refactoring, predictions WIP
This commit is contained in:
Dobromir Popov
@ -2568,6 +2568,241 @@
|
||||
# except Exception as e:
|
||||
# logger.error(f"Error handling unified stream data: {e}")
|
||||
|
||||
# def _get_cnn_pivot_predictions(self, symbol: str, df: pd.DataFrame) -> List[Dict[str, Any]]:
|
||||
# """Get CNN model predictions for next pivot points"""
|
||||
# try:
|
||||
# predictions = []
|
||||
|
||||
# if not hasattr(self, 'orchestrator') or not self.orchestrator:
|
||||
# return predictions
|
||||
#
|
||||
# # Check if orchestrator has CNN capabilities
|
||||
# if hasattr(self.orchestrator, 'pivot_rl_trainer') and self.orchestrator.pivot_rl_trainer:
|
||||
# if hasattr(self.orchestrator.pivot_rl_trainer, 'williams') and self.orchestrator.pivot_rl_trainer.williams:
|
||||
# williams = self.orchestrator.pivot_rl_trainer.williams
|
||||
#
|
||||
# if hasattr(williams, 'cnn_model') and williams.cnn_model:
|
||||
# # Get latest market data for CNN input
|
||||
# if not df.empty and len(df) >= 900: # CNN needs at least 900 timesteps
|
||||
# try:
|
||||
# # Prepare multi-timeframe input for CNN
|
||||
# current_time = datetime.now()
|
||||
#
|
||||
# # Create dummy pivot point for CNN input preparation
|
||||
# dummy_pivot = type('SwingPoint', (), {
|
||||
# 'timestamp': current_time,
|
||||
# 'price': df['close'].iloc[-1],
|
||||
# 'index': len(df) - 1,
|
||||
# 'swing_type': 'prediction_point',
|
||||
# 'strength': 1
|
||||
# })()
|
||||
#
|
||||
# # Prepare CNN input using Williams structure
|
||||
# cnn_input = williams._prepare_cnn_input(
|
||||
# dummy_pivot,
|
||||
# df.values, # OHLCV data context
|
||||
# None # No previous pivot details
|
||||
# )
|
||||
#
|
||||
# if cnn_input is not None and cnn_input.size > 0:
|
||||
# # Reshape for batch prediction
|
||||
# if len(cnn_input.shape) == 2:
|
||||
# cnn_input = np.expand_dims(cnn_input, axis=0)
|
||||
#
|
||||
# # Get CNN prediction
|
||||
# pred_output = williams.cnn_model.model.predict(cnn_input, verbose=0)
|
||||
#
|
||||
# if pred_output is not None and len(pred_output) > 0:
|
||||
# # Parse CNN output (10 outputs for 5 Williams levels)
|
||||
# # Each level has [type_probability, predicted_price]
|
||||
# current_price = df['close'].iloc[-1]
|
||||
#
|
||||
# for level_idx in range(min(5, len(pred_output[0]) // 2)):
|
||||
# type_prob = pred_output[0][level_idx * 2]
|
||||
# price_offset = pred_output[0][level_idx * 2 + 1]
|
||||
#
|
||||
# # Determine prediction type
|
||||
# is_high = type_prob > 0.5
|
||||
# confidence = abs(type_prob - 0.5) * 2 # Convert to 0-1 range
|
||||
#
|
||||
# # Calculate predicted price
|
||||
# predicted_price = current_price + (price_offset * current_price * 0.01) # Assume price_offset is percentage
|
||||
#
|
||||
# # Only include predictions with reasonable confidence
|
||||
# if confidence > 0.3:
|
||||
# prediction = {
|
||||
# 'level': level_idx + 1,
|
||||
# 'type': 'HIGH' if is_high else 'LOW',
|
||||
# 'predicted_price': predicted_price,
|
||||
# 'confidence': confidence,
|
||||
# 'timestamp': current_time,
|
||||
# 'current_price': current_price,
|
||||
# 'price_offset_pct': price_offset * 100,
|
||||
# 'model_output': {
|
||||
# 'type_prob': float(type_prob),
|
||||
# 'price_offset': float(price_offset)
|
||||
# }
|
||||
# }
|
||||
# predictions.append(prediction)
|
||||
#
|
||||
# logger.debug(f"[CNN] Generated {len(predictions)} pivot predictions for {symbol}")
|
||||
#
|
||||
# except Exception as e:
|
||||
# logger.warning(f"Error generating CNN predictions: {e}")
|
||||
#
|
||||
# return predictions
|
||||
#
|
||||
# except Exception as e:
|
||||
# logger.error(f"Error getting CNN pivot predictions: {e}")
|
||||
# return []
|
||||
|
||||
# def _add_cnn_predictions_to_chart(self, fig: go.Figure, predictions: List[Dict[str, Any]], row: int = 1):
|
||||
# """Add CNN predictions as hollow circles to the chart"""
|
||||
# try:
|
||||
# if not predictions:
|
||||
# return
|
||||
#
|
||||
# # Separate HIGH and LOW predictions
|
||||
# high_predictions = [p for p in predictions if p['type'] == 'HIGH']
|
||||
# low_predictions = [p for p in predictions if p['type'] == 'LOW']
|
||||
#
|
||||
# # Add HIGH prediction markers (hollow red circles)
|
||||
# if high_predictions:
|
||||
# # Create future timestamps for display (predictions are for future points)
|
||||
# base_time = high_predictions[0]['timestamp']
|
||||
#
|
||||
# fig.add_trace(
|
||||
# go.Scatter(
|
||||
# x=[base_time + timedelta(minutes=i*5) for i in range(len(high_predictions))],
|
||||
# y=[p['predicted_price'] for p in high_predictions],
|
||||
# mode='markers',
|
||||
# marker=dict(
|
||||
# color='rgba(255, 107, 107, 0)', # Transparent fill
|
||||
# size=[max(8, min(20, p['confidence'] * 20)) for p in high_predictions],
|
||||
# symbol='circle',
|
||||
# line=dict(
|
||||
# color='#ff6b6b', # Red border
|
||||
# width=2
|
||||
# )
|
||||
# ),
|
||||
# name='CNN HIGH Predictions',
|
||||
# showlegend=True,
|
||||
# hovertemplate='<b>CNN HIGH Prediction</b><br>' +
|
||||
# 'Price: $%{y:.2f}<br>' +
|
||||
# 'Confidence: %{customdata:.1%}<br>' +
|
||||
# 'Level: %{text}<extra></extra>',
|
||||
# customdata=[p['confidence'] for p in high_predictions],
|
||||
# text=[f"Level {p['level']}" for p in high_predictions]
|
||||
# ),
|
||||
# row=row, col=1
|
||||
# )
|
||||
#
|
||||
# # Add LOW prediction markers (hollow green circles)
|
||||
# if low_predictions:
|
||||
# base_time = low_predictions[0]['timestamp']
|
||||
#
|
||||
# fig.add_trace(
|
||||
# go.Scatter(
|
||||
# x=[base_time + timedelta(minutes=i*5) for i in range(len(low_predictions))],
|
||||
# y=[p['predicted_price'] for p in low_predictions],
|
||||
# mode='markers',
|
||||
# marker=dict(
|
||||
# color='rgba(0, 255, 136, 0)', # Transparent fill
|
||||
# size=[max(8, min(20, p['confidence'] * 20)) for p in low_predictions],
|
||||
# symbol='circle',
|
||||
# line=dict(
|
||||
# color='#00ff88', # Green border
|
||||
# width=2
|
||||
# )
|
||||
# ),
|
||||
# name='CNN LOW Predictions',
|
||||
# showlegend=True,
|
||||
# hovertemplate='<b>CNN LOW Prediction</b><br>' +
|
||||
# 'Price: $%{y:.2f}<br>' +
|
||||
# 'Confidence: %{customdata:.1%}<br>' +
|
||||
# 'Level: %{text}<extra></extra>',
|
||||
# customdata=[p['confidence'] for p in low_predictions],
|
||||
# text=[f"Level {p['level']}" for p in low_predictions]
|
||||
# ),
|
||||
# row=row, col=1
|
||||
# )
|
||||
#
|
||||
# logger.debug(f"[CHART] Added {len(high_predictions)} HIGH and {len(low_predictions)} LOW CNN predictions to chart")
|
||||
#
|
||||
# except Exception as e:
|
||||
# logger.error(f"Error adding CNN predictions to chart: {e}")
|
||||
|
||||
# def _capture_actual_pivot_data(self, actual_pivot: Dict[str, Any]) -> None:
|
||||
# """Capture actual pivot data when it occurs for training comparison"""
|
||||
# try:
|
||||
# if not hasattr(self, '_pivot_training_data'):
|
||||
# self._pivot_training_data = []
|
||||
#
|
||||
# # Store actual pivot with timestamp for later comparison with predictions
|
||||
# pivot_data = {
|
||||
# 'actual_pivot': actual_pivot,
|
||||
# 'timestamp': datetime.now(),
|
||||
# 'captured_at': datetime.now().isoformat()
|
||||
# }
|
||||
#
|
||||
# self._pivot_training_data.append(pivot_data)
|
||||
#
|
||||
# # Keep only last 1000 actual pivots
|
||||
# if len(self._pivot_training_data) > 1000:
|
||||
# self._pivot_training_data = self._pivot_training_data[-1000:]
|
||||
#
|
||||
# logger.info(f"[TRAINING] Captured actual pivot: {actual_pivot['type']} at ${actual_pivot['price']:.2f}")
|
||||
#
|
||||
# # Save to persistent storage periodically
|
||||
# if len(self._pivot_training_data) % 10 == 0:
|
||||
# self._save_pivot_training_data()
|
||||
#
|
||||
# except Exception as e:
|
||||
# logger.error(f"Error capturing actual pivot data: {e}")
|
||||
|
||||
# def _save_pivot_training_data(self) -> None:
|
||||
# """Save pivot training data to JSON file for model improvement"""
|
||||
# try:
|
||||
# if not hasattr(self, '_pivot_training_data') or not self._pivot_training_data:
|
||||
# return
|
||||
#
|
||||
# # Create data directory if it doesn't exist
|
||||
# import os
|
||||
# os.makedirs('data/cnn_training', exist_ok=True)
|
||||
#
|
||||
# # Save to timestamped file
|
||||
# timestamp = datetime.now().strftime('%Y%m%d_%H%M%S')
|
||||
# filename = f'data/cnn_training/pivot_predictions_vs_actual_{timestamp}.json'
|
||||
#
|
||||
# # Prepare data for JSON serialization
|
||||
# save_data = {
|
||||
# 'metadata': {
|
||||
# 'created_at': datetime.now().isoformat(),
|
||||
# 'total_samples': len(self._pivot_training_data),
|
||||
# 'description': 'CNN pivot predictions compared with actual market pivots'
|
||||
# },
|
||||
# 'training_samples': []
|
||||
# }
|
||||
#
|
||||
# for sample in self._pivot_training_data:
|
||||
# # Convert datetime objects to ISO strings for JSON
|
||||
# json_sample = {
|
||||
# 'actual_pivot': sample['actual_pivot'],
|
||||
# 'timestamp': sample['timestamp'].isoformat() if isinstance(sample['timestamp'], datetime) else sample['timestamp'],
|
||||
# 'captured_at': sample['captured_at']
|
||||
# }
|
||||
# save_data['training_samples'].append(json_sample)
|
||||
#
|
||||
# # Write to file
|
||||
# import json
|
||||
# with open(filename, 'w') as f:
|
||||
# json.dump(save_data, f, indent=2, default=str)
|
||||
#
|
||||
# logger.info(f"[TRAINING] Saved {len(self._pivot_training_data)} pivot training samples to {filename}")
|
||||
#
|
||||
# except Exception as e:
|
||||
# logger.error(f"Error saving pivot training data: {e}")
|
||||
|
||||
# def create_scalping_dashboard(data_provider=None, orchestrator=None, trading_executor=None):
|
||||
# """Create real-time dashboard instance with MEXC integration"""
|
||||
# return RealTimeScalpingDashboard(data_provider, orchestrator, trading_executor)
|
||||
|
||||
Reference in New Issue
Block a user