Question 180
A notebook generates a useful model but requires clicking cells in a particular manual order. What should be done before production use?
Explanation
Correct Answer
Manual cell execution can hide state and ordering assumptions that undermine repeatable scheduled production training.
Incorrect Answers
Version-control the notebook and record run metadata for each important experiment. It addresses another case: Multiple analysts edit the same notebook and cannot tell which transformation produced an unexpected plot. For that case, the distinguishing justification or acceptance check is: Source history and recorded runs distinguish changes in executed code from changes in the input data. The present case instead states: A notebook generates a useful model but requires clicking cells in a particular manual order. It calls for Refactor essential steps into a deterministic script or pipeline with explicit dependencies. Showing success for the other case would not establish: Manual cell execution can hide state and ordering assumptions that undermine repeatable scheduled production training.
The alternative, Clear sensitive outputs and review the notebook before committing it to source control, belongs to a different technical condition: A notebook is being promoted to a code-reviewed project, but its output cells contain client information. There the source-specific mechanism or proof is: Outputs can persist customer data in notebook files independently of the code that produced them. The condition asked about here is: A notebook generates a useful model but requires clicking cells in a particular manual order. A suitable correction is Refactor essential steps into a deterministic script or pipeline with explicit dependencies; the targeted justification or result is: Manual cell execution can hide state and ordering assumptions that undermine repeatable scheduled production training.
Use deterministic random seeds and log them with experiment metadata. That operation would be relevant if: A notebook's random sampling produces inconsistent model metrics across reruns, even when data and parameters are unchanged. Its corresponding reason or verification is: Controlled randomness helps distinguish real model changes from variation introduced by sampling or initialization. However the problem in this question is: A notebook generates a useful model but requires clicking cells in a particular manual order. The requirement instead calls for Refactor essential steps into a deterministic script or pipeline with explicit dependencies; the question-specific check is: Manual cell execution can hide state and ordering assumptions that undermine repeatable scheduled production training.