Practice the key verb+noun collocations used when logging experiments, comparing runs, and managing models in the ML lifecycle in English.
0 / 5 completed
1 / 5
Fill in: 'Use MLflow to ___ every experiment so you can compare hyperparameters later.'
We 'log an experiment' — 'log' is the standard ML operations collocation for capturing experiment metadata, parameters, and artifacts. 'Save' is too generic; 'record' sounds manual; 'store' focuses on persistence, not the structured capture of experiment data.
2 / 5
Fill in: 'The platform automatically ___ precision and recall ___ for every training run.'
We 'track metrics' — 'track' is the dominant collocation for continuously observing performance indicators over time. 'Follow scores' sounds informal; 'monitor values' is used for infrastructure, not ML metrics; 'capture numbers' is vague.
3 / 5
Fill in: 'We need to ___ the last three runs to choose the best model for production.'
We 'compare runs' — 'compare' is the precise collocation for evaluating experiment results side by side. 'Review' implies narrative examination; 'check' is informal; 'analyse' focuses on a single entity rather than a head-to-head comparison.
4 / 5
Fill in: 'Once validation scores are acceptable, we ___ the model in the model registry.'
We 'register a model' — 'register' is the MLOps-standard verb for formally adding a model to a versioned registry for lifecycle management. 'Save' lacks the governance implication; 'push' is used for code; 'store' does not imply formal tracking.
5 / 5
Fill in: 'If the new model's accuracy drops below the threshold, we will ___ the previous stable version.'
We 'roll back a model version' — 'roll back' is the standard MLOps collocation for reverting to a previously known-good model. 'Restore' suits data backups; 'revert to' is correct but less idiomatic in ops contexts; 're-deploy' implies deploying the same version again, not switching.