# nn_cache_size          - neural network cache size in MBs
# nn_path                - path to neural network
# nn_type                - type of neural network (0=scorpio, 1=lczero)
# device_type            - CPU or GPU
# delay                  - Sleep threads for this amount. Set this to 1
#                          if you are using way too many threads than you
#                          have physical cores (maybe when >16x more).
# float_type             - inference precison for tensorrt = FLOAT/HALF/INT8
# wdl_head               - network has WDL head (e.g JH networks do)
# win/draw/loss_weight   - weights for WDL resp in precent, range is [0...1000]          
# min_policy_value       - minimum policy value in per mill. For 1%, set to 10
#######################################
use_nn                   0
nn_cache_size            200
nn_type                  0
wdl_head                 0
nn_path                  C:\WinBoard-4.6.2\Engines\scorpio30-mcts-nn\nets-lczero\nets-lczero\ID-42850.uff
device_type              CPU
n_devices                1
delay                    0
float_type               HALF
win_weight               100
draw_weight              100
loss_weight              100
min_policy_value         10
################################################################
# Multiple nets settings
#------------------------
# The opening is played by default network specified above
#   nn_man_m/e      -- threshold piece counts for middle/end game
#   nn_type_m/e     -- neural network type for middle/end game
#                      >=0       = as specified before
#                      -1/-2/-3  = used to refer to either opening/midgame/endgame network
#   nn_path_m/e     -- path to neural networks for middle/end game
#   cpuct_init_m/e  -- cpuct for middle/endgame
#   policy_temp_m/e -- policy temperature for middle/endgame
#   wdl_head_m/e    -- wdl value head for middle/endgame
#   ensemble        -- percent of time to use ensembling [0-100]
#   ensemble_type   -- 0 = arthimetic mean, 1 = root mean cube
################################################################
nn_man_m                 24
nn_man_e                 14
nn_type_m                -1
nn_type_e                -1
nn_path_m                C:\WinBoard-4.6.2\Engines\scorpio30-mcts-nn\nets-lczero\nets-lczero\ID-42850.uff
nn_path_e                C:\WinBoard-4.6.2\Engines\scorpio30-mcts-nn\nets-maddex\ME.uff
cpuct_init_m             125
cpuct_init_e             125
policy_temp_m            235
policy_temp_e            235
wdl_head_m               0
wdl_head_e               0
ensemble                 0
ensemble_type            0

And my command line:

"Scorpio 3.0.8 MCTS" -fcp "scorpio.exe" -fd "C:\WinBoard-4.6.2\Engines\scorpio30-mcts-nn\bin\Windows" -fn "Scorpio 3.0.8 MCTS" -fNoOwnBookUCI

What could be wrong?

[Dann Corbit]

I just built it from source and I get a segfault also.

[Dann Corbit]

Code: Select all

go infinite
# rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
# [st = 12502ms, mt = 29250ms , hply = 0 , moves_left 10]

Thread 1 received signal SIGSEGV, Segmentation fault.
Python Exception <class 'NameError'> Installation error: gdb._execute_unwinders function is missing:
0x000000000040ead7 in SEARCHER::probe_neural_(bool, float*) ()
(gdb)

[Daniel Shawul]

Building from source the NN components is probably impossible for most people, need to downlaod tensorrt, tensorflow, cudnn etc
and use esoteric build systems such as bazel from google. So yeah I don't recommend that!

To get scorpio working without NN is easy however, by simply turning it off in scorpio.ini "use_nn 0". And you can use scorpio.exe directly in this case.

Installing the NN version is easy, however, using the instructions I give here.
https://github.com/dshawul/Scorpio/blob ... INSTALL.md
Basically you download only one install script, which will then download binaries and configure everything for you.

When installing it as an engine in a GUI, use the scripts (scorpio.sh or scorpio.bat) instead of scorpio.exe
This is needed for setting paths to the required cuddnn/tensorrt libraries.

If you are interested in contributing training games, go here http://scorpiozero.ddns.net and download the client.
Currently, we are two people generating games and we already managed to do 1 million in 4 days i think ...

[Daniel Shawul]

Btw, Scorpio on CPU using NN has become stronger than scorpio AB so you can still use NN on CPU.
It will install a 12b network when you only have a CPU.

[Carlos777]

Btw, Scorpio on CPU using NN has become stronger than scorpio AB so you can still use NN on CPU.
It will install a 12b network when you only have a CPU.

Thank you for your help. I just ran install.bat and got this:

Code: Select all

###################################################
# log on/off   - turn logging on and off
# resign       - resign value in centipawns
# contempt     - +ve value discourages drawishness
###################################################
log                 off
resign              700
contempt            0
##################################################################
# montecarlo - Use montecarlo tree search (MCTS) if set to 1
# treeht - Maximum size of tree to store in memory given in MB.
#          Note that this memory is not allocated at start up; it only
#          specifies the maximum limit the tree could grow to.
#          Note: Don't think you are making scorpio weak by not setting
#          this memory to high value, infact treeht=0 is the strongest 
#          MCTS setting.
# frac_freeze_tree    [0 - 100] - freeze tree after this percentage of time is spent
# frac_alphabeta      [0 - 100] - rollouts alpha-beta search percentage when using AB-rollouts + MCTS with shared tree
# frac_abrollouts     [0 - 100] - rollouts alpha-beta search percentage when using AB-rollouts + AB with shared tree
# frac_abprior        [0 - 100] - standard alpha-beta search percentage when using AB + MCTS, and AB used as prior
# alphabeta_man_c     [0 - 32]  - switch to regular alpha-beta search with this many pieces on board
# mcts_strategy_depth [0 - 64]  - Lookahead depth for MCTS rollouts
# alphabeta_depth     [0 - 64]  - Fixed search depth for standard alphabeta searches at the leaves
# evaluate_depth      [-4 - 64]  - Fixed search depth for evaluation of new children during expansion 
#                                (-4=eval,-3=see,-2=recap qsearch,-1=no-checks qsearch 0=qsearch,x=search)
# backup_type         [0 - 7]   - minmax = 0, averaging = 1, mixed = 2, for mixing score with 
#                                 previous ID use 3,4,5 resp. 6 is classic MCTS, 7 is mixing by visit count
# reuse_tree                    - reuse MCTS tree between moves
# virtual_loss                  - virtual loss for parallel search
# visit_threshold               - minimum number of visits for minimax backup
# policy_temp                   - Temperature for move policy
# cpuct_init                    - Starting cpuct value
# cpuct_base                    - Denominator of cpuct forumula
# fpu_is_loss                   - FPU is loss (1=true, 0=false)
# fpu_red                       - Reduction factor for FPU
##################################################################
montecarlo          1
treeht              12800
frac_freeze_tree    100
frac_alphabeta      0
frac_abrollouts     20
frac_abprior        30
alphabeta_man_c     20
mcts_strategy_depth 30
alphabeta_depth     16
evaluate_depth      0
backup_type         6
reuse_tree          1
virtual_loss        1
visit_threshold     3200
policy_temp         235
cpuct_init          125
cpuct_base          19652
fpu_is_loss         0
fpu_red             33
#############################################################################
# Parallel search setup
#   mt            - number of processors
#   smp_depth     - split depth for smp
#   cluster_depth - split depth for cluster search
#   message_poll_nodes - number nodes to search before checking for message
#############################################################################
mt                  1 
smp_depth           8
cluster_depth       12
message_poll_nodes  20
smp_type            ABDADA
cluster_type        YBW 
##################################################################
# Hashtable sizes in mb.
#      ht - main hash table shared by ALL threads
#     eht - evalualtion hash table allocated PER thread
#     pht - pawn hash table allocated PER thread
# Usually 1m or 2mb is enough for pawn hash table, but the others
# can be set as high as possible
##################################################################
ht                  128
eht                 4
pht                 1
##########################################################################
# EGBB
# egbb_path              - path to dll/so file for probing endgame egbbs 
# egbb_files_path        - path to endgame egbbs, if not specified it takes
#                          value of egbb_path
# egbb_cache_size        - cache size for egbbs in MBs
# egbb_ply_limit_percent - percent of plies from the root to egbb probing
# egbb_depth_limit       - depth from the leaves to egbb probing
# egbb_load_type         - egbb load type with the following values
#                    0 = none are loaded
#                    1 = all 3/4 men are loaded
#                    2 = Not implemented yet
#                    3 = all 5 men are loaded
##########################################################################
egbb_path                C:\WinBoard-4.6.2\Engines\scorpio30-mcts-nn\bin\Windows\Scorpio\nnprobe-windows-cpu
egbb_files_path          C:\WinBoard-4.6.2\Engines\scorpio30-mcts-nn\bin\Windows\Scorpio\egbb
egbb_cache_size          32
egbb_load_type           3
egbb_depth_limit         6
#######################################
# NN
# use_nn                 - turn off/on neural network
# nn_cache_size          - neural network cache size in MBs
# nn_path                - path to neural network
# nn_type                - type of neural network (0=scorpio, 1=lczero)
# device_type            - CPU or GPU
# delay                  - Sleep threads for this amount. Set this to 1
#                          if you are using way too many threads than you
#                          have physical cores (maybe when >16x more).
# float_type             - inference precison for tensorrt = FLOAT/HALF/INT8
# wdl_head               - network has WDL head (e.g JH networks do)
# win/draw/loss_weight   - weights for WDL resp in precent, range is [0...1000]          
# min_policy_value       - minimum policy value in per mill. For 1%, set to 10
#######################################
use_nn                   0
nn_cache_size            200
nn_type                  0 
wdl_head                 0 
nn_path                  C:\WinBoard-4.6.2\Engines\scorpio30-mcts-nn\bin\Windows\Scorpio\nets-scorpio/ens-net-12x128.pb 
device_type              CPU
n_devices                1 
delay                    1 
float_type               FLOAT  
win_weight               100
draw_weight              100
loss_weight              100
min_policy_value         10
################################################################
# Multiple nets settings
#------------------------
# The opening is played by default network specified above
#   nn_man_m/e      -- threshold piece counts for middle/end game
#   nn_type_m/e     -- neural network type for middle/end game
#                      >=0       = as specified before
#                      -1/-2/-3  = used to refer to either opening/midgame/endgame network
#   nn_path_m/e     -- path to neural networks for middle/end game
#   cpuct_init_m/e  -- cpuct for middle/endgame
#   policy_temp_m/e -- policy temperature for middle/endgame
#   wdl_head_m/e    -- wdl value head for middle/endgame
#   ensemble        -- percent of time to use ensembling [0-100]
#   ensemble_type   -- 0 = arthimetic mean, 1 = root mean cube
################################################################
nn_man_m                 24
nn_man_e                 14
nn_type_m                -1
nn_type_e                -1
nn_path_m                C:\WinBoard-4.6.2\Engines\scorpio30-mcts-nn\nets-lczero\nets-lczero\ID-42850.uff
nn_path_e                C:\WinBoard-4.6.2\Engines\scorpio30-mcts-nn\nets-maddex\ME.uff
cpuct_init_m             125
cpuct_init_e             125
policy_temp_m            235
policy_temp_e            235
wdl_head_m               0
wdl_head_e               0
ensemble                 0
ensemble_type            0

Should not the parameter use_nn be 1? Also, the nn_path points to a unexistent folder. After I execute install.bat, it creates a "Scorpio" empty folder inside C:\WinBoard-4.6.2\Engines\scorpio30-mcts-nn\bin\Windows

[Daniel Shawul]

It looks like you have an old installer file. The installer should automatically set "use_nn 1".
Also, I see you have ID-48250 of lczero nets etc, so I believe you are editing scorpio.ini manually.
First, try a clean install and you do not need to edit scorpio.ini at all, as it will take care of everything for you.

The steps are:

1) Choose a directory without spaces in the path and download and put install.bat there.
Please download install.bat fresh from this link. It is also recommended to do this periodically whenever you want to upgrade
as I might change stuff there.
https://github.com/dshawul/Scorpio/rele ... nstall.bat
Also, make sure there is no "Scorpio" folder in the install path, or remove all of them since you do not need older versions.


2) Run install.bat without parameters on the command line.
It will download necessary files (networks, cudnn etc) and extracts and install them.
Then, It will make a "test run" when it finishes installing to confirm it is working properly.

3) Then install scorpio.bat as the engine in your favorite GUI as a UCI or xboard engine.

That is it! You do not need to edit scorpio.ini if you follow these steps.

To install 6-men egbbs and do other advanced stuff, read the installation document
https://github.com/dshawul/Scorpio/blob ... INSTALL.md

regards,
Daniel

[Gabor Szots]

Hello Daniel,

I have followed the steps you described above, then installed Scorpio as a WB engine under Arena. I observed that the nn to use is ens-net-12x128.pb. I wonder if this net is going to change in the future, that is, the name remains the same but the contents will be different.
This is important because now, as I am preparing for testing, I had to name Scorpio as "Scorpio 3.0.8 MCTS NN=ens-net-12x128 64-bit" which is rather long but could even grow longer if I have to specify e.g. the date of the NN file too.

Gabor

[Gabor Szots]

BTW, I have difficulties assessing its strength. I have selected opponents in the range of Scorpio 2.8 but they seem too strong for 3.0.8 MCTS NN, it gets mated rather early. Is that normal or am I doing something wrong?

[Daniel Shawul]

Hi Gabor,

Unless you use a really fast time control, I suspect it is as strong as Scorpio-AB and will be much stronger at longer time controls.
Think of lc0-CPU minus some elos for my net being weaker.
I used to use 6b net on the CPU before which is faster but found out the bigger 12b is stronger even on the CPU.

Could you print what you get when you run "scorpio.bat go quit" from the "bin/Windows" directory?
I want to see how much nps it is getting from your CPU, and also if things are working properly.
Here is what I get on my laptop's CPU for comaprison:

Code: Select all

~/Scorpio-install/Scorpio/bin/Linux$ ./scorpio.sh go quit
feature done=0
Number of cores 4 of 4
treeht 83886080 X 320 = 25600.0 MB
processors [16]
ht 67108864 X 16 = 1024.0 MB
eht 262144 X 8 X 16 = 64.0 MB
pht 32768 X 24 X 16 = 12.0 MB
EgbbProbe 4.3 by Daniel Shawul
egbb_cache 4084 X 8216 = 32.0 MB
180 egbbs loaded !      
nn_cache 131072 X 1568 = 196.0 MB
Loading neural network : /home/daniel/Scorpio-install/Scorpio-03-Jun-2020/nets-scorpio/ens-net-12x128.pb
With "FLOAT" precision
Loading graph on /cpu:0
0. main_input = -1 -1 32 
1. policy/Reshape
0. value/BiasAdd
Neural network loaded !	
loading_time = 0s
# rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
# [st = 12502ms, mt = 29250ms , hply = 0 , moves_left 10]
63 25 95 301 63 315 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 a7-a6 Bb5-a4 Ng8-f6 Ke1-g1 Bf8-e7 Rf1-e1
64 31 196 654 64 333 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 a7-a6 Bb5-a4 Ng8-f6 Ke1-g1 Bf8-e7 Rf1-e1 b7-b5 Ba4-b3 d7-d6 c2-c3 Ke8-g8
65 33 284 972 65 342 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 a7-a6 Bb5-a4 Ng8-f6 Ke1-g1 Bf8-e7 Rf1-e1 b7-b5 Ba4-b3 d7-d6 c2-c3 Ke8-g8 h2-h3 Bc8-b7 d2-d4
66 33 388 1370 66 352 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 Ng8-f6 Ke1-g1 Nf6xe4 Rf1-e1 Ne4-d6 Nf3xe5 Bf8-e7 Bb5-f1 Nc6xe5
67 32 489 1795 67 366 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 Ng8-f6 Ke1-g1 Nf6xe4 Rf1-e1 Ne4-d6 Nf3xe5 Bf8-e7 Bb5-f1 Nc6xe5 Re1xe5 Ke8-g8 d2-d4
68 31 597 2262 68 378 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 Ng8-f6 Ke1-g1 Nf6xe4 Rf1-e1 Ne4-d6 Nf3xe5 Bf8-e7 Bb5-f1 Nc6xe5 Re1xe5 Ke8-g8 d2-d4 Be7-f6 Re5-e1
69 31 706 2663 69 376 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 Ng8-f6 Ke1-g1 Nf6xe4 Rf1-e1 Ne4-d6 Nf3xe5 Bf8-e7 Bb5-f1 Nc6xe5 Re1xe5 Ke8-g8 d2-d4 Be7-f6 Re5-e1 Rf8-e8 c2-c3 Re8xe1
70 31 806 3081 70 381 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 Ng8-f6 Ke1-g1 Nf6xe4 Rf1-e1 Ne4-d6 Nf3xe5 Bf8-e7 Bb5-f1 Nc6xe5 Re1xe5 Ke8-g8 d2-d4 Be7-f6 Re5-e1 Rf8-e8 c2-c3 Re8xe1 Qd1xe1
71 32 905 3450 71 380 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 Ng8-f6 Ke1-g1 Nf6xe4 Rf1-e1 Ne4-d6 Nf3xe5 Bf8-e7 Bb5-f1 Nc6xe5 Re1xe5 Ke8-g8 d2-d4 Be7-f6 Re5-e1 Rf8-e8 c2-c3 Re8xe1 Qd1xe1 Nd6-e8 Bc1-f4
72 32 1003 3810 72 379 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 Ng8-f6 Ke1-g1 Nf6xe4 Rf1-e1 Ne4-d6 Nf3xe5 Bf8-e7 Bb5-f1 Nc6xe5 Re1xe5 Ke8-g8 d2-d4 Be7-f6 Re5-e1 Rf8-e8 c2-c3 Re8xe1 Qd1xe1 Nd6-e8 Bc1-f4 d7-d5 Nb1-d2
73 32 1009 3816 73 377 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 Ng8-f6 Ke1-g1 Nf6xe4 Rf1-e1 Ne4-d6 Nf3xe5 Bf8-e7 Bb5-f1 Nc6xe5 Re1xe5 Ke8-g8 d2-d4 Be7-f6 Re5-e1 Rf8-e8 c2-c3 Re8xe1 Qd1xe1 Nd6-e8 Bc1-f4 d7-d5 Nb1-d2
# nodes 29577 <0% qnodes> time 10185ms nps 2903 eps 0 nneps 379 tbhits 0
# Tree: nodes 4928 depth 26 pps 378 visits 3826 qcalls 0 scalls 0
move e2e4
Bye Bye

It only gets 378 playouts per second, but look how long and sensible the PVs are. So it is a formidable opponent unless something is off in the installation.

You could use a name like, "Scorpio 3.8 NN-CPU-12b" for it.

regards,
Daniel