src/streamlit_app.py

# FinBERT Market Evaluation - Main Streamlit Application
"""
A confidence-aware, volatility-adjusted post-market evaluator for FinBERT sentiment
predictions against actual stock market movements.
"""

import streamlit as st
import pandas as pd
import numpy as np
import plotly.graph_objects as go
import plotly.express as px
from datetime import datetime, timedelta, date
import time
import logging

# Import our custom modules
from sentiment_analyzer import FinBERTAnalyzer
from market_data import MarketDataService
from evaluation import EvaluationEngine

# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Page configuration
st.set_page_config(
    page_title="FinBERT Market Evaluation",
    page_icon="🚀",
    layout="wide",
    initial_sidebar_state="expanded"
)

# Initialize session state for rate limiting
if 'last_request_time' not in st.session_state:
    st.session_state.last_request_time = 0

if 'evaluation_history' not in st.session_state:
    st.session_state.evaluation_history = []

# Initialize services
@st.cache_resource
def initialize_services():
    """Initialize all services with caching."""
    analyzer = FinBERTAnalyzer()
    market_service = MarketDataService()
    evaluation_engine = EvaluationEngine()
    return analyzer, market_service, evaluation_engine

def check_rate_limit():
    """Check if rate limit allows new request (30 seconds)."""
    current_time = time.time()
    time_since_last = current_time - st.session_state.last_request_time
    return time_since_last >= 30

def update_rate_limit():
    """Update the last request time."""
    st.session_state.last_request_time = time.time()

def create_das_chart(das_score: float, confidence: float, impact: float):
    """Create horizontal bar chart for DAS, confidence, and impact."""
    fig = go.Figure()

    metrics = ['DAS Score', 'Confidence', 'Impact (scaled)']
    values = [das_score, confidence, min(impact / 5.0, 1.0)]  # Scale impact to 0-1
    colors = ['#1f77b4', '#ff7f0e', '#2ca02c']

    fig.add_trace(go.Bar(
        y=metrics,
        x=values,
        orientation='h',
        marker_color=colors,
        text=[f'{v:.3f}' for v in values],
        textposition='inside'
    ))

    fig.update_layout(
        title="Evaluation Metrics",
        xaxis_title="Score",
        height=200,
        margin=dict(l=100, r=50, t=50, b=50)
    )

    return fig

def display_evaluation_result(result: dict):
    """Display comprehensive evaluation results."""
    if "error" in result:
        st.error(f"Evaluation Error: {result['error']}")
        return

    # Prominent evaluation summary first
    st.markdown(f"### {result['evaluation_summary']}")

    # Key insights in a highlighted box
    alignment_color = "green" if result['is_correct'] else "red"
    volatility_note = "🔥 Extremely High" if result['volatility_14d'] > 100 else "📊 High" if result['volatility_14d'] > 50 else "📈 Normal"

    # Calculate if movement was significant
    movement_significant = result['impact'] > result['threshold']
    significance_text = "exceeded" if movement_significant else "was below"

    st.markdown(f"""
    <div style="background-color: rgba(0,0,0,0.1); padding: 15px; border-radius: 10px; margin: 10px 0;">
    <h4>📊 Volatility-Aware Analysis:</h4>
    <ul>
    <li><strong>Stock's 14-day volatility:</strong> {result['volatility_14d']:.1f}% ({volatility_note.lower()})</li>
    <li><strong>Significance threshold:</strong> {result['threshold']:.1f}% (= 1.0 × volatility)</li>
    <li><strong>Actual movement:</strong> {result['return_24h']:+.2f}% ({result['impact']:.2f}% magnitude)</li>
    <li><strong>Movement significance:</strong> {significance_text} threshold → {'Significant' if movement_significant else 'Not significant'}</li>
    <li><strong>Directional alignment:</strong> <span style="color: {alignment_color};">{'✅ Correct direction' if result['is_correct'] else '❌ Wrong direction or insufficient magnitude'}</span></li>
    <li><strong>Model confidence:</strong> {'High' if result['confidence'] > 0.8 else 'Medium' if result['confidence'] > 0.6 else 'Low'} ({result['confidence']:.1%})</li>
    </ul>
    </div>
    """, unsafe_allow_html=True)

    # Main metrics in columns
    col1, col2, col3, col4 = st.columns(4)

    with col1:
        st.metric("DAS Score", f"{result['das_score']:.3f}", help="Directional Alignment Score (0-1, higher is better)")

    with col2:
        sentiment_emoji = {"positive": "📈", "negative": "📉", "neutral": "➡️"}
        st.metric("Sentiment", f"{sentiment_emoji.get(result['sentiment'], '❓')} {result['sentiment'].title()}")

    with col3:
        st.metric("Confidence", f"{result['confidence']:.1%}")

    with col4:
        return_color = "normal" if abs(result['return_24h']) < result['threshold'] else "inverse"
        st.metric("Same-Day Return", f"{result['return_24h']:+.2f}%", delta=f"vs {result['threshold']:.1f}% threshold")

    # Additional metrics for 24h return if available
    if result.get('return_next_24h') is not None:
        col1, col2, col3, col4 = st.columns(4)
        with col1:
            st.metric("Next 24h Return", f"{result['return_next_24h']:+.2f}%", help="Return from close of news day to close of next trading day")
        with col2:
            if result.get('alpha_24h') is not None:
                st.metric("24h Alpha", f"{result['alpha_24h']:+.2f}%", help="24h return vs market performance")
        with col3:
            # Show combined impact
            combined_impact = abs(result['return_24h']) + abs(result.get('return_next_24h', 0))
            st.metric("Combined Impact", f"{combined_impact:.2f}%", help="Total magnitude of price movement")
        with col4:
            # Show follow-through consistency
            same_direction = (result['return_24h'] * result.get('return_next_24h', 0)) > 0
            consistency = "✅ Consistent" if same_direction else "🔄 Reversal"
            st.metric("Follow-through", consistency, help="Whether 24h movement continued same direction")

    # Visualization
    chart = create_das_chart(result['das_score'], result['confidence'], result['impact'])
    # Use session state to create unique chart counter
    if 'chart_counter' not in st.session_state:
        st.session_state.chart_counter = 0
    st.session_state.chart_counter += 1
    chart_key = f"chart_{st.session_state.chart_counter}"
    st.plotly_chart(chart, use_container_width=True, key=chart_key)

    # Technical metrics (always visible)
    st.subheader("📊 Technical Metrics")

    col1, col2, col3 = st.columns(3)

    with col1:
        st.metric("Ticker", result['ticker'])
        st.metric("News Date", result['news_date'])
        st.metric("14-day Volatility", f"{result['volatility_14d']:.2f}%")
        st.metric("Significance Threshold", f"{result['threshold']:.2f}%")

    with col2:
        st.metric("Same-Day Impact", f"{result['impact']:.2f}%")
        if result.get('return_next_24h') is not None:
            st.metric("24h Impact", f"{abs(result['return_next_24h']):.2f}%")
        st.metric("WAT Weight", f"{result['wat_weight']:.3f}")
        alignment_text = "✅ Yes" if result['is_correct'] else "❌ No"
        st.metric("Alignment", alignment_text)

    with col3:
        alpha_val = result.get('alpha_adjusted', 'N/A')
        alpha_str = f"{alpha_val:+.2f}%" if isinstance(alpha_val, (int, float)) else str(alpha_val)
        st.metric("Same-Day Alpha", alpha_str)

        if result.get('alpha_24h') is not None:
            st.metric("24h Alpha", f"{result['alpha_24h']:+.2f}%")

        # Market context
        market_same = result.get('market_return', 'N/A')
        market_str = f"{market_same:+.2f}%" if isinstance(market_same, (int, float)) else str(market_same)
        st.metric("Market Return", market_str)

def main():
    """Main application function."""
    # Header
    st.title("🚀 FinBERT Market Evaluation")
    st.markdown("""
    A confidence-aware, volatility-adjusted post-market evaluator for FinBERT sentiment predictions.
    Evaluate how well FinBERT's financial news sentiment aligns with actual stock market movements.
    """)

    # Sidebar info (no user configuration needed)
    st.sidebar.header("📊 Evaluation Framework")
    st.sidebar.markdown("""
    **Dual-Period Analysis:**
    - **Same-Day**: Intraday return (Close - Open)
    - **Next 24h**: Close-to-close follow-through
    - **Combined**: Complete market reaction picture

    **Volatility-Aware Evaluation:**
    - Uses each stock's 14-day volatility
    - Threshold = 1.0 × volatility (k=1.0)
    - Adapts to stock movement patterns

    **Directional Alignment Score:**
    - Graded 0-1 score (not binary)
    - Based on same-day return vs threshold
    - Higher = better alignment

    **Alpha Analysis:**
    - Stock return vs market performance
    - Isolates stock-specific impact
    - Available for both time periods
    """)

    # Fixed research parameters (not user-configurable)
    volatility_multiplier = 1.0  # k = 1.0 as per your framework
    confidence_threshold = 0.7   # Reasonable default

    # Initialize services
    try:
        analyzer, market_service, evaluation_engine = initialize_services()
        evaluation_engine.volatility_multiplier = volatility_multiplier
        evaluation_engine.confidence_threshold = confidence_threshold
    except Exception as e:
        st.error(f"Failed to initialize services: {str(e)}")
        st.stop()

    # Main input form
    st.header("📰 News Analysis")

    with st.form("evaluation_form"):
        # News text input
        news_text = st.text_area(
            "Financial News Text",
            height=150,
            placeholder="Enter financial news headline or summary here...",
            help="Paste the financial news text you want to analyze"
        )

        col1, col2 = st.columns(2)

        with col1:
            ticker = st.text_input(
                "Stock Ticker",
                placeholder="e.g., TSLA, AAPL, MSFT",
                help="Enter the stock ticker symbol"
            ).upper()

        with col2:
            news_date = st.date_input(
                "News Publication Date",
                value=date.today() - timedelta(days=1),
                max_value=date.today() - timedelta(days=1),
                help="Date when the news was published (must be at least 1 day ago)"
            )

        submitted = st.form_submit_button("🔍 Evaluate Prediction")

    # Process evaluation
    if submitted:
        if not news_text.strip():
            st.error("Please enter some news text to analyze.")
            return

        if not ticker:
            st.error("Please enter a stock ticker symbol.")
            return

        # Rate limiting check
        if not check_rate_limit():
            remaining_time = 30 - (time.time() - st.session_state.last_request_time)
            st.warning(f"Rate limit: Please wait {remaining_time:.0f} more seconds before next request.")
            return

        # Update rate limit
        update_rate_limit()

        # Show progress
        progress_bar = st.progress(0)
        status_text = st.empty()

        try:
            # Step 1: Sentiment Analysis
            status_text.text("🤖 Analyzing sentiment with FinBERT...")
            progress_bar.progress(25)

            sentiment_result = analyzer.analyze_sentiment(news_text)

            # Step 2: Market Data
            status_text.text("📊 Fetching market data...")
            progress_bar.progress(50)

            news_datetime = datetime.combine(news_date, datetime.min.time())
            market_result = market_service.get_stock_evaluation_data(ticker, news_datetime)

            # Step 3: Evaluation
            status_text.text("⚖️ Evaluating prediction...")
            progress_bar.progress(75)

            evaluation_result = evaluation_engine.evaluate_prediction(
                sentiment_result, market_result, news_datetime
            )

            # Step 4: Display Results
            status_text.text("✅ Evaluation complete!")
            progress_bar.progress(100)

            # Clear progress indicators
            time.sleep(0.5)
            progress_bar.empty()
            status_text.empty()

            # Display results
            st.header("📊 Evaluation Results")
            display_evaluation_result(evaluation_result)

            # Add to history
            if "error" not in evaluation_result:
                st.session_state.evaluation_history.append(evaluation_result)

        except Exception as e:
            progress_bar.empty()
            status_text.empty()
            st.error(f"Evaluation failed: {str(e)}")
            logger.error(f"Evaluation error: {str(e)}")

    # Evaluation History Section
    if st.session_state.evaluation_history:
        st.header("📋 Previous Evaluations")

        # Show most recent evaluations first (reverse chronological)
        recent_evaluations = list(reversed(st.session_state.evaluation_history))

        # Show recent evaluations in expandable cards
        for i, result in enumerate(recent_evaluations):
            # Create a concise title for each evaluation
            alignment_icon = "✅" if result['is_correct'] else "❌"
            sentiment_icon = {"positive": "📈", "negative": "📉", "neutral": "➡️"}.get(result['sentiment'], "❓")

            title = f"{alignment_icon} {result['ticker']} ({result['news_date']}) - {sentiment_icon} {result['sentiment'].title()} → {result['return_24h']:+.1f}% | DAS: {result['das_score']:.3f}"

            with st.expander(title, expanded=(i==0)):  # Expand the most recent one
                display_evaluation_result(result)

        # Simple action buttons
        st.markdown("---")

        # Simple action buttons
        col1, col2 = st.columns([1, 3])

        with col1:
            if st.button("🗑️ Clear All History"):
                st.session_state.evaluation_history = []
                st.rerun()

        with col2:
            st.caption(f"📊 {len(st.session_state.evaluation_history)} evaluation(s) completed")

    # Footer
    st.markdown("---")
    st.caption("🚀 **FinBERT Market Evaluation** | Rate limit: 30s | Model: ProsusAI/finbert | Data: Yahoo Finance")

if __name__ == "__main__":
    main()