1 / 5
Fill in: 'First we ___ models on the labelled dataset.'
-
-
-
-
We 'train models' — the standard ML collocation. 'Teach', 'coach' and 'school' apply to people, not models.
2 / 5
Fill in: 'Next we ___ performance against a held-out set.'
-
-
-
-
We 'evaluate performance' — the standard collocation. 'Judge', 'grade' and 'mark' are less idiomatic in ML evaluation.
3 / 5
Fill in: 'Once it passes, we ___ the model to production.'
-
-
-
-
We 'promote to production' — the standard MLOps collocation. 'Lift', 'push' and 'raise' don't carry the promotion sense here.
4 / 5
Fill in: 'In production we continuously ___ drift.'
-
-
-
-
We 'monitor drift' — the standard collocation. 'Watch', 'see' and 'view' are weaker and less idiomatic with drift.
5 / 5
Fill in: 'When accuracy degrades we ___ the model.'
-
-
-
-
We 'retire models' — the standard lifecycle collocation for decommissioning. 'Kill', 'stop' and 'end' are blunter and less idiomatic.