Update app.py

app.py

@@ -142,30 +142,8 @@ class Chatbot:
         similar_contexts = [self.df['Context'][indices[0][i]] for i in range(self.k)]
         return similar_questions, similar_contexts, distances, indices
 
-    def predict_faiss(self, message):
-        message = message.strip()
-        question_vector = self.get_embeddings(message)
-        question_vector = self.prepare_sentences_vector([question_vector])
-        similar_questions, similar_contexts, distances, indices = self.faiss_search(question_vector)
-        Answers = [self.df['Answer'][i] for i in indices[0]]
-        Answer = Answers[0]
-
-        return Answer
-
-    def predict_faiss(self, message):
-        message = message.strip()
-        question_vector = self.get_embeddings([message])
-        question_vector = self.prepare_sentences_vector([question_vector])
-        similar_questions, similar_contexts, distances, indices = self.faiss_search(question_vector)
-        Answers = [self.df['Answer'][i] for i in indices[0]]
-        Answer = Answers[0]
-
-        return Answer
-
-    
 
-    # Function to predict using BERT embedding
-    def predict_bert_embedding(self,message):
+    def predict(self,message):
         message = message.strip()
         question_vector = self.get_embeddings(message)
         question_vector=self.prepare_sentences_vector([question_vector])
@@ -173,62 +151,11 @@ class Chatbot:
         Answer = self.model_pipeline(similar_questions, similar_contexts)
         return Answer
 
-    # def predict_semantic_search(self,message,corpus_embeddings):
-    #     message = message.strip()
-    #     query_embedding = self.embedding_model.encode(message, convert_to_tensor=True)
-    #     query_embedding = query_embedding.to('cpu')
-    #     hits = util.semantic_search(query_embedding, corpus_embeddings, top_k=1)
-    #     hit = hits[0][0]
-    #     context=self.df['Context'][hit['corpus_id']]
-    #     score="{:.4f})".format(hit['score'])
-    #     Answer = self.model_pipeline(message, context)
-    #     return Answer
-    def predict_semantic_search(self, message):
-        corpus_embeddings = bot._chatbot.prepare_sentences_vector(bot._chatbot.get_embeddings(bot._chatbot.df['Context']))
-        message = message.strip()
-        query_embedding = self.embedding_model.encode(message, convert_to_tensor=True)
-        query_embedding = query_embedding.to('cuda')
-        hits = util.semantic_search(query_embedding, corpus_embeddings, top_k=1)
-        hit = hits[0][0]
-        context = self.df['Context'][hit['corpus_id']]
-        Answer = self.model_pipeline(message, context)
-        return Answer
-
-
-
-    def predict_without_faiss(self,message):
-        MostSimilarContext = ""
-        min_distance = 1000
-        message = message.strip(' \t\n')
-        question_vector = self.get_embeddings([message])
-        question_vector=self.prepare_sentences_vector(question_vector)
-        for j, _question_vector in enumerate(self.vectors):
-            distance = euclidean_distances(question_vector, _question_vector.reshape(1, -1))[0][0]
-            if distance < min_distance:
-                min_distance = distance
-                MostSimilarContext = self.df['Context'][j]
-                similar_question = self.df['Question'][j]
-            if distance <= 0.02469331026:
-                break
-        predict_answer = self.model_pipeline(message, MostSimilarContext)
-        Answer = predict_answer.strip().replace("<unk>","@")
-        return Answer
-
 bot = ChatbotModel()
 
 """#Gradio"""
 
 
-EXAMPLE_PATH = ["หลิน ไห่เฟิง มีชื่อเรียกอีกชื่อว่าอะไร" , "ใครเป็นผู้ตั้งสภาเศรษฐกิจโลกขึ้นในปี พ.ศ. 2514 โดยทุกปีจะมีการประชุมที่ประเทศสวิตเซอร์แลนด์", "โปรดิวเซอร์ของอัลบั้มตลอดกาล ของวงคีรีบูนคือใคร", "สกุลเดิมของหม่อมครูนุ่ม นวรัตน ณ อยุธยา คืออะไร"]
-
-# demoFaiss = gr.ChatInterface(fn=bot._chatbot.predict_faiss, examples=EXAMPLE_PATH)
-# demoBert = gr.ChatInterface(fn=bot._chatbot.predict_bert_embedding,examples=EXAMPLE_PATH)
-# demoSemantic = gr.ChatInterface(fn=bot._chatbot.predict_semantic_search,examples=EXAMPLE_PATH)
-# demoWithoutFiss = gr.ChatInterface(fn=bot._chatbot.predict_without_faiss,examples=EXAMPLE_PATH)
-demoFaiss = gr.Interface(fn=bot._chatbot.predict_faiss, inputs="text", outputs="text", title="TH wiki (just Faiss)")
-demoBert = gr.Interface(fn=bot._chatbot.predict_bert_embedding, inputs="text", outputs="text", title="TH wiki (Faiss & Model)")
-demoSemantic = gr.Interface(fn=bot._chatbot.predict_semantic_search, inputs="text", outputs="text", title="TH wiki (Semantic Search & Model)")
-demoWithoutFiss = gr.Interface(fn=bot._chatbot.predict_without_faiss, inputs="text", outputs="text", title="TH wiki (just Model)")
-
-demo = gr.TabbedInterface([demoFaiss, demoWithoutFiss, demoBert, demoSemantic], ["Faiss", "Model", "Faiss & Model", "Semantic Search & Model"])
+# EXAMPLE = ["หลิน ไห่เฟิง มีชื่อเรียกอีกชื่อว่าอะไร" , "ใครเป็นผู้ตั้งสภาเศรษฐกิจโลกขึ้นในปี พ.ศ. 2514 โดยทุกปีจะมีการประชุมที่ประเทศสวิตเซอร์แลนด์", "โปรดิวเซอร์ของอัลบั้มตลอดกาล ของวงคีรีบูนคือใคร", "สกุลเดิมของหม่อมครูนุ่ม นวรัตน ณ อยุธยา คืออะไร"]
+demo = gr.Interface(fn=bot._chatbot.predict, inputs="text", outputs="text", title="CE66-04_Thai Question Answering System by using Deep Learning")
 demo.launch()