Q: 6
Your team is working on an NLP model and has just operationalized the first model. Your team makes
updates to the model, overwrites the original model, and puts this new model into operation.
However, one of the teams using the model has seen a decrease in performance and is asking to use
the original model.
What critical error did your team make?
Options
Discussion
Option B
Makes sense to pick B, not having model versioning means you can't recover the original when needed.
A or D are tempting but B is the actual issue. Without model versioning, rollback's not an option and that's what the user needed here. Similar question came up in practice, always points to B.
I don’t think it’s D. B is the issue, since without model versioning you can't restore the original when a new model underperforms. D can be tempting, but iteration doesn’t guarantee you'll keep previous deployable models. Seen similar questions in practice exams-pretty sure B is correct here, but feel free to challenge if you see it differently.
Why does PMI keep using tricky wording here? C or D
B not A or D
My pick: C from similar practice, official guide mentions retraining pipelines often.
B , not versioning means you can't go back to the original model easily. Classic ops mistake.
B Quick question though: if the requirement had been to automate reverting to previous models, would C make more sense?
D imo. If the team just replaced the old model right away, it points to skipping proper model iteration and testing before making changes live. You want to evaluate updates in stages, not overwrite directly. I could be missing something from a governance angle though, open to other takes.
Be respectful. No spam.
