app.py

from typing import List
import pandas as pd
from transformers import AutoTokenizer, AutoModel
import torch
from langchain_community.document_loaders import PyPDFLoader
from IPython.display import display
import os
os.system('apt-get install poppler-utils')
from sklearn.metrics.pairwise import cosine_similarity
import numpy as np
import streamlit as st

class PDFProcessor:
    """
    Class for processing PDF files to extract text content.
    """
    def extract_text_from_pdfs(self, file_paths: List[str]) -> List[str]:
        """
        Extract text content from a list of PDF files.

        Args:
            file_paths (List[str]): A list of file paths to the PDF documents.

        Returns:
            List[str]: A list of text content extracted from the PDF documents.
        """
        texts = []
        for file_path in file_paths:
            try:
                loader = PyPDFLoader(file_path)
                pages = loader.load_and_split()

                for page in pages:
                    if isinstance(page.page_content, bytes):
                        text = page.page_content.decode('utf-8', errors='ignore')
                    elif isinstance(page.page_content, str):
                        text = page.page_content
                    else:
                        print(f"Unexpected type: {type(page.page_content)}")
                        continue
                    texts.append(text)
            except Exception as e:
                print(f"Failed to process {file_path}: {e}")

        return texts

class EmbeddingsProcessor:
    """
    Class for processing text to obtain embeddings using a transformer model.
    """
    def __init__(self, model_name: str):
        """
        Initialize the EmbeddingsProcessor with a pre-trained model.

        Args:
            model_name (str): The name of the pre-trained model to use for generating embeddings.
        """
        self.tokenizer = AutoTokenizer.from_pretrained(model_name)
        self.model = AutoModel.from_pretrained(model_name).to('cuda')

    def get_embeddings(self, texts: List[str]) -> np.ndarray:
        """
        Generate embeddings for a list of texts.

        Args:
            texts (List[str]): A list of text strings for which to generate embeddings.

        Returns:
            np.ndarray: A NumPy array of embeddings for the provided texts.
        """
        encoded_input = self.tokenizer(texts, padding=True, truncation=True, return_tensors="pt")
        encoded_input = {k: v.to('cuda') for k, v in encoded_input.items()}
        model_output = self.model(**encoded_input)
        return model_output.last_hidden_state.mean(dim=1).detach().cpu().numpy()

def compute_similarity(template_embeddings: np.ndarray, contract_embeddings: np.ndarray) -> np.ndarray:
    """
    Compute cosine similarity between template and contract embeddings.

    Args:
        template_embeddings (np.ndarray): A NumPy array of template embeddings.
        contract_embeddings (np.ndarray): A NumPy array of contract embeddings.

    Returns:
        np.ndarray: A NumPy array of similarity scores between contracts and templates.
    """
    return cosine_similarity(contract_embeddings, template_embeddings)

def clear_folder(path):
    if not os.path.exists(path):
        os.makedirs(path)  # Create the directory if it doesn't exist
    for file in os.listdir(path):
        file_path = os.path.join(path, file)
        try:
            if os.path.isfile(file_path):
                os.unlink(file_path)
        except Exception as e:
            print(f"Failed to delete {file_path}: {e}")

def save_uploaded_file(uploaded_file, path):
    try:
        with open(os.path.join(path, uploaded_file.name), "wb") as f:
            f.write(uploaded_file.getbuffer())
        return True
    except:
        return False

# Streamlit UI
st.title('PDF Similarity Checker')

col1, col2 = st.columns(2)

# Clear the templates and contracts folders before uploading new files
templates_folder = './templates'
contracts_folder = './contracts'

clear_folder(templates_folder)
clear_folder(contracts_folder)

with col1:
    st.header("Upload Templates")
    uploaded_files_templates = st.file_uploader("Choose PDF files", accept_multiple_files=True, type=['pdf'])
    os.makedirs(templates_folder, exist_ok=True)
    for uploaded_file in uploaded_files_templates:
        if save_uploaded_file(uploaded_file, templates_folder):
            st.write(f"Saved: {uploaded_file.name}")

with col2:
    st.header("Upload Contracts")
    uploaded_files_contracts = st.file_uploader("Choose PDF files", key="contracts", accept_multiple_files=True, type=['pdf'])
    os.makedirs(contracts_folder, exist_ok=True)
    for uploaded_file in uploaded_files_contracts:
        if save_uploaded_file(uploaded_file, contracts_folder):
            st.write(f"Saved: {uploaded_file.name}")

model_name = st.selectbox("Select Model", ['sentence-transformers/multi-qa-mpnet-base-dot-v1'], index=0)

if st.button("Compute Similarities"):
    pdf_processor = PDFProcessor()
    embedding_processor = EmbeddingsProcessor(model_name)

    # Process templates
    template_files = [os.path.join(templates_folder, f) for f in os.listdir(templates_folder)]
    template_texts = [pdf_processor.extract_text_from_pdfs([f])[0] for f in template_files if pdf_processor.extract_text_from_pdfs([f])]
    template_embeddings = embedding_processor.get_embeddings(template_texts)

    # Process contracts
    contract_files = [os.path.join(contracts_folder, f) for f in os.listdir(contracts_folder)]
    contract_texts = [pdf_processor.extract_text_from_pdfs([f])[0] for f in contract_files if pdf_processor.extract_text_from_pdfs([f])]
    contract_embeddings = embedding_processor.get_embeddings(contract_texts)

    # Compute similarities
    similarities = compute_similarity(template_embeddings, contract_embeddings)

    # Display results in a table format
    similarity_data = []
    for i, contract_file in enumerate(contract_files):
        row = [i + 1, os.path.basename(contract_file)]  # SI No and contract file name
        for j in range(len(template_files)):
            if j < similarities.shape[1] and i < similarities.shape[0]:  # Check if indices are within bounds
                row.append(f"{similarities[i, j] * 100:.2f}%")  # Format as percentage
            else:
                row.append("N/A")  # Handle out-of-bounds indices gracefully
        similarity_data.append(row)

    # Create a DataFrame for the table
    columns = ["SI No", "Contract"] + [os.path.basename(template_files[j]) for j in range(len(template_files))]
    similarity_df = pd.DataFrame(similarity_data, columns=columns)

    # Display maximize option
    if st.checkbox("Maximize Table View"):
        st.write("Similarity Scores Table (Maximized):")
        st.dataframe(similarity_df)  # Maximized view
    else:
        st.write("Similarity Scores Table:")
        st.table(similarity_df.style.hide(axis="index"))  # Normal view

    # Download option
    csv = similarity_df.to_csv(index=False).encode('utf-8')
    st.download_button(
        label="Download Similarity Table as CSV",
        data=csv,
        file_name='similarity_scores.csv',
        mime='text/csv',
    )