This is a basic working qlearning example from the mountain-car environment in gym
Here are some of the paramaters you may wish to tinker around wth.
- EPISODES = 1 + 2000
- Number of episodes for the environment simulation, change the last number not the 1. The higher the number the more learning, and the better the agent gets.
- SHOW_EVERY = 500
- When to show the environment render
- SAVE_EVERY = 20
- Important for the graphs, the lower the number the significantly lower the whole simulation is
- RENDER = True
- Set to False if you don't want to render the environment. This is useful if you want to run this in jupyter notebook or Ipython.
- LEARNING_RATE = 0.1
- DISCOUNT = 0.95
- EPSILON = 0.1
- EPSILON_START = 0
- EPSILON_END = EPISODES
- EPSILON_DECAY = EPSILON / (EPSILON_END - EPSILON_START)
- Q_SIZE = [20, 20]
- The size of the Q table, the bigger it is the slower the whole process becomes, the smaller it is the lower the precision.