Phoenix21's picture
added rule based feedback system
80a8bc4
raw
history blame
13.8 kB
import os
import logging
import re
from langchain.vectorstores import Chroma
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnablePassthrough
from langchain_huggingface import HuggingFaceEmbeddings
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_groq import ChatGroq
from langchain.schema import Document
from langchain.prompts import PromptTemplate
from langchain.chains import RetrievalQA
import chardet
import gradio as gr
import pandas as pd
import json
# Enable logging for debugging
logging.basicConfig(level=logging.INFO) # Changed to INFO to reduce verbosity
logger = logging.getLogger(__name__)
# Function to clean the API key
def clean_api_key(key):
return ''.join(c for c in key if ord(c) < 128)
# Load the GROQ API key from environment variables (set as a secret in the Space)
api_key = os.getenv("GROQ_API_KEY")
if not api_key:
logger.error("GROQ_API_KEY environment variable is not set. Please add it as a secret.")
raise ValueError("GROQ_API_KEY environment variable is not set. Please add it as a secret.")
api_key = clean_api_key(api_key).strip() # Clean and strip whitespace
# Function to clean text by removing non-ASCII characters
def clean_text(text):
return text.encode("ascii", errors="ignore").decode()
# Function to load and clean documents from multiple file formats
def load_documents(file_paths):
docs = []
for file_path in file_paths:
ext = os.path.splitext(file_path)[-1].lower()
try:
if ext == ".csv":
# Handle CSV files
with open(file_path, 'rb') as f:
result = chardet.detect(f.read())
encoding = result['encoding']
data = pd.read_csv(file_path, encoding=encoding)
for index, row in data.iterrows():
content = clean_text(row.to_string())
docs.append(Document(page_content=content, metadata={"source": file_path}))
elif ext == ".json":
# Handle JSON files
with open(file_path, 'r', encoding='utf-8') as f:
data = json.load(f)
if isinstance(data, list):
for entry in data:
content = clean_text(json.dumps(entry))
docs.append(Document(page_content=content, metadata={"source": file_path}))
elif isinstance(data, dict):
content = clean_text(json.dumps(data))
docs.append(Document(page_content=content, metadata={"source": file_path}))
elif ext == ".txt":
# Handle TXT files
with open(file_path, 'r', encoding='utf-8') as f:
content = clean_text(f.read())
docs.append(Document(page_content=content, metadata={"source": file_path}))
else:
logger.warning(f"Unsupported file format: {file_path}")
except Exception as e:
logger.error(f"Error processing file {file_path}: {e}")
logger.debug("Exception details:", exc_info=True)
return docs
# Function to ensure the response ends with complete sentences
def ensure_complete_sentences(text):
# Use regex to find all complete sentences
sentences = re.findall(r'[^.!?]*[.!?]', text)
if sentences:
# Join all complete sentences to form the complete answer
return ' '.join(sentences).strip()
return text # Return as is if no complete sentence is found
# Function to check if input is valid
def is_valid_input(text):
"""
Checks if the input text is meaningful.
Returns True if the text contains alphabetic characters and is of sufficient length.
"""
if not text or text.strip() == "":
return False
# Regex to check for at least one alphabetic character
if not re.search('[A-Za-z]', text):
return False
# Additional check: minimum length
if len(text.strip()) < 5:
return False
return True
# Initialize the LLM using ChatGroq with GROQ's API
def initialize_llm(model, temperature, max_tokens):
try:
# Allocate a portion of tokens for the prompt, e.g., 20%
prompt_allocation = int(max_tokens * 0.2)
response_max_tokens = max_tokens - prompt_allocation
if response_max_tokens <= 50:
raise ValueError("max_tokens is too small to allocate for the response.")
llm = ChatGroq(
model=model,
temperature=temperature,
max_tokens=response_max_tokens, # Adjusted max_tokens
api_key=api_key # Ensure the API key is passed correctly
)
logger.info("LLM initialized successfully.")
return llm
except Exception as e:
logger.error(f"Error initializing LLM: {e}")
raise
# Create the RAG pipeline
def create_rag_pipeline(file_paths, model, temperature, max_tokens):
try:
llm = initialize_llm(model, temperature, max_tokens)
docs = load_documents(file_paths)
if not docs:
logger.warning("No documents were loaded. Please check your file paths and formats.")
return None, "No documents were loaded. Please check your file paths and formats."
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
splits = text_splitter.split_documents(docs)
# Initialize the embedding model
embedding_model = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
# Use a temporary directory for Chroma vectorstore to prevent caching issues on Hugging Face Spaces
vectorstore = Chroma.from_documents(
documents=splits,
embedding=embedding_model,
persist_directory="/tmp/chroma_db" # Temporary storage directory
)
vectorstore.persist() # Save the database to disk
logger.info("Vectorstore initialized and persisted successfully.")
retriever = vectorstore.as_retriever()
custom_prompt_template = PromptTemplate(
input_variables=["context", "question"],
template="""
You are an AI assistant with expertise in daily wellness. Your aim is to provide detailed and comprehensive solutions regarding daily wellness topics without unnecessary verbosity.
Context:
{context}
Question:
{question}
Provide a thorough and complete answer, including relevant examples and a suggested schedule. Ensure that the response does not end abruptly.
"""
)
rag_chain = RetrievalQA.from_chain_type(
llm=llm,
chain_type="stuff",
retriever=retriever,
chain_type_kwargs={"prompt": custom_prompt_template}
)
logger.info("RAG pipeline created successfully.")
return rag_chain, "Pipeline created successfully."
except Exception as e:
logger.error(f"Error creating RAG pipeline: {e}")
logger.debug("Exception details:", exc_info=True)
return None, f"Error creating RAG pipeline: {e}"
# Define positive and negative words for rule-based sentiment analysis
POSITIVE_WORDS = {
"good", "great", "excellent", "amazing", "wonderful", "fantastic", "positive",
"helpful", "satisfied", "happy", "love", "liked", "enjoyed", "beneficial",
"superb", "awesome", "nice", "brilliant", "favorable", "pleased"
}
NEGATIVE_WORDS = {
"bad", "terrible", "awful", "poor", "disappointed", "unsatisfied", "hate",
"hated", "dislike", "dislikes", "worst", "negative", "not helpful", "frustrated",
"unhappy", "dissatisfied", "unfortunate", "horrible", "annoyed", "problem", "issues"
}
# Function to handle feedback with rule-based sentiment analysis
def handle_feedback(feedback_text):
"""
Handles user feedback by analyzing its sentiment and providing a dynamic response.
Stores the feedback in a temporary file for persistence during the session.
Parameters:
- feedback_text (str): The feedback provided by the user.
Returns:
- str: Acknowledgment message based on feedback sentiment.
"""
if feedback_text and feedback_text.strip() != "":
# Normalize feedback text to lowercase for comparison
feedback_lower = feedback_text.lower()
# Count positive and negative words
positive_count = sum(word in feedback_lower for word in POSITIVE_WORDS)
negative_count = sum(word in feedback_lower for word in NEGATIVE_WORDS)
# Determine sentiment based on counts
if positive_count > negative_count:
sentiment = "positive"
acknowledgment = "Thank you for your positive feedback! We're glad to hear that you found our service helpful."
elif negative_count > positive_count:
sentiment = "negative"
acknowledgment = "We're sorry to hear that you're not satisfied. Your feedback is valuable to us, and we'll strive to improve."
else:
sentiment = "neutral"
acknowledgment = "Thank you for your feedback. We appreciate your input."
# Log the feedback with sentiment
logger.info(f"User Feedback: {feedback_text} | Sentiment: {sentiment}")
# Optionally, store feedback in a temporary file
try:
with open("/tmp/user_feedback.txt", "a") as f:
f.write(f"{feedback_text} | Sentiment: {sentiment}\n")
logger.debug("Feedback stored successfully in /tmp/user_feedback.txt.")
except Exception as e:
logger.error(f"Error storing feedback: {e}")
return acknowledgment
else:
return "No feedback provided."
# Initialize the RAG pipeline once at startup
# Define the file paths (ensure 'AIChatbot.csv' is in the root directory of your Space)
file_paths = ['AIChatbot.csv']
model = "llama3-8b-8192" # Default model name
temperature = 0.7 # Default temperature
max_tokens = 500 # Default max tokens
rag_chain, message = create_rag_pipeline(file_paths, model, temperature, max_tokens)
if rag_chain is None:
logger.error("Failed to initialize RAG pipeline at startup.")
# Depending on your preference, you might want to exit or continue. Here, we'll continue.
# Function to answer questions with input validation and post-processing
def answer_question(model, temperature, max_tokens, question, feedback):
# Validate input
if not is_valid_input(question):
logger.info("Received invalid input from user.")
return "Please provide a valid question or input containing meaningful text.", ""
# Check if the RAG pipeline is initialized
if rag_chain is None:
logger.error("RAG pipeline is not initialized.")
return "The system is currently unavailable. Please try again later.", ""
try:
answer = rag_chain.run(question)
logger.info("Question answered successfully.")
# Post-process to ensure the answer ends with complete sentences
complete_answer = ensure_complete_sentences(answer)
# Handle feedback
feedback_response = handle_feedback(feedback)
return complete_answer, feedback_response
except Exception as e_inner:
logger.error(f"Error during RAG pipeline execution: {e_inner}")
logger.debug("Exception details:", exc_info=True)
return f"Error during RAG pipeline execution: {e_inner}", ""
# Gradio Interface with Feedback Mechanism
def gradio_interface(model, temperature, max_tokens, question, feedback):
# Optionally, you can add functionality to update the RAG pipeline if model or parameters change
# For now, we'll ignore changes to model parameters after initialization
return answer_question(model, temperature, max_tokens, question, feedback)
# Define Gradio UI
interface = gr.Interface(
fn=gradio_interface,
inputs=[
gr.Textbox(
label="Model Name",
value=model,
placeholder="e.g., llama3-8b-8192"
),
gr.Slider(
label="Temperature",
minimum=0,
maximum=1,
step=0.01,
value=temperature,
info="Controls the randomness of the response. Higher values like 0.8 will make the output more random, while lower values like 0.2 will make it more focused and deterministic."
),
gr.Slider(
label="Max Tokens",
minimum=200,
maximum=2048,
step=1,
value=max_tokens,
info="Determines the maximum number of tokens in the response. Higher values allow for longer answers."
),
gr.Textbox(
label="Question",
placeholder="e.g., What is box breathing and how does it help reduce anxiety?"
),
gr.Textbox(
label="Feedback",
placeholder="Provide your feedback here...",
lines=2
)
],
outputs=[
"text",
"text"
],
title="Daily Wellness AI",
description="Ask questions about daily wellness and get detailed solutions.",
examples=[
["llama3-8b-8192", 0.7, 500, "What is box breathing and how does it help reduce anxiety?", "Great explanation!"],
["llama3-8b-8192", 0.6, 600, "Provide a daily wellness schedule incorporating box breathing techniques.", "Very helpful, thank you!"]
],
allow_flagging="never" # Disable default flagging; using custom feedback
)
# Launch Gradio app without share=True (not supported on Hugging Face Spaces)
if __name__ == "__main__":
interface.launch(server_name="0.0.0.0", server_port=7860, debug=True)