gogo2/NN/utils/realtime_analyzer.py

"""
Realtime Analyzer for Neural Network Trading System

This module implements real-time analysis of market data using trained neural network models.
"""

import logging
import time
import numpy as np
from threading import Thread
from queue import Queue
from datetime import datetime

logger = logging.getLogger(__name__)

class RealtimeAnalyzer:
    """
    Handles real-time analysis of market data using trained neural network models.
    
    Features:
    - Connects to real-time data sources (websockets)
    - Processes incoming data through the neural network
    - Generates trading signals
    - Manages risk and position sizing
    - Logs all trading decisions
    """
    
    def __init__(self, data_interface, model, symbol="BTC/USDT", timeframes=None):
        """
        Initialize the realtime analyzer.
        
        Args:
            data_interface (DataInterface): Preconfigured data interface
            model: Trained neural network model
            symbol (str): Trading pair symbol
            timeframes (list): List of timeframes to monitor
        """
        self.data_interface = data_interface
        self.model = model
        self.symbol = symbol
        self.timeframes = timeframes or ['1h']
        self.running = False
        self.data_queue = Queue()
        self.prediction_interval = 60  # Seconds between predictions
        
        logger.info(f"RealtimeAnalyzer initialized for {symbol}")

    def start(self):
        """Start the realtime analysis process."""
        if self.running:
            logger.warning("Realtime analyzer already running")
            return
            
        self.running = True
        
        # Start data collection thread
        self.data_thread = Thread(target=self._collect_data, daemon=True)
        self.data_thread.start()
        
        # Start analysis thread
        self.analysis_thread = Thread(target=self._analyze_data, daemon=True)
        self.analysis_thread.start()
        
        logger.info("Realtime analysis started")

    def stop(self):
        """Stop the realtime analysis process."""
        self.running = False
        if hasattr(self, 'data_thread'):
            self.data_thread.join(timeout=1)
        if hasattr(self, 'analysis_thread'):
            self.analysis_thread.join(timeout=1)
        logger.info("Realtime analysis stopped")

    def _collect_data(self):
        """Thread function for collecting real-time data."""
        logger.info("Starting data collection thread")
        
        # In a real implementation, this would connect to websockets/API
        # For now, we'll simulate data collection from the data interface
        while self.running:
            try:
                # Get latest data for each timeframe
                for timeframe in self.timeframes:
                    # Get recent data (simulating real-time updates)
                    X, timestamp = self.data_interface.prepare_realtime_input(
                        timeframe=timeframe,
                        n_candles=30,
                        window_size=self.data_interface.window_size
                    )
                    
                    if X is not None:
                        self.data_queue.put({
                            'timeframe': timeframe,
                            'data': X,
                            'timestamp': timestamp
                        })
                
                # Throttle data collection
                time.sleep(1)
                
            except Exception as e:
                logger.error(f"Error in data collection: {str(e)}")
                time.sleep(5)  # Wait before retrying

    def _analyze_data(self):
        """Thread function for analyzing data and generating signals."""
        logger.info("Starting analysis thread")
        
        last_prediction_time = 0
        
        while self.running:
            try:
                current_time = time.time()
                
                # Only make predictions at the specified interval
                if current_time - last_prediction_time < self.prediction_interval:
                    time.sleep(0.1)
                    continue
                
                # Get latest data from queue
                if not self.data_queue.empty():
                    data_item = self.data_queue.get()
                    
                    # Make prediction
                    prediction = self.model.predict(data_item['data'])
                    
                    # Process prediction
                    self._process_prediction(
                        prediction=prediction,
                        timeframe=data_item['timeframe'],
                        timestamp=data_item['timestamp']
                    )
                    
                    last_prediction_time = current_time
                
                time.sleep(0.1)
                
            except Exception as e:
                logger.error(f"Error in analysis: {str(e)}")
                time.sleep(1)  # Wait before retrying

    def _process_prediction(self, prediction, timeframe, timestamp):
        """
        Process model prediction and generate trading signals.
        
        Args:
            prediction: Model prediction output
            timeframe (str): Timeframe the prediction is for
            timestamp: Timestamp of the prediction
        """
        # Convert prediction to trading signal
        signal = self._prediction_to_signal(prediction)
        
        # Log the signal
        logger.info(
            f"Signal generated - Timeframe: {timeframe}, "
            f"Timestamp: {timestamp}, "
            f"Signal: {signal}"
        )
        
        # In a real implementation, we would execute trades here
        # For now, we'll just log the signals

    def _prediction_to_signal(self, prediction):
        """
        Convert model prediction to trading signal.
        
        Args:
            prediction: Model prediction output
            
        Returns:
            str: Trading signal (BUY, SELL, HOLD)
        """
        # Simple threshold-based signal generation
        if len(prediction.shape) == 1:
            # Binary classification
            return "BUY" if prediction[0] > 0.5 else "SELL"
        else:
            # Multi-class classification (3 outputs)
            class_idx = np.argmax(prediction)
            return ["SELL", "HOLD", "BUY"][class_idx]