Neural network model learning simple SC2 agent combat

Agent utilizing a neural network to learn a simple "kill all enemies" trail.

Instructions

• Install the Starcraft 2 game on a Windows machine
  • Create an account and download the Battle.net application from https://www.blizzard.com/en-us/apps/battle.net/desktop
    • Install the free Starcraft 2 game from within the Battle.net application (25.5 gigabytes)
      • The project was completed using Starcraft 2 version 4.11.1.77474, other versions may be supported, but is unconfirmed
• Add Starcraft 2 install folder to the Windows path variables
• Move the supplied simple_1v1.SC2Map file into \Maps\mini_games\
• Install Python version 3.5.2 from https://www.python.org/downloads/release/python-352/
• Install the required python packages using pip and the supplied package list
  • Run Command: "pip install -r package_list.txt"
• Download this git repository
  • My code (author: 100% me):
    • Train and save model iterativly: trainer_dqn_mlp_std_simple.py
      • Models saved in gym_ouput/
    • Run trained model: run_model.py
      • Change model_iteration variable to switch which model iteration to run
    • Code that actually interfaces with Starcraft, creates features, and executes returned model action: \customized_environments\envs\my_agent.py
• Run command to train model
  • Run command "python trainer_dqn_mlp_std_simple.py"
  • Let the trainer work for a couple of minutes to generate a few model iterations
    • If the solution does not find the map simple_1v1 map (linkage error), rename the simple_1v1.SC2Map map in \Maps\mini_games\ to one of the other maps in the folder, and give the same name to the 'map_name' variable in \customized_environments\envs\my_agent.py
• Run command to evaluate model
  • Change the 'realtime' variable in my_agent.py to true
  • Change model_iteration variable to switch which model iteration to run
  • Run command "python run_model.py"

versions of dependency used during development

• Starcraft 2 - 4.11.3.77661
• Python - 3.5.6
• Gym - 0.15.4
• protobuf - 3.11.1
• PySC2 - 3.0.0
• stable-baselines - 2.8.0