Add new intent classifier, and add args parser to main

main.py

@@ -1,9 +1,9 @@
 import logging
-import time
-import schedule
+import argparse
+
+# import schedule
 
 import config
-
 from src.rhasspy import rhasspy_mqtt as yoda_listener
 from src.rhasspy import Rhasspy
 from src.ratatouille import Ratatouille
@@ -11,30 +11,73 @@ from src.mpd import Mpd
 from src.hass import Hass
 from src.httpServer import get_server
 from src.fuzzy import fuzz_predict
-# from src.intent import AlexaIntent
+from src.intent import BertIntent
+from src.tools.simple_sentence_corrector import simple_sentence_corrector
+
+
+# --------- setup args -------------
+
+parser = argparse.ArgumentParser(
+    prog='Ratatouille',
+    description='Ratatouille le cerveau domotique !')
+
+parser.add_argument('mode')
+parser.add_argument('-i', '--ip', required=False)
+parser.add_argument('-p', '--port', required=False, type=int)
+
+args = parser.parse_args()
+
+if args.mode == "server":
+    if args.ip is None or args.port is None:
+        logging.error(" --ip or --port argument missing")
+        exit()
+
+
+# -------- setup logging ------------
 
 logging.basicConfig(
     level=10,
     format="%(asctime)s %(filename)s:%(lineno)s %(levelname)s %(message)s"
 )
 
-IP = "10.10.10.8"
-PORT = 5555
+logging.info("Loading ratatouilles modules")
+
+
+# ---------- other ---------------
 
 walle = Hass(config.hass_url, config.hass_token)
 yoda = None  # Rhasspy(config.rhasspy_url)
 mopidy = Mpd('10.10.10.8')
-ratatouille = Ratatouille(yoda, walle, mopidy, schedule)
+ratatouille = Ratatouille(yoda, walle, mopidy, None)
 # alexa = AlexaIntent() # we are not doing any request to the evil amazon but we are using one of its dataset
+bert = BertIntent()
 
 
 def answer(sentence):
     # return ratatouille.parse_alexa(alexa.predict(sentence))
-    return ratatouille.parse_fuzzy(fuzz_predict(sentence))
+    sentence_corrected = simple_sentence_corrector(sentence)
+    prediction = bert.predict(sentence_corrected)
+    return ratatouille.parse_fuzzy(prediction)
     # return "42"
 
 
-server = get_server(IP, PORT, answer)
+def run_server(ip, port):
+    server = get_server(ip, port, answer)
+    logging.info('Running server on '+ip+':'+str(port))
+    server.serve_forever()
 
-logging.info('Running server on '+IP+':'+str(PORT))
-server.serve_forever()
+
+def run_prompt():
+    question = "empty"
+    while question != "stop":
+        question = input("?")
+        print(answer(question))
+
+
+logging.info("Ratatouille is ready !")
+
+# run_server()
+if args.mode == "server":
+    run_server(str(args.ip), args.port)
+else:
+    run_prompt()

src/intent.py

@@ -1,10 +1,12 @@
 import unittest
 from transformers import AutoTokenizer, AutoModelForTokenClassification, TokenClassificationPipeline
 from transformers import AutoModelForSequenceClassification, TextClassificationPipeline
+from transformers import pipeline
 
 DOMOTIQUE_OBJ_ON = ['iot_wemo_on']
 DOMOTIQUE_OBJ_OFF = ['iot_wemo_off']
 
+
 class AlexaIntent():
 
     def __init__(self):
@@ -15,21 +17,24 @@ class AlexaIntent():
         model_name = 'qanastek/XLMRoberta-Alexa-Intents-Classification'
         tokenizer = AutoTokenizer.from_pretrained(model_name)
         model = AutoModelForSequenceClassification.from_pretrained(model_name)
-        classifier = TextClassificationPipeline(model=model, tokenizer=tokenizer)
+        classifier = TextClassificationPipeline(
+            model=model, tokenizer=tokenizer)
         return classifier
 
     def init_entities_classification(self):
-        tokenizer = AutoTokenizer.from_pretrained('qanastek/XLMRoberta-Alexa-Intents-NER-NLU')
-        model = AutoModelForTokenClassification.from_pretrained('qanastek/XLMRoberta-Alexa-Intents-NER-NLU')
+        tokenizer = AutoTokenizer.from_pretrained(
+            'qanastek/XLMRoberta-Alexa-Intents-NER-NLU')
+        model = AutoModelForTokenClassification.from_pretrained(
+            'qanastek/XLMRoberta-Alexa-Intents-NER-NLU')
         predict = TokenClassificationPipeline(model=model, tokenizer=tokenizer)
         return predict
 
-    def simple_sentence_corrector(self,sentence):    
-        sentence = sentence.replace('étant','éteins')
-        sentence = sentence.replace('dépeint','éteins')
+    def simple_sentence_corrector(self, sentence):
+        sentence = sentence.replace('étant', 'éteins')
+        sentence = sentence.replace('dépeint', 'éteins')
         return sentence
 
-    def intent_corrector(self,intents):
+    def intent_corrector(self, intents):
         for intent in intents:
             if intent['label'] in DOMOTIQUE_OBJ_ON:
                 intent['label'] = 'iot_hue_lighton'
@@ -37,7 +42,7 @@ class AlexaIntent():
                 intent['label'] = 'iot_hue_lightoff'
         return intents
 
-    def predict(self,sentence):
+    def predict(self, sentence):
         sentence = self.simple_sentence_corrector(sentence)
         return {
             'intents': self.intent_corrector(self.get_intents(sentence)),
@@ -45,6 +50,47 @@ class AlexaIntent():
         }
 
 
+class BertIntent():
+    def __init__(self):
+        self.classifier = pipeline("text-classification",
+                                   model="Tjiho/french-intents-classificaton")
+
+    def predict(self, sentence):
+        # sentence = self.simple_sentence_corrector(sentence)
+        classification = self.classifier(sentence)
+
+        if classification[0]["score"] < 0.7:  # score too low
+            return
+
+        label = classification[0]["label"]
+
+        if label == "HEURE":
+            return {'intentName': 'GetTime'}
+        elif label == "DATE":
+            return {'intentName': 'GetDate'}
+        elif label == "ETEINDRE_CUISINE":
+            return {'intentName': 'GetTime', 'intentArg': ['cuisine']}
+        elif label == "ETEINDRE_BUREAU":
+            return {'intentName': 'LightOff', 'intentArg': ['bureau']}
+        elif label == "ETEINDRE_SALON":
+            return {'intentName': 'LightOff', 'intentArg': ['salon']}
+        elif label == "ETEINDRE_CHAMBRE":
+            return {'intentName': 'LightOff', 'intentArg': ['chambre']}
+        elif label == "ALLUMER_CUISINE":
+            return {'intentName': 'LightOn', 'intentArg': ['cuisine']}
+        elif label == "ALLUMER_SALON":
+            return {'intentName': 'LightOn', 'intentArg': ['salon']}
+        elif label == "ALLUMER_BUREAU":
+            return {'intentName': 'LightOn', 'intentArg': ['bureau']}
+        elif label == "ALLUMER_CHAMBRE":
+            return {'intentName': 'LightOn', 'intentArg': ['chambre']}
+        elif label == "METEO":
+            return {'intentName': 'Meteo'}
+        elif label == "TEMPERATURE_EXTERIEUR":
+            return {'intentName': 'Temperature_ext'}
+        elif label == "TEMPERATURE_INTERIEUR":
+            return {'intentName': 'Temperature_int'}
+
 
 class TestAlexa(unittest.TestCase):
     @classmethod
@@ -60,11 +106,11 @@ class TestAlexa(unittest.TestCase):
         self.assertEqual(res['intents'][0]['label'], 'iot_hue_lighton')
         self.assertEqual(res['entities'][0]['word'], '▁cuisine')
 
-
     def test_bad_transcribe(self):
         res = self.alexa.predict("dépeint la cuisine")
         self.assertEqual(res['intents'][0]['label'], 'iot_hue_lightoff')
         self.assertEqual(res['entities'][0]['word'], '▁cuisine')
 
+
 if __name__ == '__main__':
     unittest.main()

src/tools/simple_sentence_corrector.py

@@ -0,0 +1,4 @@
+def simple_sentence_corrector(sentence):
+    sentence = sentence.replace('étant', 'éteins')
+    sentence = sentence.replace('dépeint', 'éteins')
+    return sentence