app.py

import streamlit as st
import torch
from sentence_transformers import SentenceTransformer, util
import pickle
import re

# Load the pre-trained SentenceTransformer model
model = SentenceTransformer('neuml/pubmedbert-base-embeddings')

# Load stored data
with open("embeddings_1.pkl", "rb") as fIn:
    stored_data = pickle.load(fIn)
    stored_embeddings = stored_data["embeddings"]

with open("embeddings_2.pkl", "rb") as fIn:
    stored_data_cpt = pickle.load(fIn)
    stored_embeddings_cpt = stored_data_cpt["embeddings"]

def validate_input(input_string):
    # Regular expression pattern to match letters and numbers, or letters only
    pattern = r'^[a-zA-Z0-9]+$|^[a-zA-Z]+$'
    
    # Check if input contains at least one non-numeric character
    if re.match(pattern, input_string) or input_string.isdigit():
        return True
    else:
        return False

def cpt_code(user_input):
    emb1 = model.encode(user_input.lower())
    similarities = []
    for sentence in stored_embeddings:
        similarity = util.cos_sim(sentence, emb1)
        similarities.append(similarity)

    # Filter results with similarity scores above 0.70
    result = [(code, desc, sim) for (code, desc, sim) in zip(stored_data["SBS_code"], stored_data["Description"], similarities)]

    # Sort results by similarity scores
    result.sort(key=lambda x: x[2], reverse=True)

    num_results = min(5, len(result))

    # Return top 5 entries with 'code', 'description', and 'similarity_score'
    top_5_results = []
    if num_results > 0:
        for i in range(num_results):
            code, description, similarity_score = result[i]
            top_5_results.append({"Code": code, "Description": description, "Similarity Score": similarity_score})
    else:
        top_5_results.append({"Code": "", "Description": "No match", "Similarity Score": 0.0})

    return top_5_results

def sbs_code(user_input):
    emb1 = model.encode(user_input.lower())
    similarities = []
    for sentence in stored_embeddings_cpt:
        similarity = util.cos_sim(sentence, emb1)
        similarities.append(similarity)

    # Filter results with similarity scores above 0.70
    result = [(code, desc, sim) for (code, desc, sim) in zip(stored_data_cpt["CPT_CODE"], stored_data_cpt["Description"], similarities)]

    # Sort results by similarity scores
    result.sort(key=lambda x: x[2], reverse=True)

    num_results = min(5, len(result))

    # Return top 5 entries with 'code', 'description', and 'similarity_score'
    top_5_results = []
    if num_results > 0:
        for i in range(num_results):
            code, description, similarity_score = result[i]
            top_5_results.append({"Code": code, "Description": description, "Similarity Score": similarity_score})
    else:
        top_5_results.append({"Code": "", "Description": "No match", "Similarity Score": 0.0})

    return top_5_results

def mapping_code(user_input, mode):
    if mode == "CPT_to_SBS":
        return cpt_code(user_input)
    elif mode == "SBS_to_CPT":
        return sbs_code(user_input)

# Streamlit frontend interface
def main():
    st.title("CPT-SBS Code Mapping")

    st.markdown("<font color='red'>**⚠️ Please enter the input CPT/SBS description with specific available  details for best results.**</font>", unsafe_allow_html=True)

    st.markdown("<font color='blue'>**💡 Note:** Please note that the similarity scores provided are not indicative of accuracy. Top 5 code descriptions provided should be verified with CPT/SBS descriptions by the user.</font>", unsafe_allow_html=True)


    # Dropdown for user to choose mapping direction
    mapping_mode = st.selectbox("Choose mapping direction:", ("CPT description to SBS code", "SBS description to CPT code"))

    if mapping_mode == "CPT description to SBS code":
        user_input_label = "Enter CPT description:"
        mode = "CPT_to_SBS"
    else:
        user_input_label = "Enter SBS description:"
        mode = "SBS_to_CPT"

    # Input text box for user input
    user_input = st.text_input(user_input_label, placeholder="Enter description here...")

    # Button to trigger mapping
    if st.button("Map"):
        if not user_input.strip():  # Check if input is empty or contains only whitespace
            st.error("Input box cannot be empty.")
        elif validate_input(user_input):
            st.warning("Please input correct description.")
        else:
            st.write("Please wait for a moment ...")
            # Call backend function to get mapping results
            try:
                mapping_results = mapping_code(user_input, mode)
                # Display top 5 similar sentences
                st.write("Top 5 similar entries:")
                for i, result in enumerate(mapping_results, 1):
                    st.write(f"{i}. Code: {result['Code']}, Description: {result['Description']}, Similarity Score: {float(result['Similarity Score']):.4f}")
            except ValueError as e:
                st.error(str(e))

if __name__ == "__main__":
    main()