app.py

import streamlit as st
import pandas as pd
import numpy as np
import joblib
import re
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.naive_bayes import MultinomialNB
from sklearn.linear_model import LogisticRegression
from sklearn.svm import LinearSVC
from sklearn.metrics import classification_report, accuracy_score
from sklearn.model_selection import train_test_split
import nltk
from nltk.corpus import stopwords
from nltk.stem import WordNetLemmatizer
import plotly.express as px
import plotly.graph_objects as go
from plotly.subplots import make_subplots

# Download required NLTK data
@st.cache_resource
def download_nltk_data():
    try:
        nltk.data.find('tokenizers/punkt')
        nltk.data.find('corpora/stopwords')
        nltk.data.find('corpora/wordnet')
    except LookupError:
        nltk.download('punkt', quiet=True)
        nltk.download('stopwords', quiet=True)
        nltk.download('wordnet', quiet=True)
        nltk.download('omw-1.4', quiet=True)

download_nltk_data()

# Initialize preprocessing tools
stop_words = set(stopwords.words('english'))
lemmatizer = WordNetLemmatizer()

def preprocess_text(text):
    """Clean and preprocess text for classification"""
    if pd.isna(text):
        return ""
    
    text = str(text).lower()
    text = re.sub(r'[^\w\s]', '', text)  # remove punctuation
    text = re.sub(r'\d+', '', text)      # remove digits
    
    words = text.split()
    words = [lemmatizer.lemmatize(word) for word in words if word not in stop_words and len(word) > 2]
    
    return ' '.join(words)

# Sample data for demonstration
@st.cache_data
def create_sample_data():
    """Create sample transaction data"""
    sample_data = [
        ("Monthly apartment rent payment", "rent"),
        ("Grocery shopping at walmart", "groceries"),
        ("Electric bill payment", "utilities"),
        ("Netflix monthly subscription", "subscription"),
        ("Gas station fuel", "transportation"),
        ("Restaurant dinner", "dining"),
        ("Apartment rent for december", "rent"),
        ("Weekly grocery shopping", "groceries"),
        ("Water bill payment", "utilities"),
        ("Spotify premium subscription", "subscription"),
        ("Bus fare to work", "transportation"),
        ("Coffee shop breakfast", "dining"),
        ("Monthly rent payment", "rent"),
        ("Food shopping at target", "groceries"),
        ("Internet bill", "utilities"),
        ("Amazon Prime membership", "subscription"),
        ("Uber ride home", "transportation"),
        ("Pizza delivery", "dining"),
        ("Rent for apartment", "rent"),
        ("Supermarket groceries", "groceries"),
        ("Phone bill payment", "utilities"),
        ("YouTube premium", "subscription"),
        ("Train ticket", "transportation"),
        ("Fast food lunch", "dining"),
        ("Office supplies", "shopping"),
        ("Medical appointment", "healthcare"),
        ("Gym membership", "fitness"),
        ("Book purchase", "shopping"),
        ("Doctor visit", "healthcare"),
        ("Fitness class", "fitness"),
        ("Clothing purchase", "shopping"),
        ("Pharmacy prescription", "healthcare"),
        ("Personal trainer", "fitness"),
        ("Electronics store", "shopping"),
        ("Dentist appointment", "healthcare"),
        ("Yoga class", "fitness"),
        ("Gift for friend", "shopping"),
        ("Eye exam", "healthcare"),
        ("Swimming pool fee", "fitness"),
        ("Home improvement", "shopping")
    ]
    
    df = pd.DataFrame(sample_data, columns=['purpose_text', 'transaction_type'])
    return df

@st.cache_resource
def train_models(df):
    """Train multiple models and return the best one"""
    # Preprocess data
    df['cleaned_purpose'] = df['purpose_text'].apply(preprocess_text)
    
    X = df["cleaned_purpose"]
    y = df["transaction_type"]
    
    # Train-test split
    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
    
    # TF-IDF Vectorization
    vectorizer = TfidfVectorizer(max_features=1000, ngram_range=(1, 2))
    X_train_vec = vectorizer.fit_transform(X_train)
    X_test_vec = vectorizer.transform(X_test)
    
    # Train models
    models = {
        "Naive Bayes": MultinomialNB(),
        "Logistic Regression": LogisticRegression(max_iter=1000, random_state=42),
        "SVM (LinearSVC)": LinearSVC(random_state=42)
    }
    
    results = {}
    trained_models = {}
    
    for name, model in models.items():
        model.fit(X_train_vec, y_train)
        y_pred = model.predict(X_test_vec)
        acc = accuracy_score(y_test, y_pred)
        results[name] = {
            'accuracy': acc,
            'predictions': y_pred,
            'actual': y_test
        }
        trained_models[name] = model
    
    # Find best model
    best_model_name = max(results, key=lambda x: results[x]['accuracy'])
    best_model = trained_models[best_model_name]
    
    return best_model, vectorizer, results, trained_models

def main():
    st.set_page_config(
        page_title="Transaction Classification System",
        page_icon="💳",
        layout="wide"
    )
    
    st.title("💳 Transaction Purpose Classification")
    st.markdown("---")
    
    # Sidebar
    st.sidebar.title("Navigation")
    page = st.sidebar.radio("Choose a page:", ["🏠 Home", "📊 Model Training", "🔍 Classification", "📈 Model Comparison"])
    
    # Load data
    df = create_sample_data()
    
    if page == "🏠 Home":
        st.header("Welcome to Transaction Classification System")
        
        col1, col2 = st.columns(2)
        
        with col1:
            st.subheader("📖 Project Overview")
            st.write("""
            This system classifies financial transactions based on their purpose text using machine learning.
            
            **Features:**
            - Multiple ML models (Naive Bayes, Logistic Regression, SVM)
            - Text preprocessing with NLTK
            - Interactive model comparison
            - Real-time transaction classification
            """)
            
        with col2:
            st.subheader("📊 Sample Data")
            st.dataframe(df.head(10))
            
            st.subheader("🏷️ Transaction Types")
            type_counts = df['transaction_type'].value_counts()
            fig = px.pie(values=type_counts.values, names=type_counts.index, title="Distribution of Transaction Types")
            st.plotly_chart(fig, use_container_width=True)
    
    elif page == "📊 Model Training":
        st.header("Model Training & Evaluation")
        
        # Train models
        with st.spinner("Training models..."):
            best_model, vectorizer, results, trained_models = train_models(df)
        
        col1, col2 = st.columns(2)
        
        with col1:
            st.subheader("📈 Model Performance")
            
            # Create results dataframe
            results_df = pd.DataFrame({
                'Model': list(results.keys()),
                'Accuracy': [results[model]['accuracy'] for model in results.keys()]
            })
            
            fig = px.bar(results_df, x='Model', y='Accuracy', title="Model Accuracy Comparison")
            fig.update_layout(yaxis_range=[0, 1])
            st.plotly_chart(fig, use_container_width=True)
            
            st.dataframe(results_df)
        
        with col2:
            st.subheader("🎯 Best Model Details")
            best_model_name = max(results, key=lambda x: results[x]['accuracy'])
            st.success(f"**Best Model:** {best_model_name}")
            st.metric("Accuracy", f"{results[best_model_name]['accuracy']:.3f}")
            
            # Classification report
            st.subheader("📋 Classification Report")
            y_test = results[best_model_name]['actual']
            y_pred = results[best_model_name]['predictions']
            
            report = classification_report(y_test, y_pred, output_dict=True)
            report_df = pd.DataFrame(report).transpose()
            st.dataframe(report_df.round(3))
        
        # Store models in session state
        st.session_state.best_model = best_model
        st.session_state.vectorizer = vectorizer
        st.session_state.trained_models = trained_models
    
    elif page == "🔍 Classification":
        st.header("Classify New Transaction")
        
        # Check if models are trained
        if 'best_model' not in st.session_state:
            st.warning("Please train the models first by visiting the 'Model Training' page.")
            return
        
        # Input form
        with st.form("classification_form"):
            purpose_text = st.text_area("Enter transaction purpose:", 
                                      placeholder="e.g., Monthly apartment rent payment",
                                      height=100)
            
            submitted = st.form_submit_button("Classify Transaction")
        
        if submitted and purpose_text:
            # Preprocess input
            cleaned_text = preprocess_text(purpose_text)
            
            # Make prediction
            vectorized_text = st.session_state.vectorizer.transform([cleaned_text])
            prediction = st.session_state.best_model.predict(vectorized_text)[0]
            prediction_proba = st.session_state.best_model.predict_proba(vectorized_text)[0]
            
            # Get class labels
            classes = st.session_state.best_model.classes_
            
            # Display results
            col1, col2 = st.columns(2)
            
            with col1:
                st.subheader("🎯 Classification Result")
                st.success(f"**Predicted Type:** {prediction}")
                st.info(f"**Original Text:** {purpose_text}")
                st.info(f"**Processed Text:** {cleaned_text}")
            
            with col2:
                st.subheader("📊 Prediction Confidence")
                proba_df = pd.DataFrame({
                    'Transaction Type': classes,
                    'Probability': prediction_proba
                }).sort_values('Probability', ascending=False)
                
                fig = px.bar(proba_df, x='Probability', y='Transaction Type', 
                           orientation='h', title="Prediction Probabilities")
                st.plotly_chart(fig, use_container_width=True)
    
    elif page == "📈 Model Comparison":
        st.header("Detailed Model Comparison")
        
        # Check if models are trained
        if 'trained_models' not in st.session_state:
            st.warning("Please train the models first by visiting the 'Model Training' page.")
            return
        
        # Model comparison
        st.subheader("🔍 Model Analysis")
        
        # Get sample predictions for comparison
        sample_texts = [
            "Monthly rent payment",
            "Grocery shopping",
            "Netflix subscription",
            "Gas station",
            "Restaurant dinner"
        ]
        
        comparison_data = []
        for text in sample_texts:
            cleaned = preprocess_text(text)
            vectorized = st.session_state.vectorizer.transform([cleaned])
            
            row = {'Text': text, 'Cleaned': cleaned}
            for model_name, model in st.session_state.trained_models.items():
                prediction = model.predict(vectorized)[0]
                row[model_name] = prediction
            
            comparison_data.append(row)
        
        comparison_df = pd.DataFrame(comparison_data)
        st.dataframe(comparison_df, use_container_width=True)
        
        # LLM/Transformer approach explanation
        st.subheader("🤖 Large Language Model Approach")
        
        with st.expander("Click to see LLM implementation strategy"):
            st.markdown("""
            ### Using Transformer Models for Transaction Classification
            
            **Approach:**
            1. **Pre-trained Model Selection**: Use `bert-base-uncased` or `distilbert-base-uncased`
            2. **Tokenization**: Use HuggingFace's tokenizer for the selected model
            3. **Model Architecture**: Add a classification head on top of the transformer
            4. **Fine-tuning**: Train on labeled transaction data
            
            **Code Example:**
            ```python
            from transformers import AutoTokenizer, AutoModelForSequenceClassification
            from transformers import Trainer, TrainingArguments
            
            # Load pre-trained model
            tokenizer = AutoTokenizer.from_pretrained('bert-base-uncased')
            model = AutoModelForSequenceClassification.from_pretrained(
                'bert-base-uncased', 
                num_labels=len(unique_labels)
            )
            
            # Tokenize data
            def tokenize_function(examples):
                return tokenizer(examples['purpose_text'], truncation=True, padding=True)
            
            # Fine-tune model
            training_args = TrainingArguments(
                output_dir='./results',
                num_train_epochs=3,
                per_device_train_batch_size=16,
                per_device_eval_batch_size=64,
                warmup_steps=500,
                weight_decay=0.01,
            )
            
            trainer = Trainer(
                model=model,
                args=training_args,
                train_dataset=train_dataset,
                eval_dataset=eval_dataset,
            )
            
            trainer.train()
            ```
            
            **Benefits:**
            - Better semantic understanding
            - Handles context better than TF-IDF
            - Can capture complex patterns
            - State-of-the-art performance
            
            **Drawbacks:**
            - Requires more computational resources
            - Longer training time
            - More complex deployment
            """)

if __name__ == "__main__":
    main()