/MCTSExperiments.jl

Test repository for the Colorado Julia Meetup to experiment with Monte Carlo Tree Search implementations.

Primary LanguageJulia

Setup

Open a julia 1.6 REPL in the main directory with the --project option. Then enter package mode by typing ] and type instantiate to install dependencies. After this hit backspace to enter the normal julia repl and the module will be accessible by typing using MCTSExperiments

Run example

Open a julia 1.6 REPL in the MCTS environment using the --project option. If the working directory is the main MCTS folder, you can run the code in the 'game2048.jl' file in the examples folder by typing include(joinpath(examples, "game2048.jl")) which should print out a gameboard after a series of random moves. Also the file "DecisionMakingProblems2048.jl" has more examples of implementing an MCTS policy using the 2048 environment contained in the DecisionMakingProblems repository. Some of the lines in this file are benchmarks which may take several minutes to run and can be commented out to avoid waiting.

To Do

  • Create MDP and MCTS policy for 2048 using MCTS.jl found in JuliaPOMDP
  • Compare timing and quality of policies between each implementation
  • Check for type stability and dictionary garbage collection in MCTS implementation
  • Test/compare MCTS using Dictionaries.jl instead of Base.Dict