PyTorch 1.x Reinforcement Learning Cookbook

How to do it...

Let's simulate the Cliff Walking environment by performing the following steps:

1. We import the Gym library and create an instance of the Cliff Walking environment :

>>> import gym>>> env = gym.make("CliffWalking-v0")>>> n_state = env.observation_space.n>>> print(n_state)48>>> n_action = env.action_space.n>>> print(n_action)4

2. Then, we reset the environment:

>>> env.reset() 0

The agent starts with state 36 as the bottom-left tile.

3. Then, we render the environment:

>>> env.render()

4. Let's now make a down movement regardless, even though it is not walkable:

>>> new_state, reward, is_done, info = env.step(2)>>> env.render() o  o  o  o  o  o  o  o  o  o  o  o o  o  o  o  o  o  o  o  o  o  o  o o  o  o  o  o  o  o  o  o  o  o  o x  C  C  C  C  C  C  C  C  C  C  T

The ...