AI-900 Practice Question: Describe fundamental principles of machine learning on Azure

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  A batch prediction job that scores new data against a deployed model

  Why it's wrong here

  A batch prediction job, also called batch inference or batch scoring, processes a collection of new records against an already registered or deployed model and writes the resulting predictions to storage. Because it uses existing model weights to compute outputs, it is inference or scoring, not training. A training job is the separate execution that creates or updates those weights by minimizing a loss function over labeled examples and logs training metrics.

• ✓

  A single execution of a training script that produces a trained model and tracked metrics

  Why this is correct

  A training job is the core unit of model training in Azure ML: it is a single, tracked execution of a training script on a chosen compute target, and that script performs the actual fitting of a model to data. During the run, Azure ML captures parameters, logs, and metrics (for example via MLflow or run.log_metric), and when complete it produces a model artifact that can be registered or evaluated. This aligns exactly with the definition of a training job, distinguishing it from monitoring, preprocessing, and batch-scoring workloads.

• ✗

  A scheduled report on model performance in production

  Why it's wrong here

  A scheduled report on model performance in production is part of model monitoring, not model training. It evaluates telemetry such as data drift, prediction accuracy, or latency from a deployed endpoint over time, and it does not execute a training script or update model parameters. In Azure ML, this would be implemented as a recurring monitoring job or pipeline, not as a training job.

• ✗

  A data preprocessing pipeline that cleans raw datasets

  Why it's wrong here

  A data preprocessing pipeline that cleans raw datasets typically performs operations like missing-value imputation, scaling, and one-hot encoding as a distinct stage before model fitting. These steps are often orchestrated as pipeline steps in Azure ML and serve to feed clean data into a training script, but they do not themselves learn model weights or produce a trained model artifact. A training job, by contrast, is the specific compute execution that instantiates an estimator and calls fit() on the prepared data.

Go deeper

Related to this question