73 lines
2.2 KiB
Markdown
73 lines
2.2 KiB
Markdown
# NO SIMULATION CODE POLICY
|
|
|
|
## CRITICAL RULE: NEVER CREATE SIMULATION CODE
|
|
|
|
**Date:** 2025-10-23
|
|
**Status:** PERMANENT POLICY
|
|
|
|
## What Was Removed
|
|
|
|
We deleted `ANNOTATE/core/training_simulator.py` which contained simulation/mock training code.
|
|
|
|
## Why This Is Critical
|
|
|
|
1. **Real Training Only**: We have REAL training implementations in:
|
|
- `NN/training/enhanced_realtime_training.py` - Real-time training system
|
|
- `NN/training/model_manager.py` - Model checkpoint management
|
|
- `core/unified_training_manager.py` - Unified training orchestration
|
|
- `core/orchestrator.py` - Core model training methods
|
|
|
|
2. **No Shortcuts**: Simulation code creates technical debt and masks real issues
|
|
3. **Production Quality**: All code must be production-ready, not simulated
|
|
|
|
## What To Use Instead
|
|
|
|
### For Model Training
|
|
Use the real training implementations:
|
|
|
|
```python
|
|
# Use EnhancedRealtimeTrainingSystem for real-time training
|
|
from NN.training.enhanced_realtime_training import EnhancedRealtimeTrainingSystem
|
|
|
|
# Use UnifiedTrainingManager for coordinated training
|
|
from core.unified_training_manager import UnifiedTrainingManager
|
|
|
|
# Use orchestrator's built-in training methods
|
|
orchestrator.train_models()
|
|
```
|
|
|
|
### For Model Management
|
|
```python
|
|
# Use ModelManager for checkpoint management
|
|
from NN.training.model_manager import ModelManager
|
|
|
|
# Use CheckpointManager for saving/loading
|
|
from utils.checkpoint_manager import get_checkpoint_manager
|
|
```
|
|
|
|
## If You Need Training Features
|
|
|
|
1. **Extend existing real implementations** - Don't create new simulation code
|
|
2. **Add to orchestrator** - Put training logic in the orchestrator
|
|
3. **Use UnifiedTrainingManager** - For coordinated multi-model training
|
|
4. **Integrate with EnhancedRealtimeTrainingSystem** - For online learning
|
|
|
|
## NEVER DO THIS
|
|
|
|
Create files with "simulator", "simulation", "mock", "fake" in the name
|
|
Use placeholder/dummy training loops
|
|
Return fake metrics or results
|
|
Skip actual model training
|
|
|
|
## ALWAYS DO THIS
|
|
|
|
Use real model training methods
|
|
Integrate with existing training systems
|
|
Save real checkpoints
|
|
Track real metrics
|
|
Handle real data
|
|
|
|
---
|
|
|
|
**Remember**: If data is unavailable, return None/empty/error - NEVER simulate it!
|