Change to TemplateNLG

ConvLab-3

@@ -0,0 +1 @@
+Subproject commit 60f4e5641f93e99b8d61b49cf5fd6dc818a83c4c

DialoguePolicy.py

@@ -1,20 +1,12 @@
 from collections import defaultdict
 import json
-import random
-import string
 from copy import deepcopy
 from convlab.policy.policy import Policy
 
 db_path = './hotels_data.json'
 
 
-def generate_reference_number(length=8):
-    letters_and_digits = string.ascii_uppercase + string.digits
-    reference_number = ''.join(random.choice(letters_and_digits) for _ in range(length))
-    return reference_number
-
 class DialoguePolicy(Policy):
-    info_dict = None
     def __init__(self):
         Policy.__init__(self)
         self.db = self.load_database(db_path)
@@ -48,8 +40,7 @@ class DialoguePolicy(Policy):
         if any(True for slots in user_action.values() for (slot, _) in slots if
                slot in ['book stay', 'book day', 'book people']):
             if self.results:
-                reference_number = generate_reference_number()
-                system_action = {('Booking', 'Book'): [["Ref", reference_number]]}
+                system_action = {('Booking', 'Book'): [["Ref", self.results[0].get('Ref', 'N/A')]]}
 
         system_acts = [[intent, domain, slot, value] for (domain, intent), slots in system_action.items() for
                        slot, value in slots]
@@ -58,29 +49,28 @@ class DialoguePolicy(Policy):
 
     def update_system_action(self, user_act, user_action, state, system_action):
         domain, intent = user_act
-        if domain in state['belief_state']:
-            constraints = [(slot, value) for slot, value in state['belief_state'][domain]['info'].items() if value != '']
-            # print(f"Constraints: {constraints}")
-            self.results = deepcopy(self.query(domain.lower(), constraints))
-            # print(f"Query results: {self.results}")
+        constraints = [(slot, value) for slot, value in state['belief_state'][domain]['info'].items() if value != '']
+        # print(f"Constraints: {constraints}")
+        self.results = deepcopy(self.query(domain.lower(), constraints))
+        # print(f"Query results: {self.results}")
 
-            if intent == 'request':
-                if len(self.results) == 0:
-                    system_action[(domain, 'NoOffer')] = []
-                else:
-                    for slot in user_action[user_act]:
-                        if slot[0] in self.results[0]:
-                            system_action[(domain, 'Inform')].append([slot[0], self.results[0].get(slot[0], 'unknown')])
+        if intent == 'request':
+            if len(self.results) == 0:
+                system_action[(domain, 'NoOffer')] = []
+            else:
+                for slot in user_action[user_act]:
+                    if slot[0] in self.results[0]:
+                        system_action[(domain, 'Inform')].append([slot[0], self.results[0].get(slot[0], 'unknown')])
 
-            elif intent == 'inform':
-                if len(self.results) == 0:
-                    system_action[(domain, 'NoOffer')] = []
-                else:
-                    system_action[(domain, 'Inform')].append(['Choice', str(len(self.results))])
-                    choice = self.results[0]
+        elif intent == 'inform':
+            if len(self.results) == 0:
+                system_action[(domain, 'NoOffer')] = []
+            else:
+                system_action[(domain, 'Inform')].append(['Choice', str(len(self.results))])
+                choice = self.results[0]
 
-                    if domain in ["hotel"]:
-                        system_action[(domain, 'Recommend')].append(['Name', choice['name']])
-                        for slot in state['belief_state'][domain]['info']:
-                            if choice.get(slot):
-                                state['belief_state'][domain]['info'][slot] = choice[slot]
+                if domain in ["hotel"]:
+                    system_action[(domain, 'Recommend')].append(['Name', choice['name']])
+                    for slot in state['belief_state'][domain]['info']:
+                        if choice.get(slot):
+                            state['belief_state'][domain]['info'][slot] = choice[slot]

DialogueStateTracker.py

@@ -27,12 +27,11 @@ def default_state():
         'user_action': [],
         'system_action': [],
         'terminated': False,
-        'booked': {}
+        'booked': []
     }
 
 
 class DialogueStateTracker(DST):
-    info_dict = None
     def __init__(self):
         DST.__init__(self)
         self.state = default_state()

MachineLearningNLG.py

@@ -33,15 +33,14 @@ class MachineLearningNLG:
         return translated_response
 
     def generate(self, action):
-        # Przyjmujemy, że 'action' jest formatowanym stringiem, który jest przekazywany do self.nlg
         return self.nlg(action)
 
     def init_session(self):
-        pass  # Dodanie pustej metody init_session
+        pass
 
 
 # Przykład użycia
 if __name__ == "__main__":
     nlg = MachineLearningNLG()
-    system_act = "inform(date.from=15.07, date.to=22.07)"
+    system_act = "inform(hotel='Four Seasons Hotel')"
     print(nlg.nlg(system_act))

Main.py

@@ -1,22 +1,17 @@
 from NaturalLanguageAnalyzer import NaturalLanguageAnalyzer
 from DialoguePolicy import DialoguePolicy
 from DialogueStateTracker import DialogueStateTracker
-from NaturalLanguageGeneration import NaturalLanguageGeneration
 from convlab.dialog_agent import PipelineAgent
-import warnings
-warnings.filterwarnings("ignore")
-
+from MachineLearningNLG import MachineLearningNLG  # Importujemy nowy komponent NLG
+from convlab.nlg.template.multiwoz import TemplateNLG
 
 if __name__ == "__main__":
+    text = "chciałbym zarezerwować drogi hotel z parkingiem 1 stycznia w Warszawie w centrum"
     nlu = NaturalLanguageAnalyzer()
     dst = DialogueStateTracker()
     policy = DialoguePolicy()
-    nlg = NaturalLanguageGeneration()
+    nlg = TemplateNLG(is_user=False)
 
     agent = PipelineAgent(nlu=nlu, dst=dst, policy=policy, nlg=nlg, name='sys')
-
-    print("Witam, jestem systemem do rezerwowania pokoi hotelowych. W czym mogę Ci pomóc?")
-    while True:
-        text = input()
-        response = agent.response(text)
-        print(response)
+    response = agent.response(text)
+    print(response)

NaturalLanguageAnalyzer.py

@@ -14,7 +14,6 @@ def translate_text(text, target_language='en'):
 
 
 class NaturalLanguageAnalyzer:
-    info_dict = None
     def predict(self, text, context=None):
         # Inicjalizacja modelu NLU
         model_name = "ConvLab/t5-small-nlu-multiwoz21"

NaturalLanguageGeneration.py

@@ -1,28 +1,12 @@
-import requests
-
-from convlab.nlg.template.multiwoz import TemplateNLG
-
-
-def translate_text(text, target_language='pl'):
-    url = 'https://translate.googleapis.com/translate_a/single?client=gtx&sl=auto&tl={}&dt=t&q={}'.format(
-        target_language, text)
-    response = requests.get(url)
-    if response.status_code == 200:
-        translated_text = response.json()[0]
-        translated_text_joined = ''.join([sentence[0] for sentence in translated_text])
-        return translated_text_joined
-    else:
-        return None
+import re
 
 class NaturalLanguageGeneration:
-    info_dict = None
-    def generate(self, system_act):
-        if len(system_act) == 0:
-            return "Nie rozumiem."
-        tnlg = TemplateNLG(is_user=False)
-        response_en = tnlg.generate(system_act)
-        return translate_text(response_en)
 
-
-    def init_session(self):
-        pass  # Dodanie pustej metody init_session
+    def nlg(self, system_act):
+        response = None
+        pattern = r'inform\(name=([^\)]+)\)'
+        matching = re.search(pattern, system_act)
+        if matching:
+            name = matching.group(1)
+            response = f"Witaj, nazywam się {name}"
+        return response

train_nlg.py

@@ -11,7 +11,8 @@ for file_name in os.listdir(translated_data_directory):
     if file_name.endswith('.tsv'):
         file_path = os.path.join(translated_data_directory, file_name)
         df = pd.read_csv(file_path, sep='\t')
-        dfs.append(df)
+        df_user = df[df['role'] == 'system'].drop('role', axis=1)
+        dfs.append(df_user)
 combined_df = pd.concat(dfs, ignore_index=True)
 
 # Przygotowanie zbioru danych do trenowania
@@ -49,7 +50,7 @@ training_args = Seq2SeqTrainingArguments(
     per_device_eval_batch_size=16,
     predict_with_generate=True,
     learning_rate=5e-5,
-    num_train_epochs=3,
+    num_train_epochs=10,
     evaluation_strategy="epoch",
     save_strategy="epoch",
     save_total_limit=None,  # Wyłącz rotację punktów kontrolnych