experiment_params.yaml

batch_size: 32 # How many experiences to use for each training step.
update_freq: 2 # How often to perform a training step.
gamma: .99  # Discount factor on the target Q-values
startE: 1  # Starting chance of random action
endE: 0.1  # Final chance of random action
num_episodes: 100000 # How many episodes of game environment to train network with.
load_model: False  # Whether to load a saved model.
h_size: 256 # The size of the final convolutional layer before splitting it into Advantage and Value streams.
tau: 0.001  # Rate to update target network toward primary network
max_hr: 10