Question 80
A release must identify which version of a feature-engineering component generated a model's inputs. What metadata should be retained?
Explanation
Correct Answer
Component versions and job links make the source of engineered features auditable across later model deployments.
Incorrect Answers
Declare a component output and write the artifact to that output location. This is useful for the distinct issue: A pipeline's preprocessing step writes a model artifact to its temporary working directory, so the downstream training step cannot find it. The relevant technical rationale or evidence for that issue is: Declared outputs provide the handoff path between component steps; ephemeral local files are not automatic pipeline outputs. Here the reported issue is: A release must identify which version of a feature-engineering component generated a model's inputs. The relevant response is: Record the registered component name, version, inputs, and producing job in model lineage. The first result cannot verify this different correction: Component versions and job links make the source of engineered features auditable across later model deployments.
Publish machine-readable evaluation metrics as a declared component output asset. It addresses another case: A pipeline expects model metrics from an evaluation component, but the step prints them only to stdout. For that case, the distinguishing justification or acceptance check is: Structured outputs are accessible to dependent steps; console text alone is not a dependable typed pipeline contract. The present case instead states: A release must identify which version of a feature-engineering component generated a model's inputs. It calls for Record the registered component name, version, inputs, and producing job in model lineage. Showing success for the other case would not establish: Component versions and job links make the source of engineered features auditable across later model deployments.
The alternative, Separate preprocessing and training into independently parameterized pipeline components, belongs to a different technical condition: A pipeline repeatedly retrains with different model hyperparameters, but the preprocessing component stays unchanged. There the source-specific mechanism or proof is: Distinct components let parameter variations affect the correct stage without repeatedly editing unrelated preparation logic. The condition asked about here is: A release must identify which version of a feature-engineering component generated a model's inputs. A suitable correction is Record the registered component name, version, inputs, and producing job in model lineage; the targeted justification or result is: Component versions and job links make the source of engineered features auditable across later model deployments.