Question 60
A training run fails after a teammate upgrades scikit-learn in a mutable notebook environment. The team can modify the relevant resource but must not broaden unrelated permissions or access. Before rollout, the engineer must confirm that a rerun records the approved package versions and reproduces the prior job. Which configuration should the team change first?
Explanation
Correct Answer
A reproducible environment fixes dependency compatibility independently of the user's current notebook state. A test run should show that a rerun records the approved package versions and reproduces the prior job.
Incorrect Answers
The unresolved environment reference can still select a dependency version different from the approved scikit-learn build.
Packages installed in a local Conda environment do not automatically appear in the remotely submitted Azure ML job.
Repairing a broken image build is not the reported failure. The requirement is a repeatable, pinned runtime for subsequent training jobs.