assert (0.8 <= np.mean(total_rewards) <= 0.95)
print('Well done!')
if __name__ == '__main__':
visualize = True
mdp = FrozenLakeEnv(map_name='8x8', slip_chance=0.1)
mdp.render()
gamma = 0.9
num_iter = 100
min_difference = 1e-5
# Play in Frozen Lake Env
state_values = {state: 0
for state in mdp.get_all_states()
} # Initialize state_values
# Run value iteration algo!
state_values, _ = rl_value_iteration(mdp, gamma, num_iter, min_difference,
state_values)
# See how our agent performs - e.g. render what is going on when agent choose `optimal` value
s = mdp.reset()
mdp.render()
rewards = [] # Save all rewards to see mean reward.
for _ in range(num_iter):
action = get_optimal_action(mdp, state_values, s, gamma)
new_state, reward, done, _ = mdp.step(action)
rewards += [reward]
else:
assert (0.8 <= np.mean(total_rewards) <= 0.95)
print('Well done!')
if __name__ == '__main__':
visualize = False
mdp = FrozenLakeEnv(map_name='8x8', slip_chance=0.1)
mdp.render()
gamma = 0.9
num_iter = 100
min_difference = 1e-5
# Play in Frozen Lake Env
state_values = mdp.get_all_states() # Initialize state_values
# Run value iteration algo!
init_values = {s: 0 for s in state_values}
state_values, _ = rl_value_iteration(mdp, gamma, num_iter, min_difference,
init_values)
# See how our agent performs - e.g. render what is going on when agent choose `optimal` value
s = mdp.reset()
mdp.render()
rewards = [] # Save all rewards to see mean reward.
# Your code here!
for t in range(num_iter):
s, r, done, _ = mdp.step(
get_optimal_action(mdp, state_values, s, gamma))
if __name__ == '__main__':
visualize = True
mdp = FrozenLakeEnv(map_name='8x8', slip_chance=0.1)
mdp.render()
gamma = 0.9
num_iter = 100
min_difference = 1e-5
visualize = True
mdp = FrozenLakeEnv(map_name='8x8', slip_chance=0.1)
# Play in Frozen Lake Env
state_values = {s: 0
for s in mdp.get_all_states()} # Initialize state_values
# Run value iteration algo!
state_values, _ = rl_value_iteration(mdp, gamma, num_iter, min_difference,
state_values)
if visualize:
draw_policy(mdp,
state_values,
filename='frozen_lake_visualization.png')
# Let's see how it is improving in time.
visualize_step_by_step(mdp, gamma, num_iter, min_difference)
# Express test!
rewards = mass_gaming(mdp, gamma, num_iter, 1000,
100) # Save all rewards to see mean reward.
state_values = value_iteration(mdp)
s = mdp.reset()
mdp.render()
for t in range(100):
a = get_optimal_action(mdp, state_values, s, gamma)
print(a, end='\n\n')
s, r, done, _ = mdp.step(a)
mdp.render()
if done:
break
#use draw_policy(mdp, state_values):
state_values = {s: 0 for s in mdp.get_all_states()}
for i in range(10):
print("after iteration %i" % i)
state_values = value_iteration(mdp, state_values, num_iter=1)
draw_policy(mdp, state_values)
# please ignore iter 0 at each step
from IPython.display import clear_output
from time import sleep
mdp = FrozenLakeEnv(map_name='8x8', slip_chance=0.1)
state_values = {s: 0 for s in mdp.get_all_states()}
for i in range(30):
clear_output(True)
print("after iteration %i" % i)