import json from adapt.engine import DomainIntentDeterminationEngine from padatious import IntentContainer from jarvis.utils import utils adapt_engine = DomainIntentDeterminationEngine() padatious_intents_container = IntentContainer('intent_cache') intents_handlers_adapt = dict() intents_handlers_padatious = dict() def register_entity_adapt(entity_value, entity_type, domain): adapt_engine.register_entity(entity_value=entity_value, entity_type=entity_type, domain=domain) # print("[Adapt]: Added entity with type " + entity_type + " for " + domain) def register_entity_padatious(entity_name, file_lines_list): padatious_intents_container.add_entity(entity_name, file_lines_list) # print("[Padatious]: Added entity with name " + entity_name + " with " str(len(list)) + "examples.") def register_regex_adapt(regex, domain): adapt_engine.register_regex_entity(regex, domain) # print("[Adapt]: Added new regex for " + domain) def register_intent_adapt(intent, domain): adapt_engine.register_intent_parser(intent, domain=domain) print("[Adapt]: Registered new intent " + intent.name + " for skill " + domain + ".") def register_intent_padatious(intent_name, list_of_intent_examples): padatious_intents_container.add_intent(intent_name, list_of_intent_examples) print("[Padatious]: Registered new intent " + intent_name + " with " + str( len(list_of_intent_examples)) + " examples.") def train_padatious(): padatious_intents_container.train() def load_all_skills(): for handler in intents_handlers_adapt: function_handler = intents_handlers_adapt.get(handler) intent_builder = getattr(function_handler[0], "_data", [])[0] skill_name = function_handler[1] register_intent_adapt(intent_builder.build(), domain=skill_name) for handler in intents_handlers_padatious: function_handler = intents_handlers_padatious.get(handler) intent_data_examples = function_handler[1] register_intent_padatious(handler, intent_data_examples) def handle(intent_name, data): module_path_str = None handler_method_name = None if intent_name in intents_handlers_adapt: # something like handler_play_song_spotify (used to call the handler method from the skill imported below) handler_method_name = intents_handlers_adapt.get(intent_name)[2] # something like jarvis.skill.entertainment.spotify (used to import the create_skill method to create a new object) module_path_str = intents_handlers_adapt.get(intent_name)[3] if intent_name in intents_handlers_padatious: handler_method_name = intents_handlers_padatious.get(intent_name)[0] module_path_str = intents_handlers_padatious.get(intent_name)[2] if module_path_str is not None and handler_method_name is not None: # import the create_skill method from the skill using the skill module path create_skill_method = utils.import_method_from_string(module_path_str, "create_skill") skill_init_data = {'client_ip': data['client_ip'], 'client_port': data['client_port']} # create a new object of the right skill for the utterance skill = create_skill_method(skill_init_data) # import and call the handler method from the skill getattr(skill, handler_method_name)(data=data) def train_padatious(): print("Training PADATIOUS intents models, can take a few minutes (first time) or a few seconds (startup)") padatious_intents_container.train(timeout=120) def recognise(sentence, client_ip=None, client_port=None): sentence = sentence.lower() print(sentence) data = dict() data['client_ip'] = client_ip data['client_port'] = client_port data['utterance'] = sentence best_intent_adapt = get_best_intent_adapt(sentence) best_intent_padatious = get_best_intent_padatious(sentence) confidence_adapt = get_confidence(best_intent_adapt) confidence_padatious = get_confidence(best_intent_padatious) if confidence_adapt < 0.2 and confidence_padatious < 0.2: return "I didn't understand..." else: return handle_intent(data, best_intent_adapt if confidence_adapt > confidence_padatious else best_intent_padatious) def get_confidence(intent): if intent is None: return 0 if 'confidence' in intent: return intent['confidence'] elif hasattr(intent, 'conf'): return intent.conf else: return 0 def get_best_intent_adapt(sentence): if len(intents_handlers_adapt) > 0: try: best_intents = adapt_engine.determine_intent(sentence, 100) best_intent = next(best_intents) return best_intent except StopIteration: pass return None # No match (Adapt) def get_best_intent_padatious(sentence): if len(intents_handlers_padatious) > 0: result = padatious_intents_container.calc_intent(sentence) return result else: return None # No match (Padatious) def handle_intent(data, intent): if 'intent_type' in intent: return handle_adapt_intent(data, intent) elif hasattr(intent, 'name'): return handle_padatious_intent(data, intent) def handle_adapt_intent(data, best_intent): for key, val in best_intent.items(): if key != 'intent_type' and key != 'target' and key != 'confidence': data[key] = val handle(best_intent['intent_type'], data=data) return best_intent def handle_padatious_intent(data, result): data.update(result.matches) # adding the matches from padatious to the data handle(result.name, data) return json.dumps(str(result))