def test_agent_planner():
rs_pomcp = POMCP(rs_type, horizon=5)
agent.planner = rs_pomcp
while agent.planning_horizon >= 0:
obs, reward = agent.execute_action
print(f'Action {agent.actions[len(agent.actions)-1].name}, observation: {obs.name} and reward {reward}')
print(f'Total reward accumulated: {agent.agent_type.frame.oc.get_current_reward()}')
def best_response(self):
tom_zero_worker_pomcp = POMCP(
self.worker_type,
horizon=self.planning_horizon,
exploration_bonus=self.worker_agent.exploration_bonus)
self.worker_agent.planner = tom_zero_worker_pomcp
action, q_value = self.worker_agent.compute_optimal_policy
return action, q_value
def test_manager_planner():
tom_zero_manager_pomcp = POMCP(manager_type, horizon=5)
manager.planner = tom_zero_manager_pomcp
while manager.planning_horizon >= 0:
obs, reward = manager.execute_action
print(
f'Action {manager.actions[len(manager.actions)-1].name}, observation: {obs.name} and reward {reward}'
)
if obs.value:
break
print(
f'Total reward accumulated: {manager.agent_type.frame.oc.get_current_reward()}'
)
belief_dict = {
is_: belief_e
for belief_e, is_ in zip(initial_belief, is_list_cur_level)
} # check might be source of bug when intial belief is <TL, TR>
for frame_cur_level in frames_cur_level:
#cur_model = Model.create_or_get_model(belief_dict, frame_cur_level)
if cur_level == 0:
environment = Environment(
physical_states, frame_cur_level.observations,
frame_cur_level.actions,
frame_cur_level.observation_table,
frame_cur_level.reward_table,
frame_cur_level.transition_table)
solver = POMCP(environment)
else:
environment = MultiAgentEnvironment(
physical_states, frame_cur_level.observations,
frame_cur_level.actions,
frame_cur_level.observation_table,
frame_cur_level.reward_table,
frame_cur_level.transition_table)
solver = IPOMCP(environment)
cur_model = Model(belief_dict, "", frame_cur_level, solver)
model_list_cur_level.append(cur_model)
if cur_level == 0:
model_list_0_level = model_list_cur_level
model_list_prev_level = model_list_cur_level
observations = ['GL', 'GR']
sensor_accuracy = 0.85
reward_listen = -1
reward_gold = 10
reward_tiger = -100
noisy = True
beta = None
gamma = 0.9
epsilon = 0.01
horizon = 3
environment = TigerProblem(states, observations, actions, sensor_accuracy,
reward_listen, reward_gold, reward_tiger, noisy,
beta)
pomcp = POMCP(environment, gamma, epsilon, horizon)
history = ''
initial_belief = {0: 0.5, 1: 0.5}
#testing first best action
#root_node, best_action=pomcp.search(history, initial_belief, i = 10000)
#print("Root Node: "+ str(root_node))
#print("Best Action: "+ str(best_action))
#sys.exit()
#finite horizon, averaged across multiple episodes
'''
num_episodes = 100
rewards = []
#run for all time-steps
def test_pomcp():
root_node = ObservationNode(None, '', '')
tiger_pomcp = POMCP(tiger_type, horizon=3)
br_node, br_value = tiger_pomcp.search(root_node)
obs, reward = tiger_problem.pomdp_step(br_node)