app.py

import numpy as np # linear algebra
import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)
import pprint
import os
import ast
import gradio as gr
from gradio.themes.base import Base
import weaviate
from weaviate.embedded import EmbeddedOptions
from langchain_community.vectorstores import Weaviate
from langchain.prompts import ChatPromptTemplate
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain.schema import Document
from langchain_community.chat_models import ChatOpenAI
from kaggle_secrets import UserSecretsClient
from langchain.schema.runnable import RunnablePassthrough
from langchain.schema.output_parser import StrOutputParser
from langchain_core.messages import HumanMessage, SystemMessage

df = pd.read_csv('./RAW_recipes.csv')

# Variables
max_length = 231637 #total number of recipes aka rows
curr_len = 10000 # how much we want to process and embed

#Concatenate all rows into one string
curr_i = 0
recipe_info = []
for index, row in df.iterrows():
    if curr_i >= curr_len:
        break
    curr_i+=1
    name, id, minutes, contributor_id, submitted, tags, nutrition, n_steps, steps, description, ingredients, n_ingredients = row

    #convert to list
    nutrition = ast.literal_eval(nutrition)
    steps = ast.literal_eval(steps)
    
    #format nutrition
    nutrition_map = ["Calorie"," Total Fat", 'Sugar', 'Sodium', 'Protein', 'Saturated Fat', 'Total Carbohydrate']
    nutrition_labeled = []
    for label, num in zip(nutrition_map, nutrition):
        if label == "Calorie":
            nutrition_labeled.append(f"{label} : {num} per serving")
        else:
            nutrition_labeled.append(f"{label} : {num} % daily value")

    #format steps
    for i in range(len(steps)):
        steps[i] = f"{i+1}. " + steps[i]
    recipe_info.append(f'''
    {name} : {minutes} minutes, submitted on {submitted} 
    description: {description},
    ingredients: {ingredients}
    number of ingredients: {n_ingredients}
    tags: {tags}, nutrition: {nutrition_labeled}, total steps: {n_steps}
    steps: {steps}
    '''.replace("\r", "").replace("\n", ""))


text_splitter = RecursiveCharacterTextSplitter(chunk_size=1500, chunk_overlap=150)

#split into recipe_info into chunks
docs = []
for doc in recipe_info:
    # Wrap each string in a Document object
    document = Document(page_content=doc)  # create a Document object with the content
    chunk = text_splitter.split_documents([document])  # Pass a list of Document objects
    docs.append(chunk)

# merge all chunks into one 
merged_documents = []
for doc in docs:
  merged_documents.extend(doc)

# Hugging Face model for embeddings.
model_name = "sentence-transformers/all-MiniLM-L6-v2"
model_kwargs = {'device': 'cpu'}
embeddings = HuggingFaceEmbeddings(
    model_name=model_name,
    model_kwargs=model_kwargs,
)

#initialize weaviate client
client = weaviate.Client(
  embedded_options = EmbeddedOptions()
)


vector_search = Weaviate.from_documents(
    client = client,
    documents = merged_documents,
    embedding = embeddings,
    by_text = False
)


# Instantiate Weaviate Vector Search as a retriever

# Basic RAG.
# k to search for only the 25 most relevant documents.
# score_threshold to use only documents with a relevance score above 0.77.
k = 10
score_threshold = 0.77

retriever = vector_search.as_retriever(
   search_type = "mmr", 
   search_kwargs = {
      "k": k,
      "score_threshold": score_threshold
   }
)

template = """
You are an assistant for question-answering tasks.
Use the following pieces of retrieved context to answer the question at the end.
The following pieces of retrieved context are recipes.
If you don't know the answer, just say that you don't know. Don't try to make up an answer.
Dont say anthing mean or offensive.

Context: {context}

Question: {question}
"""

custom_rag_prompt = ChatPromptTemplate.from_template(template)

llm = ChatOpenAI(
    model_name="gpt-3.5-turbo",
    temperature=0.2)

# Regular chain format: chain = prompt | model | output_parser
rag_chain = (
    {"context": retriever,  "question": RunnablePassthrough()}
    | custom_rag_prompt
    | llm
    | StrOutputParser()
)


def get_response(query):
  return rag_chain.invoke(query)


with gr.Blocks(theme=Base(), title="RAG Recipe AI") as demo:
    gr.Markdown("""
        # RAG Recipe AI
        
        This model will answer all your recipe-related questions. 
        Enter a question about a recipe, and the system will return an answer based on 10,000 food.com recipes stored in the vector database. \n
        Features Considered: \n
        \t - Cook Time
        \t - Nutrition Information
        \t - Steps
        \t - Ingredients
        \t - Dish Description 
        Sample Queries: \n
        \t - What is an easy dessert I can make with apples?
        \t - What is the nutritional information of a Caesar salad?
        \t - How many calories is in an average American burger?
    """)
    textbox = gr.Textbox(label="Question:")
    with gr.Row():
        button = gr.Button("Submit", variant="primary")
    with gr.Column():
        output1 = gr.Textbox(lines=1, max_lines=10, label="Answer:")
    # Call get_response function upon clicking the Submit button.
    button.click(get_response, textbox, outputs=[output1])

demo.launch()