ML Experiment Tracking
Track and compare experiments.
What to Track
Parameters: hyperparameter values. Metrics: training/validation scores. Artifacts: models, visualizations. Code: version.
Tools
MLflow: tracking server, model registry. Weights & Biases: beautiful UI, collaboration. TensorBoard: for TensorFlow/PyTorch.
Best Practices
Structured naming. Consistent logging. Compare with parallel coordinates.
Key Takeaways
- Track parameters, metrics, artifacts
- MLflow, W&B, TensorBoard popular
- Enables experiment comparison