def run(parameter):
"""
The entry function of this code.
Args:
parameter: the super-parameter
"""
print(json.dumps(parameter, indent=2))
time.sleep(2)
slot_set = pickle.load(file=open(parameter["slot_set"], "rb"))
action_set = pickle.load(file=open(parameter["action_set"], "rb"))
disease_symptom = pickle.load(file=open(parameter["disease_symptom"], "rb"))
steward = RunningSteward(parameter=parameter,checkpoint_path=parameter["checkpoint_path"])
print('action_set', action_set)
warm_start = parameter.get("warm_start")
warm_start_epoch_number = parameter.get("warm_start_epoch_number")
train_mode = parameter.get("train_mode")
agent_id = parameter.get("agent_id")
simulate_epoch_number = parameter.get("simulate_epoch_number")
# Warm start.
if warm_start == True and train_mode == True:
print("warm starting...")
agent = AgentRule(action_set=action_set,slot_set=slot_set,disease_symptom=disease_symptom,parameter=parameter)
steward.dialogue_manager.set_agent(agent=agent)
steward.warm_start(epoch_number=warm_start_epoch_number)
# exit()
if agent_id.lower() == 'agentdqn':
agent = AgentDQN(action_set=action_set,slot_set=slot_set,disease_symptom=disease_symptom,parameter=parameter)
elif agent_id.lower() == 'agentrandom':
agent = AgentRandom(action_set=action_set,slot_set=slot_set,disease_symptom=disease_symptom,parameter=parameter)
elif agent_id.lower() == 'agentrule':
agent = AgentRule(action_set=action_set,slot_set=slot_set,disease_symptom=disease_symptom,parameter=parameter)
elif agent_id.lower() == 'agenthrl':
agent = AgentHRL(action_set=action_set, slot_set=slot_set, disease_symptom=disease_symptom, parameter=parameter)
elif agent_id.lower() == 'agentwithgoaljoint':
agent = AgentWithGoalJoint(action_set=action_set, slot_set=slot_set, disease_symptom=disease_symptom, parameter=parameter)
elif agent_id.lower() == 'agentwithgoal':
agent = AgentWithGoal(action_set=action_set, slot_set=slot_set, disease_symptom=disease_symptom, parameter=parameter)
elif agent_id.lower() == 'agentwithgoal2':
agent = AgentWithGoal2(action_set=action_set, slot_set=slot_set, disease_symptom=disease_symptom, parameter=parameter)
elif agent_id.lower() == 'agentwithgoal3':
from src.dialogue_system.agent.agent_with_goal_3 import AgentWithGoal as AgentWithGoal3
agent = AgentWithGoal3(action_set=action_set, slot_set=slot_set, disease_symptom=disease_symptom,
parameter=parameter)
else:
raise ValueError('Agent id should be one of [AgentRule, AgentDQN, AgentRandom, AgentHRL, AgentWithGoal, AgentWithGoal2, AgentWithGoalJoint].')
steward.dialogue_manager.set_agent(agent=agent)
if train_mode is True: # Train
steward.simulate(epoch_number=simulate_epoch_number, train_mode=train_mode)
else: # test
for index in range(simulate_epoch_number):
res = steward.evaluate_model(dataset='test', index=index)
return res
def __init__(self, parameter, checkpoint_path):
self.epoch_size = parameter.get("simulation_size", 100)
self.parameter = parameter
self.checkpoint_path = checkpoint_path
self.learning_curve = {}
slot_set = pickle.load(file=open(parameter["slot_set"], "rb"))
action_set = pickle.load(file=open(parameter["action_set"], "rb"))
goal_set = pickle.load(file=open(parameter["goal_set"], "rb"))
disease_symptom = pickle.load(
file=open(parameter["disease_symptom"], "rb"))
user = User(goal_set=goal_set,
disease_syptom=disease_symptom,
parameter=parameter)
agent = AgentRule(action_set=action_set,
slot_set=slot_set,
disease_symptom=disease_symptom,
parameter=parameter)
self.dialogue_manager = DialogueManager(user=user,
agent=agent,
parameter=parameter)
self.best_result = {
"success_rate": 0.0,
"average_reward": 0.0,
"average_turn": 0,
"average_wrong_disease": 10
}
def run():
slot_set = pickle.load(file=open(parameter["slot_set"], "rb"))
action_set = pickle.load(file=open(parameter["action_set"], "rb"))
disease_symptom = pickle.load(
file=open(parameter["disease_symptom"], "rb"))
steward = RunningSteward(parameter=parameter,
checkpoint_path=checkpoint_path)
warm_start = parameter.get("warm_start")
warm_start_epoch_number = parameter.get("warm_start_epoch_number")
train_mode = parameter.get("train_mode")
agent_id = parameter.get("agent_id")
simulate_epoch_number = parameter.get("simulate_epoch_number")
# Warm start.
if warm_start == 1 and train_mode == 1:
print("warm starting...")
agent = AgentRule(action_set=action_set,
slot_set=slot_set,
disease_symptom=disease_symptom,
parameter=parameter)
steward.warm_start(agent=agent, epoch_number=warm_start_epoch_number)
# exit()
if agent_id == 1:
agent = AgentDQN(action_set=action_set,
slot_set=slot_set,
disease_symptom=disease_symptom,
parameter=parameter)
elif agent_id == 2:
agent = AgentActorCritic(action_set=action_set,
slot_set=slot_set,
disease_symptom=disease_symptom,
parameter=parameter)
elif agent_id == 3:
agent = AgentRandom(action_set=action_set,
slot_set=slot_set,
disease_symptom=disease_symptom,
parameter=parameter)
elif agent_id == 0:
agent = AgentRule(action_set=action_set,
slot_set=slot_set,
disease_symptom=disease_symptom,
parameter=parameter)
steward.simulate(agent=agent,
epoch_number=simulate_epoch_number,
train_mode=train_mode)
def __init__(self, parameter, checkpoint_path):
self.epoch_size = parameter.get("simulation_size", 100)
self.parameter = parameter
self.checkpoint_path = checkpoint_path
self.learning_curve = {}
slot_set = pickle.load(file=open(parameter["slot_set"], "rb"))
action_set = pickle.load(file=open(parameter["action_set"], "rb"))
goal_set = pickle.load(file=open(parameter["goal_set"], "rb"))
disease_symptom = pickle.load(
file=open(parameter["disease_symptom"], "rb"))
user = User(goal_set=goal_set,
disease_syptom=disease_symptom,
parameter=parameter)
agent = AgentRule(action_set=action_set,
slot_set=slot_set,
disease_symptom=disease_symptom,
parameter=parameter)
if parameter.get("use_all_labels"):
self.dialogue_manager = DialogueManager_HRL(user=user,
agent=agent,
parameter=parameter)
else:
self.dialogue_manager = DialogueManager(user=user,
agent=agent,
parameter=parameter)
if self.parameter.get("disease_as_action") == False:
if self.parameter.get("classifier_type") == "machine_learning":
self.dialogue_manager.train_ml_classifier()
print(
"############ the machine learning model is training ###########"
)
elif self.parameter.get("classifier_type") == "deep_learning":
self.dialogue_manager.build_deep_learning_classifier()
else:
raise ValueError(
"the classifier type is not among machine_learning and deep_learning"
)
self.best_result = {
"success_rate": 0.0,
"average_reward": 0.0,
"average_turn": 0,
"average_wrong_disease": 10
}
def run(parameter):
"""
The entry function of this code.
Args:
parameter: the super-parameter
"""
agent_id = parameter.get("agent_id")
dqn_id = parameter.get("dqn_id")
disease_number = parameter.get("disease_number")
max_turn = parameter.get("max_turn")
if agent_id == 1:
checkpoint_path = "./../model/dqn/checkpoint/checkpoint_d" + str(
disease_number) + "_" + str(agent_id) + "_dqn" + str(
dqn_id) + "_T" + str(max_turn) + "/"
else:
checkpoint_path = "./../model/dqn/checkpoint/checkpoint_d" + str(
disease_number) + "_" + str(agent_id) + "_T" + str(max_turn) + "/"
print(json.dumps(parameter, indent=2))
time.sleep(1)
slot_set = pickle.load(file=open(parameter["slot_set"], "rb"))
action_set = pickle.load(file=open(parameter["action_set"], "rb"))
disease_symptom = pickle.load(
file=open(parameter["disease_symptom"], "rb"))
steward = RunningSteward(parameter=parameter,
checkpoint_path=checkpoint_path)
warm_start = parameter.get("warm_start")
warm_start_epoch_number = parameter.get("warm_start_epoch_number")
train_mode = parameter.get("train_mode")
agent_id = parameter.get("agent_id")
simulate_epoch_number = parameter.get("simulate_epoch_number")
# Warm start.
if warm_start == True and train_mode == True:
print("warm starting...")
agent = AgentRule(action_set=action_set,
slot_set=slot_set,
disease_symptom=disease_symptom,
parameter=parameter)
steward.warm_start(agent=agent, epoch_number=warm_start_epoch_number)
# exit()
if agent_id.lower() == 'agentdqn':
agent = AgentDQN(action_set=action_set,
slot_set=slot_set,
disease_symptom=disease_symptom,
parameter=parameter)
elif agent_id.lower() == 'agentactorcritic':
agent = AgentActorCritic(action_set=action_set,
slot_set=slot_set,
disease_symptom=disease_symptom,
parameter=parameter)
elif agent_id.lower() == 'agentrandom':
agent = AgentRandom(action_set=action_set,
slot_set=slot_set,
disease_symptom=disease_symptom,
parameter=parameter)
elif agent_id.lower() == 'agentrule':
agent = AgentRule(action_set=action_set,
slot_set=slot_set,
disease_symptom=disease_symptom,
parameter=parameter)
else:
raise ValueError(
'Agent id should be one of [AgentRule, AgentDQN, AgentActorCritic, AgentRandom].'
)
steward.simulate(agent=agent,
epoch_number=simulate_epoch_number,
train_mode=train_mode)
def run(parameter, main_checkpoint_path):
agent_id = parameter.get("agent_id")
dqn_id = parameter.get("dqn_id")
disease_number = parameter.get("disease_number")
lr = parameter.get("dqn_learning_rate")
reward_for_success = parameter.get("reward_for_success")
reward_for_fail = parameter.get("reward_for_fail")
reward_for_not_come_yet = parameter.get("reward_for_not_come_yet")
reward_for_inform_right_symptom = parameter.get("reward_for_inform_right_symptom")
max_turn = parameter.get("max_turn")
minus_left_slots = parameter.get("minus_left_slots")
gamma = parameter.get("gamma")
epsilon = parameter.get("epsilon")
run_id = parameter.get('run_id')
if agent_id == 1:
checkpoint_path = main_checkpoint_path + "checkpoint_d" + str(disease_number) + "_agt" + str(agent_id) + \
"_dqn" + str(dqn_id) + "_T" + str(max_turn) + "_lr" + str(lr) + "_RFS" + str(reward_for_success) +\
"_RFF" + str(reward_for_fail) + "_RFNCY" + str(reward_for_not_come_yet) + "_RFIRS" + str(reward_for_inform_right_symptom) +\
"_mls" + str(minus_left_slots) + "_gamma" + str(gamma) + "_epsilon" + str(epsilon) + "_RID" + str(run_id) + "/"
else:
checkpoint_path = main_checkpoint_path + "checkpoint_d" + str(disease_number) + "_agt" + str(agent_id) + \
"_T" + str(max_turn) + "_lr" + str(lr) + "_RFS" + str(reward_for_success) + \
"_RFF" + str(reward_for_fail) + "_RFNCY" + str(reward_for_not_come_yet) + "_RFIRS" + str(reward_for_inform_right_symptom) + \
"_mls" + str(minus_left_slots) + "_gamma" + str(gamma) + "_epsilon" + str(epsilon) + "_RID" + str(run_id) + "/"
print(json.dumps(parameter, indent=2))
time.sleep(8)
slot_set = pickle.load(file=open(parameter["slot_set"], "rb"))
action_set = pickle.load(file=open(parameter["action_set"], "rb"))
disease_symptom = pickle.load(file=open(parameter["disease_symptom"], "rb"))
steward = RunningSteward(parameter=parameter, checkpoint_path=checkpoint_path)
warm_start = parameter.get("warm_start")
warm_start_epoch_number = parameter.get("warm_start_epoch_number")
train_mode = parameter.get("train_mode")
agent_id = parameter.get("agent_id")
simulate_epoch_number = parameter.get("simulate_epoch_number")
# Warm start.
if warm_start == 1 and train_mode == 1:
print("warm starting...")
agent = AgentRule(action_set=action_set, slot_set=slot_set, disease_symptom=disease_symptom,
parameter=parameter)
steward.warm_start(agent=agent, epoch_number=warm_start_epoch_number)
# exit()
if agent_id == 1:
agent = AgentDQN(action_set=action_set, slot_set=slot_set, disease_symptom=disease_symptom, parameter=parameter)
elif agent_id == 2:
agent = AgentActorCritic(action_set=action_set, slot_set=slot_set, disease_symptom=disease_symptom,
parameter=parameter)
elif agent_id == 3:
agent = AgentRandom(action_set=action_set, slot_set=slot_set, disease_symptom=disease_symptom,
parameter=parameter)
elif agent_id == 0:
agent = AgentRule(action_set=action_set, slot_set=slot_set, disease_symptom=disease_symptom,
parameter=parameter)
steward.simulate(agent=agent, epoch_number=simulate_epoch_number, train_mode=train_mode)