def test_intent_classification():
nludatafile = "/media/nvidia/ssd/catkin_ws/src/chu_bot_source/chubot/data/nlu_greet.json"
chubot = ChuBotBrain("greet_en", language='en')
chubot.load_data(nludatafile)
meta = chubot.train_intent_classification()
print(meta)
inmessage = "sad but great. want a dog picture"
intent_probs = chubot.predict_intent(inmessage)
print(intent_probs)
def test_nlu_predict():
nludatafile = "/media/nvidia/ssd/catkin_ws/src/chu_bot_source/chubot/data/nlu_greet.json"
chubot = ChuBotBrain("greet_en", language='en')
inmessage = "a bird and a dog"
tagged_entities = chubot.predict_entity(inmessage)
print(inmessage)
print(tagged_entities)
inmessage = "sad but great. want a dog picture"
intent_probs = chubot.predict_intent(inmessage)
print(inmessage)
print(intent_probs)
def test_intent_train(name):
chubot = ChuBotBrain(name, language='vi')
chubot.load_data("data/train.json")
# chubot.load_data("data/vi_nlu_ask_way.json")
meta = chubot.train_intent_classification()
test_link = "data/test.txt"
with open(test_link,'r',encoding='utf=8') as f:
rows = f.readlines()
intent_list_test = []
intent_list_result = []
data_list_test = []
for row in rows:
parts = row.split(',')
intent_list_test.append(encode_intent(parts[0]))
data_list_test.append(parts[1])
for data in data_list_test:
responses = chubot.predict_intent(data)
(prob, intent) = responses[0]
intent_list_result.append(encode_intent(intent))
print("accuracy",accuracy_score(intent_list_test,intent_list_result))
class ChuBotAction():
def __init__(self, botname, language='vi'):
self.botname = botname
self.chubot = ChuBotBrain(botname, language)
self.followup_actions = None
self.action_templates = None
self.action_custom = None
self.slots = None
self.start_flag = True
self.active_entities = {}
# TODO perhaps if folder exists -> load metadata file
def load_domain(self, domain_file):
"""Load domain info to bot
"""
# TODO handle exception here
# setter -hmm, is this okay? should use @property or not?
with io.open(domain_file) as f:
action_domain = json.loads(
open(domain_file, encoding="utf-8").read())
# data fields
self.followup_actions = action_domain["action_domain_data"][
"followup_actions"]
self.action_templates = action_domain["action_domain_data"][
"action_templates"]
self.action_custom = action_domain["action_domain_data"][
"action_custom"]
self.slots = action_domain["action_domain_data"]["slots"]
def handle_action(self, action, **kwargs):
"""Handle a response action
Select a random template for the action
or fill the slot
return a list of responses based on action
"""
import random
import re
# TODO Handle KeyError dictionary
response = ""
if action in self.action_templates:
# if action is a template
templates = self.action_templates.get(action)
template = random.choice(templates)
# TODO handle multiple detected entity but template only one?
# ->should in custom action
# TODO handle multiple slots
regex_pattern = r"{{(.*)}}"
template_entity = re.search(regex_pattern, template, re.M | re.I)
if not template_entity:
# no found
# return template as response
return [template]
else:
active_entities = kwargs['active_entities']
# TODO multiple slot filling
entity = template_entity.group(1)
for ent in active_entities:
if ent['entity'] == entity:
value = ent['value']
return [template.replace(template_entity.group(), value)]
elif action in self.action_custom:
import custom_actions
func = getattr(custom_actions,
self.action_custom[action].get("func_name"))
response = func(
action_args=self.action_custom[action].get("kwargs"), **kwargs)
return response
def begin_conversation(self, **kwargs):
"""Say welcome
"""
# TODO should fix name of this action or not?
start_flag = False
return self.handle_action("bot_welcome")
def handle_message(self,
inmessage,
debug=False,
intent_prob_threshold=0.3,
**kwargs):
"""Simple rule_based response
"""
import itertools
# TODO handle first welcome, repetitive input, etc -> state of conversation
entities_pred = self.chubot.predict_entity(inmessage)
intents_pred = self.chubot.predict_intent(inmessage)
if debug:
print("Predicted entity: ", entities_pred)
print("Predicted intent: ", intents_pred)
# select the first ranked predicted entity/intents
# handle entities with different values but same entity type
active_entities = {"active_entities": entities_pred}
# select highest probable intent
(prob, intent) = intents_pred[0]
if prob < intent_prob_threshold:
bot_actions = ["default_fallback"]
else:
# TODO handle key error?
bot_actions = self.followup_actions.get(intent)
if debug:
print(bot_actions)
bot_responses = [
self.handle_action(action, **active_entities)
for action in bot_actions
]
bot_responses = list(itertools.chain(*bot_responses))
return bot_responses
def run_commandline_bot(self, **kwargs):
"""A commandline bot
"""
# TODO handle action_domain_file doesnt match entities and intent in nludata file
print("Type stop to end.\n -------------------")
begin = self.begin_conversation()
print("Chubot:", begin)
while True:
sys.stdout.write("You > ")
inmessage = input()
if inmessage == 'stop':
break
responses = self.handle_message(inmessage, debug=False)
for response in responses:
print("Chubot:", response)
def custom_handle_message(self,
inmessage,
debug=False,
intent_prob_threshold=0.3,
**kwargs):
"""Simple rule_based response
"""
import itertools
# TODO handle first welcome, repetitive input, etc -> state of conversation
entities_pred = self.chubot.predict_entity(inmessage)
intents_pred = self.chubot.predict_intent(inmessage)
if debug:
print("Predicted entity: ", entities_pred)
print("Predicted intent: ", intents_pred)
# select the first ranked predicted entity/intents
# handle entities with different values but same entity type
active_entities = {"active_entities": entities_pred}
# select highest probable intent
(prob, intent) = intents_pred[0]
if prob < intent_prob_threshold:
bot_actions = ["default_fallback"]
else:
# TODO handle key error?
bot_actions = self.followup_actions.get(intent)
if debug:
print(bot_actions)
bot_responses = [
self.handle_action(action, **active_entities)
for action in bot_actions
]
bot_responses = list(itertools.chain(*bot_responses))
return bot_responses