

In [1]:

    
import gym



In [2]:

    
env = gym.make('FrozenLake-v0') # initialize the Frozen Lake environment









    



[2017-10-26 17:15:45,498] Making new env: FrozenLake-v0



In [3]:

    
env.render() # check out what the initial state of the environment looks like









    



\e[1mSFFF
FHFH
FFFH
HFFG







    Out[3]:





<ipykernel.iostream.OutStream at 0x7f106d4527d0>

What we see here is a layout of a 'frozen lake'. The agent can move in this world. Some tiles of the lake are walkable (frozen, 'F'), and others lead to the agent falling into the water (hole, 'H'). Additionally, the movement direction of the agent is uncertain and only partially depends on the chosen direction. The agent is rewarded for finding a walkable path to a goal tile. The available actions are walking one step in any of the four directions (up, down, left and right). The current state is highlighted in red.



In [4]:

    
n_states = env.observation_space.n # number of states
n_actions = env.action_space.n # number of actions
print("This environment has %s states and %s actions." % (n_states, n_actions))









    



This environment has 16 states and 4 actions.

Let us now try to move around in this environment a little bit.



In [5]:

    
s0 = env.reset()
print("Initially the agent is in state %s." % s0)

s1 = env.step(env.action_space.sample()) # take a random step
print('''
    The agent is now in state %s and has recieved a reward %s. 
    Episode done? %s.
    Additional information (transition probability of landing in this state): %s.
    '''
    % s1)
env.render() # render current state









    



Initially the agent is in state 0.

    The agent is now in state 0 and has recieved a reward 0.0. 
    Episode done? False.
    Additional information (transition probability of landing in this state): {'prob': 0.3333333333333333}.
    
SFFF
FHFH
FFFH
HFFG
  (Left)






    Out[5]:





<ipykernel.iostream.OutStream at 0x7f44924d5750>