def run_value_iteration(transition_probabilities=np.ones((5, 2)) * 0.5,
rewards=[1, 0, 0, 0, 10],
horizon=10,
gamma=0.9):
v = np.zeros(5)
env = MarsRover(transition_probabilities, rewards, horizon)
done = False
state = env.reset()
i = 0
while not done:
i += 1
print(f"This is step {i}")
r1 = (rewards[state - 1] + gamma * v[state - 1])
r2 = (rewards[state + 1] + gamma * v[state + 1])
action = np.argmax([r1, r2])
if r1 == r2:
action = np.random.randint(2)
new_state, reward, done = env.step(action)
v, converged = update_value_function(v, state, new_state, reward)
if converged:
break
state = new_state
final_reward = evaluate_agent(v, env)
print(
f"Your agent achieved a final accumulated reward of {final_reward} after {i} update steps."
)
return v, i, final_reward
def run_policy_iteration(transition_probabilities=np.ones((5, 2)),
rewards=[1, 0, 0, 0, 10],
horizon=10):
env = MarsRover(transition_probabilities, rewards, horizon)
qs = np.zeros((5, 2))
pi = np.random.randint(0, 2, size=5)
done = False
state = env.reset()
i = 0
while not done:
i += 1
print(f"This is step {i}")
action = pi[state]
new_state, reward, done = env.step(action)
qs, pi, converged = update_policy(qs, pi, state, new_state, action,
reward)
if converged:
break
state = new_state
final_reward = evaluate_policy(pi, env)
print(
f"Your policy achieved a final accumulated reward of {final_reward} after {i} update steps."
)
return pi, i, final_reward
def run_value_iteration(transition_probabilities=np.ones((5, 2)) * 0.5,
rewards=[1, 0, 0, 0, 10],
horizon=10):
env = MarsRover(transition_probabilities, rewards, horizon)
done = False
state = env.reset()
i = 0
while not done:
i += 1
print(f"This is step {i}")
new_state, reward, done = env.step(action)
final_reward = evaluate_agent(v, env)
print(
f"Your agent achieved a final accumulated reward of {final_reward} after {i} update steps."
)
return v, i, final_reward
def evaluate_policy_dp(pi=np.random.randint(2, size=5),
transition_probabilities=np.ones((5, 2)),
rewards=[1, 0, 0, 0, 10]):
env = MarsRover(transition_probabilities=transition_probabilities,
rewards=rewards)
i = 0
while True:
i += 1
print(f"Policy was evaluated in {i} steps with resulting v {v}")
return v, i