AI-300 Practice Question 31: MLTable Multi Location Data Asset
Microsoft Machine Learning Operations Engineer Associate (AI-300) · Test 4 · question 1 of 10
A team trains a churn model on Parquet files that live in two different Azure Storage containers. Before every training run, the same steps must be applied: read both locations as one table, drop two unused columns, and keep only rows with a positive tenure. The team wants this loading logic stored once, and each registered version of the data must stay reproducible when new files arrive later. Which type of Azure Machine Learning data asset should they create?
Show answer and explanation
Answer:
Azure Machine Learning documents three data asset types. The Table (`mltable`) type is meant for when *"you have a complex schema subject to frequent changes, or you need a subset of large tabular data,"* and for *"unstructured data ... spread across multiple storage locations."* An MLTable file declares the paths and a list of transformations (for example filter and drop_columns), so the loading logic is stored once. The paths are declared in the file, so the documentation explains that a version keeps reading only what it lists, which *"ensures reproducibility"* when new files arrive; a new version declares the new paths (option C). Data asset versions are also immutable.
- A is wrong because a uri_file references a single file, while the data is spread across many files in two containers. - B is wrong because a uri_folder reads a folder of files as they are. It has no place to declare column drops or row filters, and it points at what is in the folder now, so new files change what a run reads. - D is wrong because a datastore URI is a valid way to start, but it leaves the filtering in each training script and gives no versioned, reusable definition of the data.
Reference: Create Data Assets - Azure Machine Learning