app.py

import streamlit as st
import pandas as pd
import torch
from transformers import TapexTokenizer, BartForConditionalGeneration
import xml.etree.ElementTree as ET
from io import StringIO
import logging
from datetime import datetime
import time

# Configure logging
logging.basicConfig(
  level=logging.INFO,
  format='%(asctime)s - %(levelname)s - %(message)s'
)
logger = logging.getLogger(__name__)

@st.cache_resource
def load_model():
  """
  Load and cache the TAPEX model and tokenizer using Streamlit's caching
  """
  try:
      tokenizer = TapexTokenizer.from_pretrained(
          "microsoft/tapex-large-finetuned-wtq",
          model_max_length=1024
      )
      model = BartForConditionalGeneration.from_pretrained(
          "microsoft/tapex-large-finetuned-wtq"
      )
      device = 'cuda' if torch.cuda.is_available() else 'cpu'
      model = model.to(device)
      model.eval()
      return tokenizer, model
  except Exception as e:
      st.error(f"Error loading model: {str(e)}")
      return None, None

def parse_xml_to_dataframe(xml_string: str):
  """
  Parse XML string to DataFrame with error handling
  """
  try:
      tree = ET.parse(StringIO(xml_string))
      root = tree.getroot()

      data = []
      columns = set()

      # First pass: collect all possible columns
      for record in root.findall('.//record'):
          columns.update(elem.tag for elem in record)

      # Second pass: create data rows
      for record in root.findall('.//record'):
          row_data = {col: None for col in columns}
          for elem in record:
              row_data[elem.tag] = elem.text
          data.append(row_data)

      df = pd.DataFrame(data)

      # Convert numeric columns (automatically detect)
      for col in df.columns:
          try:
              df[col] = pd.to_numeric(df[col])
          except:
              continue

      return df, None
  except Exception as e:
      return None, f"Error parsing XML: {str(e)}"

def process_query(tokenizer, model, df, query: str):
  """
  Process a single query using the TAPEX model
  """
  try:
      start_time = time.time()

      # Handle direct DataFrame operations for common queries
      query_lower = query.lower()
      if "highest" in query_lower or "maximum" in query_lower:
          for col in df.select_dtypes(include=['number']).columns:
              if col.lower() in query_lower:
                  return df.loc[df[col].idxmax()].to_dict()
      elif "average" in query_lower or "mean" in query_lower:
          for col in df.select_dtypes(include=['number']).columns:
              if col.lower() in query_lower:
                  return f"Average {col}: {df[col].mean():.2f}"
      elif "total" in query_lower or "sum" in query_lower:
          for col in df.select_dtypes(include=['number']).columns:
              if col.lower() in query_lower:
                  return f"Total {col}: {df[col].sum():.2f}"

      # Use TAPEX for more complex queries
      with torch.no_grad():
          encoding = tokenizer(
              table=df.astype(str),
              query=query,
              return_tensors="pt",
              padding=True,
              truncation=True
          )
          outputs = model.generate(**encoding)
          answer = tokenizer.batch_decode(outputs, skip_special_tokens=True)[0]

      processing_time = time.time() - start_time
      return f"Answer: {answer} (Processing time: {processing_time:.2f}s)"

  except Exception as e:
      return f"Error processing query: {str(e)}"

def main():
  st.title("XML Data Query System")
  st.write("Upload your XML data and ask questions about it!")

  # Initialize session state for XML input and query if not exists
  if 'xml_input' not in st.session_state:
      st.session_state.xml_input = ""
  if 'current_query' not in st.session_state:
      st.session_state.current_query = ""

  # Load model
  with st.spinner("Loading TAPEX model... (this may take a few moments)"):
      tokenizer, model = load_model()
      if tokenizer is None or model is None:
          st.error("Failed to load the model. Please refresh the page.")
          return

  # XML Input
  xml_input = st.text_area(
      "Enter your XML data here:",
      value=st.session_state.xml_input,
      height=200,
      help="Paste your XML data here. Make sure it's properly formatted."
  )

  # Sample XML button
  if st.button("Load Sample XML"):
      st.session_state.xml_input = """<?xml version="1.0" encoding="UTF-8"?>
      <data>
          <records>
              <record>
                  <company>Apple</company>
                  <revenue>365.7</revenue>
                  <employees>147000</employees>
                  <year>2021</year>
              </record>
              <record>
                  <company>Microsoft</company>
                  <revenue>168.1</revenue>
                  <employees>181000</employees>
                  <year>2021</year>
              </record>
              <record>
                  <company>Amazon</company>
                  <revenue>386.1</revenue>
                  <employees>1608000</employees>
                  <year>2021</year>
              </record>
          </records>
      </data>"""
      st.rerun()

  if xml_input:
      df, error = parse_xml_to_dataframe(xml_input)
      if error:
          st.error(error)
      else:
          st.success("XML parsed successfully!")

          # Display DataFrame
          st.subheader("Parsed Data:")
          st.dataframe(df)

          # Query input
          query = st.text_input(
              "Enter your question about the data:",
              value=st.session_state.current_query,
              help="Example: 'Which company has the highest revenue?'"
          )

          # Process query
          if query:
              with st.spinner("Processing query..."):
                  result = process_query(tokenizer, model, df, query)
                  st.write(result)

          # Sample queries
          st.subheader("Sample Questions (Click to use):")
          sample_queries = [
              "Which company has the highest revenue?",
              "What is the average revenue of all companies?",
              "How many employees does Microsoft have?",
              "Which company has the most employees?",
              "What is the total revenue of all companies?"
          ]

          # Create columns for sample query buttons
          cols = st.columns(len(sample_queries))
          for idx, (col, sample_query) in enumerate(zip(cols, sample_queries)):
              with col:
                  if st.button(f"Query {idx + 1}", help=sample_query, key=f"query_btn_{idx}"):
                      st.session_state.current_query = sample_query
                      st.rerun()

          # Display the sample queries as text for reference
          with st.expander("View all sample questions"):
              for idx, query in enumerate(sample_queries, 1):
                  st.write(f"{idx}. {query}")

if __name__ == "__main__":
  main()