app.py

import streamlit as st
from transformers import RagTokenizer, RagRetriever, RagTokenForGeneration, DPRQuestionEncoder, DPRQuestionEncoderTokenizer
import torch
import faiss
import numpy as np

# Streamlit app layout
st.title("Real Estate Recommendation System")
st.write("Use this app to predict property values, rent prices, and find matching properties for potential buyers.")

# Load models function
@st.cache_resource
def load_models():
    try:
        # Load RAG model components
        rag_tokenizer = RagTokenizer.from_pretrained("facebook/rag-token-nq")
        rag_retriever = RagRetriever.from_pretrained("facebook/rag-token-nq", index_name="exact", use_dummy_dataset=True)
        rag_model = RagTokenForGeneration.from_pretrained("facebook/rag-token-nq", retriever=rag_retriever)
        
        # Load DPR components for question encoding
        dpr_tokenizer = DPRQuestionEncoderTokenizer.from_pretrained("facebook/dpr-question_encoder-single-nq-base")
        dpr_model = DPRQuestionEncoder.from_pretrained("facebook/dpr-question_encoder-single-nq-base")

        return rag_tokenizer, rag_retriever, rag_model, dpr_tokenizer, dpr_model
    except Exception as e:
        st.error(f"Error loading models: {e}")
        st.stop()

# Load models
rag_tokenizer, rag_retriever, rag_model, dpr_tokenizer, dpr_model = load_models()

# User input for property details
st.sidebar.header("Enter Property Details:")
property_location = st.sidebar.text_input("Property Location", "San Francisco, CA")
property_size = st.sidebar.number_input("Property Size (sq ft)", 500, 10000, 1500)
bedrooms = st.sidebar.number_input("Number of Bedrooms", 1, 10, 3)
bathrooms = st.sidebar.number_input("Number of Bathrooms", 1, 5, 2)
budget = st.sidebar.number_input("Buyer's Budget ($)", 50000, 10000000, 500000)

# Property query
query = f"Find properties in {property_location} with {bedrooms} bedrooms, {bathrooms} bathrooms, and approximately {property_size} sq ft."

# Encoding the query using DPR
input_ids = dpr_tokenizer(query, return_tensors="pt")["input_ids"]
embeddings = dpr_model(input_ids).pooler_output

# Display the query
st.write(f"### Query: {query}")

# Using the RAG model for generating recommendations
rag_input_dict = rag_tokenizer.prepare_seq2seq_batch(query, return_tensors="pt")
generated_ids = rag_model.generate(input_ids=rag_input_dict["input_ids"], num_beams=2, max_length=100)
output = rag_tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]

# Display the recommendations
st.write("### Recommended Properties:")
st.write(output)