While this application can be used in multiple ways, the primary purpose of this application is to run allow the computer to learn how to play snake on its own. The user can also choose to play snake on their own however that is a side feature that was implemented for initial testing purposes.
There are 2 ways in which someone can run the application where the computer learns to play snake. First, regardless of which way they want to run the program, the user should install tensorflow, tf-agents, and pygame to their computer and create a virtual environment which can be used to run tensorflow. Then the user can either run the Main.py file or they can run load the Computer.py file. If they run the Main.py file, they should then click the "Computer" button which will show up in the top right of their screen. If the Computer.py file was run, the user should then run the "simulate" method which will start the simulation.
As the simulation runs, it will tell the user how many steps of data have been collected by the agent and how well the agent is doing in evaluation games which are run every time the agent collects 5000 steps of data. The simulation runs for a total of 30000 steps, and the agent evaluates a total of 35 games. On average the simulation takes around 5 to 10 minutes to run in total and after around 10000 steps the snake will average around 30 total fruit when evaluating afterwards. The highest score the snake ever achieved in our runs was 70 fruit and this run is saved in the simulation file which is in our latest commit. After the simulation finishes running, a matlab plot will be displayed of the evaluation results of our simulation.
All of the runs by the agent will be saved to the file in the simulations folder. For the user to actually see what the computer did during these runs, they can run the Persistence.py file which will display a GUI which will show the computer playing the game in a very sped up version.
