The program displays 48 boxes. In each one of them is a blue and a green circle. The green circle is controlled by a neural network (separate for each one) and the blue one is static “food”. Every time a green circle touches a blue circle, the network responsible is rewarded with a point, and the blue circle is relocated. After a chosen time period 10 networks with the highest number of points are chosen, and copied with small mutations. Thus, a new generation of 48 networks is created. The process repeats.
After each generation, the terminal outputs the score achieved by the best network within the generation. Example:
Generation number: 112 best score: 7
Generation number: 113 best score: 10
Generation number: 114 best score: 6
Generation number: 115 best score: 9
Generation number: 116 best score: 15
Generation number: 117 best score: 18
Generation number: 118 best score: 20
Generation number: 119 best score: 17
Generation number: 120 best score: 21
In order to perform compilation, a sfml library is needed, which can be downloaded from this site:
https://www.sfml-dev.org/download/sfml/2.5.1/
On linux the command for installation is sudo apt-get install libsfml-dev.
For compilation on linux the script compile.sh can be used ./compile.sh
By clicking n you can cycle between 60fps, 120fps, unlimited fps and no display. Turning off the display increases simulation speed dramatically.
Pauses the simulation, while pausing the simulation, networks can also be saved/loaded to/from the file.
Saves network marked with cursor to a file. The name of the file to which the network is saved needs to be written in the terminal.
Loads network marked with cursor to a file. The name of the file from which the network is loaded needs to be written in the terminal.
The algorithm is very time and memory efficient. It doesn't use any allocation on the heap as all data structures needed in order to build a population of neural networks are written using templates and their size is known during compilation.