Training Workflow

Steps

1. Edit experiment name in train.run:

   # Example: "paper-2025-06"
   

2. Submit training job:

   sbatch train.run
   

3. Monitor in MLflow:

4. Experiment name: paper-2025-06_training
5. Each scenario logged with ID
6. Metrics, parameters, and artifacts tracked

Experiment Naming Convention

• Manual experiment names: paper-2025-06
• Training experiment: paper-2025-06_training
• Inpainting experiment: paper-2025-06_inpainting

Scenario Management

Scenarios defined in epiframework.py using dataclasses: - UNet architecture configuration - Dataset selection - Transform specifications - All parameters logged to MLflow