AI-300 Practice Question 48: Evaluation Data Mapping Columns
Microsoft Machine Learning Operations Engineer Associate (AI-300) · Test 5 · question 8 of 10
A team has an evaluation dataset whose columns are named user_question, bot_answer and source_docs. The Foundry evaluator they want to run expects the inputs query, response and context. They do not want to rewrite the dataset. What should they configure?
Show answer and explanation
Answer:
In a Foundry evaluation run, the testing criteria can include a data_mapping that connects dataset columns to evaluator inputs. The documentation shows the {{item.<field_name>}} syntax for this, for example "query": "{{item.query}}". For this dataset, the mapping query: {{item.user_question}}, response: {{item.bot_answer}} and context: {{item.source_docs}} connects the columns to the evaluator's inputs without changing the data (option A).
- B is wrong because the built-in evaluators define their own input names, and creating a new evaluator version to match the columns is unnecessary work compared with mapping the columns. - C is wrong because a renaming pipeline adds a step to maintain and a second copy of the data, which the mapping avoids. - D is wrong because a deployment parameter has no role in evaluation inputs. Data mapping is defined per evaluator in the evaluation configuration.
Reference: Custom Evaluators - Microsoft Foundry