Flappy Bird game which has functionality for machine learning through Supervised Learning as well as Reinforcement Learning. The agent makes decisions using an Artificial Neural Network. The neural network should be initially trained by simply playing the game properly until a reasonable amount of data has been collected (≈2000 datapoints should work, use the p key to print data collection progress). Once a neural network has been trained, the agent can play autonomously by using the a key, and can further speed up data collection by pressing the r key.
It is highly recommended that you use conda for simply installing all of the Python requirements to a virtual environment. Conda installation instructions can be found here. Once you have installed conda, follow the instructions below.
- From the root directory, create the conda environment using the bash command:
conda env create -f environment.yml -n <ENVIRONMENT NAME> - Activate the conda environment:
source activate <ENVIRONMENT NAME> - Start the game:
python flappy.pyThe agent will gather data appropriately from the player, then the following commands can be used to execute different procedures:
p: Print the current state of the gathered training data to the terminal.
t: Train the neural network on the collected data so far.
a: Switch autonomous mode on/off.
r: Switch reinforcement learning speed-up on/off.