Python implementation of the game Flappy Bird, played by a neural network
In the script there are some areas which need to be filled in because they are specific to my file locations, however the script will not run unless they are changed to your specific file locations.
The following list can found in the script on line 13:
BIRD_IMGS = [pygame.transform.scale2x(pygame.image.load(os.path.join(r"IMAGE_FOLDER",r"BIRD_1"))),
pygame.transform.scale2x(pygame.image.load(os.path.join(r"IMAGE_FOLDER", r"BIRD_2"))),
pygame.transform.scale2x(pygame.image.load(os.path.join(r"IMAGE_FOLDER", r"BIRD_3")))
]
The image folder is called imgs, to insert the location of the image folder you must locate to it and press the shift key on your keyboard and then right click. Once that is done select copy as path and return back to the script. Once you are in the script delete the part that says IMAGE_FOLDER and replace it with the folder path that you just copied. To paste the path press ctrl and v on your keyboard or alternitivly right click and select paste. BIRD_1, BIRD_2 and BIRD_3 are all located in the imgs folder. do what you did with the imgs folder before to BIRD_1, BIRD_2 and BIRD_3.
The following variable can be found in the script on line 14.
PIPE_IMG = pygame.transform.scale2x(pygame.image.load(os.path.join(r"IMAGE_FOLDER", r"PIPE_IMAGE")))
The image folder is called imgs, to insert the location of the image folder you must locate to it and press the shift key on your keyboard and then right click. Once that is done select copy as path and return back to the script. Once you are in the script delete the part that says IMAGE_FOLDER and replace it with the folder path that you just copied. To paste the path press ctrl and v on your keyboard or alternitivly right click and select paste. PIPE_IMAGE is located in the imgs folder. do what you did with the imgs folder before to PIPE_IMAGE.
The following variable can be found in the script on line 15
BASE_IMG = pygame.transform.scale2x(pygame.image.load(os.path.join(r"IMAGE_FOLDER", r"BASE_IMAGE")))
The image folder is called imgs, to insert the location of the image folder you must locate to it and press the shift key on your keyboard and then right click. Once that is done select copy as path and return back to the script. Once you are in the script delete the part that says IMAGE_FOLDER and replace it with the folder path that you just copied. To paste the path press ctrl and v on your keyboard or alternitivly right click and select paste. BASE_IMAGE is located in the imgs folder. do what you did with the imgs folder before to BASE_IMAGE.
The following variable can be found in the script on line 16
BG_IMG = pygame.transform.scale2x(pygame.image.load(os.path.join(r"IMAGE_FOLDER", r"BG_IMAGE")))
The image folder is called imgs, to insert the location of the image folder you must locate to it and press the shift key on your keyboard and then right click. Once that is done select copy as path and return back to the script. Once you are in the script delete the part that says IMAGE_FOLDER and replace it with the folder path that you just copied. To paste the path press ctrl and v on your keyboard or alternitivly right click and select paste. BG_IMAGE is located in the imgs folder. do what you did with the imgs folder before to BG_IMAGE.
The following variable can be found in the script on line 270
config_path = os.path.join(local_dir, r"CONFIG_FILE")
The config file is called neatCONFIG.txt. Replace the CONFIG_FILE in the script with the path of the neatCONFIG.txt
Training time can vary from 10 minutes to 10 microseconds. If you want to see the process of its training set the pop_size to 5.
However if you do not want to see how it works and proggresivly gets faster set the pop_size to 100.
[NEAT]
fitness_criterion = min
fitness_threshold = 500
pop_size = 5
reset_on_extinction = False
Higher the pop_size the faster the training.
pop_size and other parameters can be changed withing the neatCONFIG.txt file.
A pop_size of 5 takes roughly 100 generations until it is a good AI. A pop_size of 100 takes roughly 1 generation until it is a good AI.
The AI works by using a simple neural network with 3 inputs (y_movement, height_of_top_pipe, height_of_bottom_pipe) it could theoreticly work with just 2 input but that would require more computer power since the neural network would need to calculate the distance between each pipe therefore increasing training times. The neurel network only has 1 output, should the bird go up or down. The bird only needs to move up or down becasue it does not move forward instead the environment moves backward. The problem with this is that because the neural network so simple it can not play the game without dying in the first few seconds. To conquer this what the script gradually adds or changes layers and neurons it also adjustes the weights and biasses until you have a neurel network that has been perfectly made and with enough training would never loose at the game flappy bird.
The game and easily be adjusted to be played by a human instead of a neural network
To do this you must make a function called anything you want and comment out or delete all code that relates to the AI. In the function you must create an if statement that when space is clicked call the jump function.
This may help:
import keyboard
if keyboard.is_pressed('a'):
print('a key has ben pressed')
Try and figure out if you can get it to work when space is pressed instead of A
Special thanks to Tech With Tim