app.py

import streamlit as st
import pandas as pd
import json
import numpy as np
import pinecone
from sentence_transformers import SentenceTransformer
import time

#pinecone.init(api_key='f5112f8c-f27d-4af1-b427-0c0953c113b5', environment='asia-southeast1-gcp')

pinecone.init(api_key='5c5b5687-b73d-47e9-9cc8-e184ff72cc45', environment='us-central1-gcp')

def process_string(s):
    return s.lower().replace('&', 'and')

index = pinecone.Index('ingradientsearch')

@st.cache_data
def load_model():
    return SentenceTransformer(r"finetiuned_model")

from concurrent.futures import ThreadPoolExecutor

def process_embedding(ingredient, model):
    processed_ingredient = process_string(ingredient)
    return model.encode([processed_ingredient]).tolist()

def pinecone_query(xq, index, top_k=1, includeMetadata=True):
    return index.query(xq, top_k=top_k, includeMetadata=includeMetadata)

def get_top_matches(ingredients):
    loaded_model = load_model()
    matches = []
    scores = []

    # Generate embeddings in parallel
    with ThreadPoolExecutor() as executor:
        embeddings = list(executor.map(lambda ing: process_embedding(ing, loaded_model), ingredients))

    # Query Pinecone in parallel
    results = []
    with ThreadPoolExecutor() as executor:
        results = list(executor.map(lambda xq: pinecone_query(xq, index), embeddings))

    # Extract matches and scores
    for result in results:
        if result['matches']:
            match = result['matches'][0]
            matches.append(match['metadata']['Ingredient'])
            scores.append(round(match['score'], 2))

    return matches, scores


def main():
    st.set_page_config(page_title="Ingredients Matching App", page_icon=":smiley:", layout="wide")
    st.title("Ingredients name matching App :smiley:")

    st.header("Matches using embeddings (semantic search)")
    st.write("Enter the JSON input:")
    json_input = st.text_area("")

    if st.button("Process"):
        start_time = time.time()
        with st.spinner("Processing..."):
            try:
                input_data = json.loads(json_input)
                for menu_item in input_data:
                    ingredients = menu_item.get("ingredients", [])
                    matches, scores = get_top_matches(ingredients)
                    menu_item["Ingradients_matched"] = matches
                    menu_item["scores"] = scores
    
                st.write("Processed JSON:")
                #st.write(json.dumps(input_data, indent=2))
                st.write("<pre>" + json.dumps(input_data, indent=4) + "</pre>", unsafe_allow_html=True)
            except json.JSONDecodeError:
                st.error("Invalid JSON input. Please check and try again.")

        end_time = time.time()
        st.write(f"Processing time: {end_time - start_time:.2f} seconds")

if __name__ == "__main__":
    main()