Update app.py

app.py

@@ -1,203 +1,165 @@
 import logging
-import os
 import gradio as gr
 from transformers import pipeline
 from sentence_transformers import SentenceTransformer, util
 import PyPDF2
 
-# Set up logging: write logs to a writable directory (/tmp)
-log_file_path = "/tmp/support_bot_log.txt"
-logging.basicConfig(filename=log_file_path, level=logging.INFO, format='%(asctime)s - %(message)s')
-
-def flush_logs():
-    for handler in logging.getLogger().handlers:
-        handler.flush()
-
-class SupportBotAgent:
-    def __init__(self, document_path):
-        # Load a pre-trained question-answering model
-        self.qa_model = pipeline("question-answering", model="distilbert-base-uncased-distilled-squad")
-        # Set up an embedding model for finding relevant sections
-        self.embedder = SentenceTransformer('all-MiniLM-L6-v2')
-        # Load the document text and split it into sections (by paragraphs)
-        self.document_text = self.load_document(document_path)
-        self.sections = self.document_text.split('\n\n')
-        flush_logs()
-        # Log document length for debugging
-        logging.info(f"Document length: {len(self.document_text)} characters")
-        flush_logs()
-        # Create embeddings for all sections
-        self.section_embeddings = self.embedder.encode(self.sections, convert_to_tensor=True)
-        logging.info(f"Loaded document: {document_path}")
-        flush_logs()
-
-    def load_document(self, path):
-        """Loads and extracts text from a TXT or PDF file."""
-        if path.lower().endswith(".txt"):
-            file_type = "Text File"
-            with open(path, 'r', encoding='utf-8') as file:
-                text = file.read()
-        elif path.lower().endswith(".pdf"):
-            file_type = "PDF File"
-            text = ""
-            with open(path, "rb") as file:
-                pdf_reader = PyPDF2.PdfReader(file)
-                for page in pdf_reader.pages:
-                    page_text = page.extract_text()
-                    if page_text:
-                        text += page_text + "\n"
-        else:
-            file_type = "Unsupported Format"
-            logging.error(f"Unsupported file format: {path}")
-            flush_logs()
-            raise ValueError("Unsupported file format. Please provide a TXT or PDF file.")
-        logging.info(f"Loaded {file_type}: {path}")
-        flush_logs()
-        return text
-
-    def find_relevant_section(self, query):
-        """
-        Uses semantic similarity first, falling back to keyword search if needed.
-        """
-        stopwords = {"and", "the", "is", "for", "to", "a", "an", "of", "in", "on", "at", "with", "by", "it", "as", "so", "what"}
-        query_embedding = self.embedder.encode(query, convert_to_tensor=True)
-        similarities = util.cos_sim(query_embedding, self.section_embeddings)[0]
-        best_idx = similarities.argmax().item()
-        best_section = self.sections[best_idx]
-        similarity_score = similarities[best_idx].item()
-        SIMILARITY_THRESHOLD = 0.4  # Adjust if needed
-
-        if similarity_score >= SIMILARITY_THRESHOLD:
-            logging.info(f"Found relevant section using embeddings for query: {query} (score: {similarity_score})")
-            flush_logs()
-            return best_section
-
-        logging.info(f"Low similarity ({similarity_score}) for query: {query}. Falling back to keyword search.")
-        flush_logs()
-        query_words = {word for word in query.lower().split() if word not in stopwords}
-        for section in self.sections:
-            section_words = {word for word in section.lower().split() if word not in stopwords}
-            common_words = query_words.intersection(section_words)
-            if len(common_words) >= 2:
-                logging.info(f"Keyword match for query: {query} with common words: {common_words}")
-                flush_logs()
-                return section
-
-        logging.info("No good keyword match found. Returning default response.")
-        flush_logs()
-        return "I don’t have enough information to answer that."
-
-    def answer_query(self, query):
-        context = self.find_relevant_section(query)
-        if not context:
-            answer = "I don’t have enough information to answer that."
-        else:
-            result = self.qa_model(question=query, context=context, max_answer_len=50)
-            answer = result["answer"]
-        logging.info(f"Answer for query '{query}': {answer}")
-        flush_logs()
-        return answer
-
-    def adjust_response(self, query, response, feedback):
-        """Adjusts the response based on feedback."""
-        if feedback == "too vague":
-            context = self.find_relevant_section(query)
-            adjusted_response = f"{response}\n\n(More details:\n{context[:500]}...)"
-        elif feedback == "not helpful":
-            adjusted_response = self.answer_query(query + " Please provide more detailed information with examples.")
-        else:
-            adjusted_response = response
-        logging.info(f"Adjusted answer for query '{query}': {adjusted_response}")
-        flush_logs()
-        return adjusted_response
-
-# --- Gradio Functions and App Workflow ---
-
+# Set up logging with immediate writing
+logging.basicConfig(
+    filename='support_bot_log.txt',
+    level=logging.INFO,
+    format='%(asctime)s - %(message)s',
+    force=True  # Ensures any existing handlers are replaced and logging starts fresh
+)
+logger = logging.getLogger()
+
+# Load models
+qa_model = pipeline("question-answering", model="distilbert-base-uncased-distilled-squad")
+embedder = SentenceTransformer('all-MiniLM-L6-v2')
+
+# Helper function to extract text from PDF
+def extract_text_from_pdf(file_path):
+    text = ""
+    with open(file_path, "rb") as file:
+        pdf_reader = PyPDF2.PdfReader(file)
+        for page in pdf_reader.pages:
+            text += page.extract_text() + "\n"
+    return text
+
+# Find the most relevant section in the document
+def find_relevant_section(query, sections, section_embeddings):
+    stopwords = {"and", "the", "is", "for", "to", "a", "an", "of", "in", "on", "at", "with", "by", "it", "as", "so", "what"}
+    
+    # Semantic search
+    query_embedding = embedder.encode(query, convert_to_tensor=True)
+    similarities = util.cos_sim(query_embedding, section_embeddings)[0]
+    best_idx = similarities.argmax().item()
+    best_section = sections[best_idx]
+    similarity_score = similarities[best_idx].item()
+    
+    SIMILARITY_THRESHOLD = 0.4
+    if similarity_score >= SIMILARITY_THRESHOLD:
+        logger.info(f"Found relevant section using embeddings for query: {query}")
+        return best_section
+    
+    logger.info(f"Low similarity ({similarity_score}). Falling back to keyword search.")
+    
+    # Keyword-based fallback search with stopword filtering
+    query_words = {word for word in query.lower().split() if word not in stopwords}
+    for section in sections:
+        section_words = {word for word in section.lower().split() if word not in stopwords}
+        common_words = query_words.intersection(section_words)
+        if len(common_words) >= 2:
+            logger.info(f"Keyword match found for query: {query} with common words: {common_words}")
+            return section
+    
+    logger.info(f"No good keyword match found. Returning default fallback response.")
+    return "I don’t have enough information to answer that."
+
+# Process the uploaded file with detailed logging
 def process_file(file, state):
-    """Handles file upload and initializes the SupportBotAgent."""
-    logging.info("Received file upload request")
-    flush_logs()
     if file is None:
-        logging.info("No file uploaded")
-        flush_logs()
-        return [("Bot", "Please upload a TXT or PDF file.")], state
-
-    # Save the uploaded file to /tmp. Handle both file objects and NamedString.
-    temp_path = os.path.join("/tmp", file.name)
-    with open(temp_path, "wb") as f:
-        if hasattr(file, "read"):
-            content = file.read()
+        logger.info("No file uploaded.")
+        return [("Bot", "Please upload a file.")], state
+    
+    file_path = file.name
+    if file_path.lower().endswith(".pdf"):
+        logger.info(f"Uploaded PDF file: {file_path}")
+        text = extract_text_from_pdf(file_path)
+    elif file_path.lower().endswith(".txt"):
+        logger.info(f"Uploaded TXT file: {file_path}")
+        with open(file_path, 'r', encoding='utf-8') as f:
+            text = f.read()
+    else:
+        logger.error(f"Unsupported file format: {file_path}")
+        return [("Bot", "Unsupported file format. Please upload a PDF or TXT file.")], state
+    
+    sections = text.split('\n\n')
+    section_embeddings = embedder.encode(sections, convert_to_tensor=True)
+    state['document_text'] = text
+    state['sections'] = sections
+    state['section_embeddings'] = section_embeddings
+    state['current_query'] = None
+    state['feedback_count'] = 0
+    state['mode'] = 'waiting_for_query'
+    state['chat_history'] = [("Bot", "File processed. You can now ask questions.")]
+    logger.info(f"Processed file: {file_path}")
+    return state['chat_history'], state
+
+# Handle user input (queries and feedback)
+def handle_input(user_input, state):
+    if state['mode'] == 'waiting_for_upload':
+        state['chat_history'].append(("Bot", "Please upload a file first."))
+        logger.info("User attempted to interact without uploading a file.")
+    elif state['mode'] == 'waiting_for_query':
+        query = user_input
+        state['current_query'] = query
+        state['feedback_count'] = 0
+        context = find_relevant_section(query, state['sections'], state['section_embeddings'])
+        if context == "I don’t have enough information to answer that.":
+            answer = context
         else:
-            content = file
-        if isinstance(content, str):
-            content = content.encode("utf-8")
-        f.write(content)
-    logging.info(f"Saved uploaded file to {temp_path} (size: {os.path.getsize(temp_path)} bytes)")
-    flush_logs()
-
-    try:
-        state["agent"] = SupportBotAgent(temp_path)
-    except Exception as e:
-        logging.error(f"Error processing file: {str(e)}")
-        flush_logs()
-        return [("Bot", f"Error processing file: {str(e)}")], state
-
-    state["chat_history"] = [("Bot", "File loaded successfully. Enter your query (or type 'exit' to end):")]
-    state["mode"] = "query"
-    state["last_query"] = ""
-    state["last_answer"] = ""
-    state["feedback_count"] = 0
-    return state["chat_history"], state
-
-def process_input(user_input, state):
-    """
-    Processes user input: as a query or feedback.
-    Typing 'exit' stops the session.
-    """
-    if state.get("mode", "query") == "ended":
-        return state["chat_history"], state
-
-    if user_input.lower() == "exit":
-        state["chat_history"].append(("Bot", "Session ended. You may now download the log file."))
-        state["mode"] = "ended"
-        flush_logs()
-        return state["chat_history"], state
-
-    if state["mode"] == "query":
-        state["last_query"] = user_input
-        answer = state["agent"].answer_query(user_input)
-        state["last_answer"] = answer
-        state["feedback_count"] = 0
-        state["chat_history"].append(("User", user_input))
-        state["chat_history"].append(("Bot", f"Answer: {answer}\nPlease provide feedback (good, too vague, not helpful):"))
-        state["mode"] = "feedback"
-    elif state["mode"] == "feedback":
+            result = qa_model(question=query, context=context)
+            answer = result["answer"]
+        state['last_answer'] = answer
+        state['mode'] = 'waiting_for_feedback'
+        state['chat_history'].append(("User", query))
+        state['chat_history'].append(("Bot", f"Answer: {answer}\nPlease provide feedback: good, too vague, not helpful."))
+        logger.info(f"Query: {query}, Answer: {answer}")
+    elif state['mode'] == 'waiting_for_feedback':
         feedback = user_input.lower()
-        state["chat_history"].append(("User", feedback))
-        if feedback == "good" or state["feedback_count"] >= 1:
-            state["chat_history"].append(("Bot", "Thank you for your feedback. Enter your next query (or type 'exit' to end):"))
-            state["mode"] = "query"
+        state['chat_history'].append(("User", feedback))
+        logger.info(f"Feedback: {feedback}")
+        if feedback == "good" or state['feedback_count'] >= 2:
+            state['mode'] = 'waiting_for_query'
+            if feedback == "good":
+                state['chat_history'].append(("Bot", "Thank you for your feedback. You can ask another question."))
+                logger.info("Feedback accepted as 'good'. Waiting for next query.")
+            else:
+                state['chat_history'].append(("Bot", "Maximum feedback iterations reached. You can ask another question."))
+                logger.info("Max feedback iterations reached. Waiting for next query.")
         else:
-            new_answer = state["agent"].adjust_response(state["last_query"], state["last_answer"], feedback)
-            state["last_answer"] = new_answer
-            state["feedback_count"] += 1
-            state["chat_history"].append(("Bot", f"Updated Answer: {new_answer}\nPlease provide feedback (good, too vague, not helpful):"))
-    flush_logs()
-    return state["chat_history"], state
-
-# --- Gradio UI Setup ---
-
+            query = state['current_query']
+            context = find_relevant_section(query, state['sections'], state['section_embeddings'])
+            if feedback == "too vague":
+                adjusted_answer = f"{state['last_answer']}\n\n(More details:\n{context[:500]}...)"
+            elif feedback == "not helpful":
+                adjusted_answer = qa_model(question=query + " Please provide more detailed information with examples.", context=context)['answer']
+            else:
+                state['chat_history'].append(("Bot", "Please provide valid feedback: good, too vague, not helpful."))
+                logger.info(f"Invalid feedback received: {feedback}")
+                return state['chat_history'], state
+            state['last_answer'] = adjusted_answer
+            state['feedback_count'] += 1
+            state['chat_history'].append(("Bot", f"Updated answer: {adjusted_answer}\nPlease provide feedback: good, too vague, not helpful."))
+            logger.info(f"Adjusted answer: {adjusted_answer}")
+    return state['chat_history'], state
+
+# Initial state
+initial_state = {
+    'document_text': None,
+    'sections': None,
+    'section_embeddings': None,
+    'current_query': None,
+    'feedback_count': 0,
+    'mode': 'waiting_for_upload',
+    'chat_history': [("Bot", "Please upload a PDF or TXT file to start.")],
+    'last_answer': None
+}
+
+# Gradio interface
 with gr.Blocks() as demo:
-    state = gr.State({"mode": "idle"})
-    gr.Markdown("## Customer Support Bot with Document Training")
-    file_upload = gr.File(label="Upload TXT or PDF file")
+    state = gr.State(initial_state)
+    file_upload = gr.File(label="Upload PDF or TXT file")
     chat = gr.Chatbot()
-    user_input = gr.Textbox(label="Enter your query or feedback")
+    user_input = gr.Textbox(label="Your query or feedback")
     submit_btn = gr.Button("Submit")
-    # Provide a file component to download the log file
-    log_file = gr.File(label="Download Log File", file_count="single", interactive=False, value=log_file_path)
+    log_file = gr.File(label="Download Log File", value="support_bot_log.txt")  # Added for log download
 
+    # Process file upload
     file_upload.upload(process_file, inputs=[file_upload, state], outputs=[chat, state])
-    submit_btn.click(process_input, inputs=[user_input, state], outputs=[chat, state]).then(lambda: "", None, user_input)
 
-demo.launch(share=True)
+    # Handle user input and clear the textbox
+    submit_btn.click(handle_input, inputs=[user_input, state], outputs=[chat, state]).then(lambda: "", None, user_input)
+
+demo.launch(share=True)