AI-300 Practice Question 24: Automated ML Baseline
Microsoft Machine Learning Operations Engineer Associate (AI-300) · Test 3 · question 4 of 10
A business analyst with limited machine learning background needs a baseline classifier over a tabular dataset by the end of the week. They want the service to try several algorithms and preprocessing options on its own, rank the attempts by a metric they choose, and have every attempt tracked in the Azure Machine Learning workspace like any other job. What should they use?
Show answer and explanation
Answer:
Automated ML "automates the time-consuming, iterative tasks of machine learning model development" so that "data scientists, analysts, and developers can build machine learning models at scale with efficiency and productivity". You configure "the number of iterations over different models, hyperparameter settings, advanced preprocessing and featurization options, and the metrics to evaluate when determining the best model", and for automated ML experiments "featurization happens automatically" (option C). The trials run as jobs in the workspace, so they are tracked like any other experiment.
- A is wrong because a sweep job explores hyperparameters for one algorithm that you have already chosen and written a script for. It cannot compare different model families or preprocessing strategies. - B is wrong because it gets to the same comparison only after the analyst writes and maintains a component per algorithm, exactly the bespoke engineering work this scenario is trying to avoid. - D is wrong because splitting endpoint traffic compares models that are already deployed in production. It is a release strategy, not a way to discover a first baseline model.
Reference: What is automated ML? AutoML - Azure Machine Learning