popov/gogo2
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

loss /performance display

Browse Source
This commit is contained in:
Dobromir Popov
2025-07-02 03:29:38 +03:00
parent 29e4076638
commit 11718bf92f
3 changed files with 128 additions and 48 deletions
Download Patch File Download Diff File Expand all files Collapse all files

51
web/clean_dashboard.py
View File

@ -3981,19 +3981,45 @@ class CleanTradingDashboard:
async def _on_trading_decision(self, decision):
"""Handle trading decision from orchestrator."""
try:
action = getattr(decision, 'action', decision.get('action'))
# Handle both object and dict formats
if hasattr(decision, 'action'):
action = getattr(decision, 'action', 'HOLD')
symbol = getattr(decision, 'symbol', 'ETH/USDT')
confidence = getattr(decision, 'confidence', 0.0)
else:
action = decision.get('action', 'HOLD')
symbol = decision.get('symbol', 'ETH/USDT')
confidence = decision.get('confidence', 0.0)
if action == 'HOLD':
return
symbol = getattr(decision, 'symbol', decision.get('symbol', 'ETH/USDT'))
if 'ETH' not in symbol.upper():
return
dashboard_decision = asdict(decision) if not isinstance(decision, dict) else decision.copy()
dashboard_decision['timestamp'] = datetime.now()
dashboard_decision['executed'] = False
# Convert to dict format for dashboard storage
if hasattr(decision, '__dict__'):
dashboard_decision = {
'action': action,
'symbol': symbol,
'confidence': confidence,
'timestamp': datetime.now(),
'executed': False
}
# Add any other attributes from the decision object
for attr in ['price', 'quantity', 'reasoning', 'model_source']:
if hasattr(decision, attr):
dashboard_decision[attr] = getattr(decision, attr)
else:
dashboard_decision = decision.copy()
dashboard_decision['timestamp'] = datetime.now()
dashboard_decision['executed'] = False
self.recent_decisions.append(dashboard_decision)
if len(self.recent_decisions) > 200:
self.recent_decisions.pop(0)
logger.info(f"[ORCHESTRATOR SIGNAL] Received: {action} for {symbol}")
logger.info(f"[ORCHESTRATOR SIGNAL] Received: {action} for {symbol} (confidence: {confidence:.3f})")
except Exception as e:
logger.error(f"Error handling trading decision: {e}")
@ -4523,6 +4549,10 @@ class CleanTradingDashboard:
network.train()
action_logits, predicted_confidence = network(features_tensor)
# Ensure predicted_confidence has a batch dimension if it doesn't already
if predicted_confidence.dim() == 0:
predicted_confidence = predicted_confidence.unsqueeze(0)
# Calculate losses
action_loss = nn.CrossEntropyLoss()(action_logits, action_target_tensor)
confidence_loss = nn.MSELoss()(predicted_confidence, confidence_target_tensor)
@ -4559,7 +4589,10 @@ class CleanTradingDashboard:
performance_metrics = {
'loss': avg_loss,
'training_samples': training_samples,
'model_parameters': sum(p.numel() for p in network.parameters())
'model_parameters': sum(p.numel() for p in network.parameters()),
'loss_improvement': 1.0 / (1.0 + avg_loss), # Higher is better
'training_iterations': loss_count,
'average_confidence': confidence_target if 'confidence_target' in locals() else 0.5
}
metadata = save_checkpoint(
@ -4577,7 +4610,9 @@ class CleanTradingDashboard:
logger.error(f"Error saving decision fusion checkpoint: {e}")
if training_samples > 0:
logger.info(f"DECISION TRAINING: Processed {training_samples} decision fusion samples")
avg_loss_info = f", avg_loss={total_loss/loss_count:.6f}" if loss_count > 0 else ""
performance_score = 100 / (1 + (total_loss/loss_count)) if loss_count > 0 else 0.1
logger.info(f"DECISION TRAINING: Processed {training_samples} decision fusion samples{avg_loss_info}, perf_score={performance_score:.4f}")
except Exception as e:
logger.error(f"Error in real decision fusion training: {e}")