Art057_Python_005.py

# SARSA parameters
alpha = 0.1   # learning rate
gamma = 0.95  # discount factor

# Training parameters
n_episodes = 100000  # number of episodes to use for training
n_max_steps = 100   # maximum number of steps per episode

# Exploration / Exploitation parameters
start_epsilon = 1.0  # start training by selecting purely random actions
min_epsilon = 0.0   # the lowest epsilon allowed to decay to
decay_rate = 0.00001   # epsilon will gradually decay so we do less exploring and more exploiting as Q-function improves