The Simulation

How the Reproduction Works

"Splicing" (Or Crossing Over)

The Fitness Evaluation Function

The Neural Network

Running the Simulation

You can use make run or run the binary directly with ./game to run the simulation. After you launch it, the program will ask you for some parameters:

• Tamanho do Grid: defines how many rows and columns the game's internal grid has. Each agent in the game occupies a single square inside the game's grid.

• Tipo de Cruzamento: defines the breeding type (what type of reproduction will be used). It can be:

  • "0" Elitist: the best agent in the current generation breeds with every other agent;

  • "1" N_Best: the "N" best agents of the current generation breeds between themselves (the value of N is decided internally by the program based on how many agents are in a generation); or

  • "2" Assexual: there is no breeding, each agent creates a mutated copy of themselves for the new generation.

• Número de Indivíduos + Quantidade de Frames + Número de Gerações: in this part you will have to input 3 values separated by spaces:

  • The amount of agents in each generation of the simulation (if you chose the N_Best type of reproduction, the number of agents needs to be at least 36);

  • The amount of frames until a generation ends (if all the agents die before the generation reaches this amount of frames the generation will end anyway); and

  • How many generations you want to simulate.

• Número de Raios + Quantos Frames até Matar: in this part you will input 2 values separated by spaces:

  • The amount of beans you want being generated at the same time; and

  • The amount of frames until it actually kills an agent (if the agent is under it).

• Modo: this is the type of "splicing" (crossing-over), it can be:

  • "0" Average;

  • "1" Splicing_half; or

  • "2" Splicing_rand.

During the simulation you can use the "F" key in your keyboard to make the simulation run faster (up to 8x faster), or use the "S" key to make the simulation run slower (down to 0.1x slower).

You can also use the "Q" key to quit the simulation, this will make so that the progress made is saved to an csv file, this way you won't lose all the simulation data like you would if you just killed the program.

**Be aware that if you quit the simulation, the data from the generation you were at the moment until the generation the simulation was supposed to run may be filled with garbage and may not make sense. Only the data from the previous generations are guaranteed to make sense.

Suggested Values to Test

I suggest you test the following values:

• 100x100 grid: 100;
• N_BEST: 1;
• Agents: 400, Frames: 100000, Generations: 100;
• Beans: 100, Frames: 10;
• SPLICING_RAND: 2.

Plotting the Results

Use the make plot command to run the plot_csv.py script, or run it directly by typing the python3 plot_csv.py command. If you decide to run the script directly and you are on a Wayland system, you have to remember to set the QT_QPA_PLATFORM=wayland environment variable beforehand.

**You need the matplotlib and PyQt5 python packages installed to be able to use the plot functionality.

Building the Application

The building process is pretty straight forward. First you'll need to clone the repository with:

git clone https://github.com/m-almeida0/SquaresAndBeans.git

cd SquaresAndBeans/

Next you'll need to download some dependencies. In the following sections we have the commands you can use on two popular GNU/Linux distributions (if you're using Windows or MacOS, well... good luck :D. You can use WSL on Windows if you want though).

Fedora

sudo dnf install make g++ freeglut-devel python3-matplotlib-qt5

Ubuntu

sudo apt install make g++ freeglut3-dev python3-matplotlib python3-pyqt5

After having installed the needed dependencies, you can run make to build the application. In the next section, each possible make command is listed and explained.

Compiling

• make clean: deletes all previous compilation artifacts;

• make game: generate squares.o and NNetworks.o object files if it does not exists, or if it is outdated, and then compiles the game binary if it also does not exists or is outdated;

• make/make all: does clean and then game;

• make run: does game and then runs the simulation.