MLOps engineers use precise collocations throughout the model lifecycle. From monitoring drift to rolling back versions, this quiz covers the vocabulary of deploying and maintaining machine learning models in production.
0 / 5 completed
1 / 5
After training, the ML team is ready to ___ models to the production serving infrastructure.
Deploy models is the standard MLOps collocation for making a trained model available in a production environment. 'Release' is used for software versions. 'Push' is for code. 'Ship' is informal and startup-culture. Deploy models is the precise term used in MLOps documentation, kubeflow pipelines, and ML platform engineering.
2 / 5
Production models are observed continuously to ___ drift in input data distributions.
Monitor drift is the MLOps collocation for continuously observing changes in data or prediction distributions that signal model degradation. 'Detect' is a one-time action. 'Track' is for metrics over time. 'Catch' is informal. Monitor drift is the standard phrase in ML model observability, used by tools like Evidently AI and Arize.
3 / 5
When model performance degrades, the team schedules a job to ___ the model on fresh data.
Retrain periodically is the MLOps collocation for running the full training pipeline again on updated data to restore model accuracy. 'Train' is for the initial process. 'Update' is vague. 'Fine-tune' implies a partial update with a smaller dataset. Retrain specifically means restarting training to refresh a production model.
4 / 5
A/B testing and shadow scoring are used to ___ model performance before full rollout.
Evaluate performance is the standard ML collocation for systematically measuring model quality using defined metrics like precision, recall, or RMSE. 'Assess' is more qualitative. 'Test' is used in software. 'Measure' is a step within evaluation. Evaluate performance is the universal term in model validation and MLOps pipelines.
5 / 5
If a new model causes prediction errors in production, the team must quickly ___ to the previous version.
Roll back versions is the MLOps collocation for reverting a production model to a previously stable version when a new deployment fails. 'Revert' is used in version control for code. 'Return' is informal. 'Switch back' is conversational. Roll back is the precise operational term used in deployment pipelines, feature stores, and model registries.