Microsoft Certified: Azure Data Scientist Associate Practice Exam 2025: Latest Questions

You are preparing a dataset in Azure Machine Learning. You need to ensure that training and test splits are reproducible across runs and across different compute targets. Which approach should you use?

A data scientist wants to track metrics, parameters, and artifacts for each experiment run in Azure Machine Learning while iterating in a notebook. What is the recommended feature to use?

You are selecting a compute option in Azure Machine Learning for interactive development and debugging in Jupyter notebooks. Which compute type is most appropriate?

You have trained a model and need to use it for batch scoring of a large dataset daily. Which deployment option in Azure Machine Learning is typically most suitable?

Your training data is highly imbalanced (1% positive class). You want a single metric in Azure Machine Learning to better reflect performance on the minority class. Which metric is the best choice?

You plan to operationalize model training as a reusable workflow that includes data preparation, training, and evaluation steps. You also want the workflow to be runnable on a schedule. Which Azure Machine Learning capability best fits this requirement?

You trained a scikit-learn model locally. In Azure Machine Learning, you want to deploy it to a managed online endpoint and ensure the scoring environment has the exact required Python dependencies. What should you create and register with the deployment?

A model deployed to an online endpoint returns HTTP 500 errors after a new deployment. Logs show an exception: "ModuleNotFoundError" for a library used in the scoring script. What is the most likely root cause?

Your organization requires that training jobs cannot exfiltrate data to the public internet. You must run Azure Machine Learning training on managed compute while restricting network traffic to private endpoints. Which design best meets this requirement?

A regulated environment requires you to deploy a model with a repeatable, auditable process. You must ensure the exact model artifact, code, and environment used for production can be traced back to the training run. Which approach best satisfies this requirement in Azure Machine Learning?

You are creating an Azure Machine Learning pipeline to train a model weekly. You want each run to use the exact same Python dependencies to ensure reproducibility. What is the recommended approach?

You are exploring a large dataset in a notebook using pandas. The dataset does not fit into memory on your compute instance. What should you use in Azure Machine Learning to work with the data at scale while keeping a similar dataframe-style experience?

You registered a model in Azure Machine Learning. You want to deploy it for real-time inference using a managed service with autoscaling and minimal infrastructure management. Which compute target should you choose?

You are training a model and notice that one category accounts for 98% of the target labels. Your model achieves 98% accuracy but performs poorly on the minority class. Which evaluation metric is most appropriate to focus on?

You have a training script that logs metrics using MLflow. You want to compare multiple runs and choose the best model based on a metric threshold before registering it. Which Azure Machine Learning capability best supports this workflow?

A regulated enterprise requires that training data and model artifacts do not traverse the public internet. You need to use Azure Machine Learning while keeping traffic private. What should you implement?

You deployed a real-time endpoint. Requests succeed, but latency is highly variable and occasionally spikes. You suspect the model is being loaded repeatedly. Which configuration change most directly addresses this issue?

You need to run hyperparameter tuning for a scikit-learn model and parallelize trials across a compute cluster. You also want to use an early termination policy to stop poorly performing trials. Which approach should you use in Azure Machine Learning?

You deployed a model to a managed online endpoint. After deployment, the service returns HTTP 500 errors for some inputs. You need to identify whether the failures are due to schema mismatches in the request payload versus model runtime errors. What is the best next step?

Your team wants to ensure that a deployed real-time endpoint always uses the same preprocessing logic as training, even as the code evolves. You have feature transformations implemented in Python. What is the most robust design pattern in Azure Machine Learning to reduce training-serving skew?

You are training a classification model in Azure Machine Learning. The target class occurs in only 2% of rows. Your initial model shows high accuracy but very low recall for the minority class. What is the BEST next step to evaluate the model in a way that reflects the business need to catch the minority class?

You must deploy a trained model to a managed online endpoint in Azure Machine Learning. Company policy requires that the endpoint cannot be publicly reachable from the internet and must only be accessible from within the corporate virtual network. What should you configure?

You are building an Azure Machine Learning pipeline. A feature engineering step takes 40 minutes and rarely changes, while the training step is rerun frequently with different hyperparameters. You want to minimize pipeline runtime and compute cost across runs. What should you do?

After deploying a model to a managed online endpoint, you notice prediction quality slowly degrades over several weeks. You suspect data drift between training data and production inference data. What is the most appropriate approach in Azure Machine Learning to detect and investigate this?

You are using Automated ML for a time-series forecasting problem. The dataset contains multiple stores and products, and you need the model to learn separate patterns per store-product combination while still training as a single experiment. What should you configure in Automated ML?