Experiment Tracking Quiz

Answers and explanations appear as you take the quiz.

1. What is the primary purpose of experiment tracking in machine learning?
   • To make runs reproducible and comparable by recording their full context
   • To make models train faster on the GPU
   • To automatically deploy models to production
   • To compress model weights for storage
2. Which combination is essential for making a single training run truly reproducible?
   • Only the final accuracy number
   • Code version, dataset, and hyperparameters
   • The GPU temperature and fan speed
   • The name of the data scientist
3. Which of these is a popular experiment tracking tool?
   • Photoshop
   • Kubernetes
   • MLflow
   • PostgreSQL
4. How do most experiment trackers typically capture metrics during training?
   • By manually copying numbers into a spreadsheet afterward
   • By taking screenshots of the terminal
   • By reading the model's source code
   • Through logging API calls inserted into the training loop that stream to a backend
5. Where are large artifacts like saved model weights usually stored by a tracking system?
   • In object/artifact storage, with references kept in the metadata store
   • Inside the metrics database row
   • In the GPU memory permanently
   • They are not stored at all
6. How is experiment tracking evolving in the era of large language models?
   • It is being abandoned because LLMs need no tracking
   • It now also captures prompt versions, evaluation traces, and qualitative outputs
   • It only tracks GPU costs now
   • It has merged entirely into video editing software
7. Why record a dataset version or content hash with an ML run?
   • To prove the model is unbiased
   • To guarantee identical GPU arithmetic
   • So the run can be linked to the exact data snapshot used
   • To replace the code version
8. In MLflow Tracking, where are large model files typically stored relative to parameters and metrics?
   • Only inside the metric names
   • Always embedded in the database row for each parameter
   • They are not recorded by a tracking system
   • In an artifact store, while run metadata is kept in a backend store