app.py

import streamlit as st
import torch
import torch.nn as nn
from transformers import AutoTokenizer
import yfinance as yf
import pandas as pd
import numpy as np
from datetime import datetime, timedelta
import plotly.graph_objects as go
import plotly.express as px
from plotly.subplots import make_subplots
import time
import logging

# Configurazione pagina
st.set_page_config(
    page_title="Financial Transformer Analysis",
    page_icon="📈",
    layout="wide",
    initial_sidebar_state="expanded"
)

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

# Import delle classi dal modulo principale
@st.cache_resource
def load_model_components():
    """Carica i componenti del modello con cache"""
    
    class MultiLayerSemanticExtractor(nn.Module):
        def __init__(self, input_dim: int, hidden_dims: list, output_dim: int):
            super().__init__()
            self.layers = nn.ModuleList()
            
            prev_dim = input_dim
            for hidden_dim in hidden_dims:
                self.layers.append(nn.Sequential(
                    nn.Linear(prev_dim, hidden_dim),
                    nn.LayerNorm(hidden_dim),
                    nn.ReLU(),
                    nn.Dropout(0.1)
                ))
                prev_dim = hidden_dim
            
            self.output_layer = nn.Linear(prev_dim, output_dim)
            
        def forward(self, x):
            layer_outputs = []
            for layer in self.layers:
                x = layer(x)
                layer_outputs.append(x)
            final_output = self.output_layer(x)
            return final_output, layer_outputs

    class FinancialTransformer(nn.Module):
        def __init__(self, vocab_size=10000, d_model=512, nhead=8, num_layers=6, 
                     feature_dim=6, semantic_dims=[256, 128, 64]):
            super().__init__()
            
            self.d_model = d_model
            self.feature_dim = feature_dim
            
            self.embedding = nn.Embedding(vocab_size, d_model)
            self.pos_encoding = nn.Parameter(torch.randn(1000, d_model))
            
            encoder_layer = nn.TransformerEncoderLayer(
                d_model=d_model, nhead=nhead, dim_feedforward=d_model * 4,
                dropout=0.1, batch_first=True
            )
            self.transformer = nn.TransformerEncoder(encoder_layer, num_layers)
            
            self.semantic_extractor = MultiLayerSemanticExtractor(
                input_dim=feature_dim, hidden_dims=semantic_dims, output_dim=d_model
            )
            
            self.feature_projection = nn.Linear(d_model, d_model)
            self.price_predictor = nn.Linear(d_model, 1)
            self.trend_classifier = nn.Linear(d_model, 3)
            self.volatility_predictor = nn.Linear(d_model, 1)
            
        def forward(self, text_tokens, financial_features, attention_mask=None):
            batch_size, seq_len = text_tokens.shape
            
            text_emb = self.embedding(text_tokens)
            pos_emb = self.pos_encoding[:seq_len].unsqueeze(0).repeat(batch_size, 1, 1)
            text_emb = text_emb + pos_emb
            
            financial_emb, semantic_layers = self.semantic_extractor(financial_features)
            financial_emb = self.feature_projection(financial_emb)
            
            if len(financial_emb.shape) == 2:
                financial_emb = financial_emb.unsqueeze(1).repeat(1, seq_len, 1)
            
            combined_emb = text_emb + financial_emb
            transformer_output = self.transformer(combined_emb, src_key_padding_mask=attention_mask)
            
            if attention_mask is not None:
                mask_expanded = attention_mask.unsqueeze(-1).expand_as(transformer_output)
                transformer_output = transformer_output * mask_expanded
                pooled_output = transformer_output.sum(1) / mask_expanded.sum(1)
            else:
                pooled_output = transformer_output.mean(1)
            
            predictions = {
                'price_change': self.price_predictor(pooled_output),
                'trend': self.trend_classifier(pooled_output),
                'volatility': self.volatility_predictor(pooled_output),
                'semantic_layers': semantic_layers,
                'transformer_output': transformer_output
            }
            
            return predictions
    
    # Carica tokenizer
    tokenizer = AutoTokenizer.from_pretrained("distilbert-base-uncased")
    if tokenizer.pad_token is None:
        tokenizer.pad_token = tokenizer.eos_token
    
    # Inizializza modello
    model = FinancialTransformer(
        vocab_size=tokenizer.vocab_size,
        d_model=512,
        nhead=8,
        num_layers=6,
        feature_dim=6,
        semantic_dims=[256, 128, 64]
    )
    
    return model, tokenizer

def calculate_technical_indicators(data):
    """Calcola indicatori tecnici"""
    indicators = {}
    
    # Media mobile semplice
    indicators['sma_20'] = data['Close'].rolling(window=20).mean().fillna(data['Close'].mean())
    indicators['sma_50'] = data['Close'].rolling(window=50).mean().fillna(data['Close'].mean())
    
    # RSI
    delta = data['Close'].diff()
    gain = (delta.where(delta > 0, 0)).rolling(window=14).mean()
    loss = (-delta.where(delta < 0, 0)).rolling(window=14).mean()
    rs = gain / loss
    indicators['rsi'] = 100 - (100 / (1 + rs)).fillna(50)
    
    # Volatilità
    indicators['volatility'] = data['Close'].rolling(window=20).std().fillna(data['Close'].std())
    
    # Volume relativo
    indicators['volume_ratio'] = (data['Volume'] / data['Volume'].rolling(window=20).mean()).fillna(1)
    
    # Trend
    indicators['price_change'] = data['Close'].pct_change().fillna(0)
    
    return indicators

def extract_semantic_features(data, indicators):
    """Estrae features semantiche"""
    features = []
    
    # Normalizza i prezzi
    price_norm = (data['Close'] - data['Close'].mean()) / (data['Close'].std() + 1e-8)
    features.append(price_norm.values)
    
    # Aggiungi indicatori normalizzati
    for key, values in indicators.items():
        if key == 'rsi':
            normalized = (values - 50) / 50
        else:
            mean_val = values.mean()
            std_val = values.std()
            normalized = (values - mean_val) / (std_val + 1e-8)
        features.append(normalized.values)
    
    feature_matrix = np.column_stack(features)
    return feature_matrix

def create_market_context(symbol, data):
    """Crea contesto testuale"""
    if len(data) < 2:
        return f"Stock {symbol} trading data available."
    
    latest = data.iloc[-1]
    prev = data.iloc[-2]
    
    change = ((latest['Close'] - prev['Close']) / prev['Close']) * 100
    direction = "increased" if change > 0 else "decreased"
    
    context = f"Stock {symbol} has {direction} by {abs(change):.2f}% " \
             f"trading at ${latest['Close']:.2f} with volume {latest['Volume']:,}. " \
             f"High: ${latest['High']:.2f}, Low: ${latest['Low']:.2f}."
    
    return context

def analyze_symbol(symbol, model, tokenizer):
    """Analizza un simbolo"""
    try:
        # Recupera dati
        ticker = yf.Ticker(symbol)
        data = ticker.history(period="5d", interval="1m")
        
        if data.empty:
            return None
        
        # Calcola indicatori
        indicators = calculate_technical_indicators(data)
        
        # Estrai features
        features = extract_semantic_features(data, indicators)
        
        # Crea contesto
        context = create_market_context(symbol, data)
        
        # Tokenizza
        tokens = tokenizer(
            context, padding=True, truncation=True, 
            max_length=512, return_tensors="pt"
        )
        
        # Features finanziarie
        financial_features = torch.FloatTensor(features[-1:])
        
        # Predizione
        model.eval()
        with torch.no_grad():
            predictions = model(
                tokens['input_ids'],
                financial_features,
                attention_mask=tokens['attention_mask']
            )
        
        # Interpreta risultati
        price_change = predictions['price_change'].item()
        trend_probs = torch.softmax(predictions['trend'], dim=1)
        volatility = predictions['volatility'].item()
        
        trend_labels = ['Down', 'Stable', 'Up']
        predicted_trend = trend_labels[trend_probs.argmax().item()]
        
        return {
            'symbol': symbol,
            'current_price': data['Close'].iloc[-1],
            'predicted_price_change': price_change,
            'predicted_trend': predicted_trend,
            'trend_confidence': trend_probs.max().item(),
            'predicted_volatility': volatility,
            'market_context': context,
            'data': data,
            'indicators': indicators
        }
        
    except Exception as e:
        st.error(f"Errore nell'analisi di {symbol}: {str(e)}")
        return None

def create_price_chart(data, symbol):
    """Crea grafico dei prezzi"""
    fig = make_subplots(
        rows=2, cols=1,
        shared_xaxes=True,
        vertical_spacing=0.1,
        subplot_titles=(f'{symbol} Price', 'Volume'),
        row_width=[0.7, 0.3]
    )
    
    # Candlestick
    fig.add_trace(
        go.Candlestick(
            x=data.index,
            open=data['Open'],
            high=data['High'],
            low=data['Low'],
            close=data['Close'],
            name=symbol
        ),
        row=1, col=1
    )
    
    # Volume
    fig.add_trace(
        go.Bar(
            x=data.index,
            y=data['Volume'],
            name='Volume',
            marker_color='rgba(0,100,80,0.6)'
        ),
        row=2, col=1
    )
    
    fig.update_layout(
        title=f'{symbol} Real-Time Analysis',
        xaxis_title='Time',
        yaxis_title='Price ($)',
        height=600,
        showlegend=False
    )
    
    return fig

def create_indicators_chart(data, indicators):
    """Crea grafico degli indicatori"""
    fig = make_subplots(
        rows=2, cols=2,
        subplot_titles=('RSI', 'Moving Averages', 'Volatility', 'Volume Ratio')
    )
    
    # RSI
    fig.add_trace(
        go.Scatter(x=data.index, y=indicators['rsi'], name='RSI'),
        row=1, col=1
    )
    fig.add_hline(y=70, line_dash="dash", line_color="red", row=1, col=1)
    fig.add_hline(y=30, line_dash="dash", line_color="green", row=1, col=1)
    
    # Moving Averages
    fig.add_trace(
        go.Scatter(x=data.index, y=data['Close'], name='Close', line=dict(color='blue')),
        row=1, col=2
    )
    fig.add_trace(
        go.Scatter(x=data.index, y=indicators['sma_20'], name='SMA 20', line=dict(color='orange')),
        row=1, col=2
    )
    fig.add_trace(
        go.Scatter(x=data.index, y=indicators['sma_50'], name='SMA 50', line=dict(color='red')),
        row=1, col=2
    )
    
    # Volatility
    fig.add_trace(
        go.Scatter(x=data.index, y=indicators['volatility'], name='Volatility'),
        row=2, col=1
    )
    
    # Volume Ratio
    fig.add_trace(
        go.Scatter(x=data.index, y=indicators['volume_ratio'], name='Volume Ratio'),
        row=2, col=2
    )
    fig.add_hline(y=1, line_dash="dash", line_color="gray", row=2, col=2)
    
    fig.update_layout(height=600, showlegend=False)
    return fig

# Interfaccia principale
def main():
    st.title("📈 Financial Transformer Real-Time Analysis")
    st.markdown("---")
    
    # Sidebar
    st.sidebar.header("⚙️ Configuration")
    
    # Selezione simboli
    popular_symbols = ['AAPL', 'GOOGL', 'MSFT', 'TSLA', 'AMZN', 'META', 'NVDA']
    selected_symbols = st.sidebar.multiselect(
        "Select Symbols", 
        popular_symbols, 
        default=['AAPL', 'GOOGL', 'MSFT']
    )
    
    # Simbolo custom
    custom_symbol = st.sidebar.text_input("Custom Symbol (optional)")
    if custom_symbol:
        selected_symbols.append(custom_symbol.upper())
    
    # Parametri
    st.sidebar.subheader("Parameters")
    update_interval = st.sidebar.slider("Update Interval (seconds)", 30, 300, 60)
    show_charts = st.sidebar.checkbox("Show Charts", True)
    show_indicators = st.sidebar.checkbox("Show Technical Indicators", True)
    
    # Carica modello
    with st.spinner("Loading model..."):
        model, tokenizer = load_model_components()
    
    # Pulsante di analisi
    if st.sidebar.button("🚀 Start Analysis"):
        if not selected_symbols:
            st.error("Please select at least one symbol")
            return
        
        # Placeholder per risultati
        results_placeholder = st.empty()
        charts_placeholder = st.empty()
        
        # Loop di analisi
        for iteration in range(10):  # Limitato per demo
            st.subheader(f"Analysis Iteration {iteration + 1}")
            
            results = []
            
            # Analizza ogni simbolo
            for symbol in selected_symbols:
                with st.spinner(f"Analyzing {symbol}..."):
                    result = analyze_symbol(symbol, model, tokenizer)
                    if result:
                        results.append(result)
            
            if results:
                # Mostra risultati in tabella
                results_df = pd.DataFrame([{
                    'Symbol': r['symbol'],
                    'Current Price': f"${r['current_price']:.2f}",
                    'Predicted Change': f"{r['predicted_price_change']:.4f}",
                    'Trend': r['predicted_trend'],
                    'Confidence': f"{r['trend_confidence']:.2f}",
                    'Volatility': f"{r['predicted_volatility']:.4f}"
                } for r in results])
                
                st.table(results_df)
                
                # Mostra grafici se richiesto
                if show_charts:
                    cols = st.columns(len(results))
                    for i, result in enumerate(results):
                        with cols[i]:
                            st.plotly_chart(
                                create_price_chart(result['data'], result['symbol']),
                                use_container_width=True
                            )
                
                # Mostra indicatori tecnici
                if show_indicators:
                    st.subheader("📊 Technical Indicators")
                    for result in results:
                        st.subheader(f"{result['symbol']} Indicators")
                        st.plotly_chart(
                            create_indicators_chart(result['data'], result['indicators']),
                            use_container_width=True
                        )
            
            # Attendi prossimo update
            if iteration < 9:  # Non aspettare nell'ultima iterazione
                time.sleep(update_interval)
    
    # Informazioni
    st.sidebar.markdown("---")
    st.sidebar.markdown("### ℹ️ About")
    st.sidebar.markdown("""
    Questo strumento utilizza un transformer multi-layer per analizzare 
    dati finanziari in tempo reale e generare predizioni.
    
    **Features:**
    - Analisi semantica multi-layer
    - Indicatori tecnici avanzati
    - Predizioni trend e volatilità
    - Visualizzazioni interattive
    """)
    
    # Disclaimer
    st.sidebar.markdown("---")
    st.sidebar.markdown("### ⚠️ Disclaimer")
    st.sidebar.markdown("""
    **ATTENZIONE**: Questo strumento è solo per scopi educativi. 
    Non costituisce consulenza finanziaria. Gli investimenti comportano rischi.
    """)

if __name__ == "__main__":
    main()