Monte Carlo Tree Search

Implemented by

• Andrea Mascaretti ([email protected])
• Michele Pellegrino ([email protected])

(Source code available on GitHub)

Requirements

• C++11 compiler (tested with g++ and clang)
• MPI (for parallelization)

Optional:

• R (for data analysis and plots)
  • the tidyverse package
  • the forecats package

Compiling

• Run $ make [options] in the main directory.
  • Selecting DEBUG=yes (DEBUG=no) enables (disable) debugging (with GDB, for instance)

Running

• Parallel:

  • run $ mpiexec -np [number of cores] [executable] to run the parallel scripts.

• Speedup test:

  • run test_gen.sh to run all the tests in the folder speedup_test.

• Serial:

  • run $ ./[executable] to execute the serial scripts.

Description

• src: contains the source code and the template classes that implement both the games and the search tree
• examples: contains examples of matches, namely ai vs. random player, ai vs. ai and ai vs. human player
• test: contains various tests concerning data structures (struct_test), games (nim_test and oxo_test), parallel speedup (speedup_test) and data analysis (analysis)

Developed and tested on Debian, CentOs and MacOS X