README

[ KataGo Config ]

Example KataGo setup to use as external joseki engine.

This is intended for raspberry pi (CPU build), so using small network and few visits.
Feel free to experiment with other networks / settings.


- Build:

    $ git clone https://github.com/lightvector/KataGo
    $ cd KataGo
    $ cd cpp
    $ cmake . -DUSE_BACKEND=EIGEN

  Should give you a CPU build (Eigen backend) once you have all the prerequisites installed.


- Network:

  Get 10b network with extended training from:
  
    https://katagoarchive.org/g170/neuralnets/index.html   (g170e-b10c128-s1141046784-d204142634.bin.gz)  or
    https://katagotraining.org/networks/                   (kata1-b10c128-s1141046784-d204142634.txt.gz)
    
  .bin.gz version preferred as it will load faster, otherwise they're the same.

  Rename or symlink it as 'default_model.bin.gz' in the directory where katago runs.


- Config:

  Copy 'default_gtp.cfg' (this directory) in the directory where katago runs.

  Note: This is for KataGo v1.11.0.
  If using different version start from its default config file and adapt settings. 

  Settings we care about:
    - Don't resign		(allowResignation = false)
    - Log to file		(logToStderr = false   and   logFile = /tmp/kata.log)
    - Few visits		(maxPlayouts = 15)
    - Number of threads		(numSearchThreads = 4   : Number of cores you have)