Question 50
Two training jobs use different dataset versions but the same run name. Which metadata should link predictions to the actual source data?
Explanation
Correct Answer
Explicit source versions make downstream predictions traceable even when experiment labels or models share friendly names.
Incorrect Answers
The alternative, Change the component input to a compatible uri_folder asset for the required directory structure, belongs to a different technical condition: A training task expects a directory containing nested subfolders, but its input asset uses uri_file and points to one manifest. There the source-specific mechanism or proof is: An input's asset type must match the component's expected file-versus-folder contract, not merely share a similar path name. The condition asked about here is: Two training jobs use different dataset versions but the same run name. A suitable correction is Record the specific registered data asset name and version with each run; the targeted justification or result is: Explicit source versions make downstream predictions traceable even when experiment labels or models share friendly names.
Register the data location as a governed Azure ML data asset with ownership and version metadata. That operation would be relevant if: A pipeline input points directly to an individual's personal blob URL, which is likely to disappear. Its corresponding reason or verification is: Managed data assets make a training input discoverable and reusable without depending on an employee's transient location. However the problem in this question is: Two training jobs use different dataset versions but the same run name. The requirement instead calls for Record the specific registered data asset name and version with each run; the question-specific check is: Explicit source versions make downstream predictions traceable even when experiment labels or models share friendly names.
Register the input as a versioned uri_file data asset. A separate source case supports that approach when: An Azure ML pipeline needs to pass one versioned parquet file rather than an entire mounted directory. The supporting rationale or observable result is: A uri_file represents a single addressed file; treating it as an unrestricted directory would misstate the component's expected input. This question describes a different issue: Two training jobs use different dataset versions but the same run name. Address it through Record the specific registered data asset name and version with each run, with the required justification or evidence: Explicit source versions make downstream predictions traceable even when experiment labels or models share friendly names.