def get_optimal_policy(self):
for state_index, action_distribution in self.policy.policy_table.items(
):
q_values = {}
for action_index, _ in action_distribution.items():
q_values[action_index] = self._get_value_of_action(
state_index, action_index)
greedy_distibution = create_distribution_greedily()(q_values)
self.policy.policy_table[state_index] = greedy_distibution
return self.policy
def __init__(self,
q_table,
table_policy,
env,
steps,
statistics,
episodes,
step_size=0.1,
discount=1.0):
self.q_table = q_table
self.policy = table_policy
self.env = env
self.episodes = episodes
self.step_size = step_size
self.discount = discount
self.create_distribution_greedily = create_distribution_greedily()
self.steps = steps
self.statistics = statistics
def __init__(self, behavior_policy, target_policy, critic):
self.target_policy = target_policy
self.behavior_policy = behavior_policy
self.critic = critic
self.create_distribution_greedily = create_distribution_greedily()
def __init__(self, policy, critic):
self.policy = policy
self.critic = critic
self.create_distribution_greedily = create_distribution_greedily()
def __init__(self, policy,critic,epsilon=0.1):
self.policy = policy
self.critic = critic
self.create_distribution_epsilon_greedily = create_distribution_epsilon_greedily(epsilon)
self.create_distribution_greedily = create_distribution_greedily()