Machine Learning Associate · 38% of the exam

Databricks Machine Learning: free practice questions

5 sample questions from our 100-question bank for this domain — answers and explanations included. These are the same scenario-based style as the real Databricks exam.

1. Which statement about Databricks Repos in an ML workflow is TRUE?

  • A. Repos integrate Git into the workspace for versioning and collaboration on ML code✓ Correct
  • B. Repos train models automatically
  • C. Repos store the feature tables
  • D. Repos replace MLflow
Explanation

Databricks Repos bring Git version control into the workspace for collaborating on and versioning ML code. They don't train models (B), store feature tables (C), or replace MLflow (D).

2. A data scientist is new to Databricks and wants to understand how MLflow is integrated into the platform. Which of the following statements BEST describes the default MLflow tracking behavior when running a training script in a Databricks notebook?

  • A. MLflow tracking must be explicitly enabled by calling `mlflow.enable_autologging()` before any metrics are captured.
  • B. Databricks automatically provides a managed MLflow tracking server; runs are logged to the workspace's MLflow tracking URI without any additional configuration.✓ Correct
  • C. MLflow runs are stored only in the driver node's local filesystem and must be manually exported to a persistent store.
  • D. Databricks requires users to configure an external MLflow tracking server URI before any experiment data can be persisted.
Explanation

Correct: Databricks provides a fully managed MLflow tracking server integrated into every workspace. The tracking URI is pre-configured, so experiments and runs are automatically persisted without any extra setup. Wrong - A: `mlflow.enable_autologging()` is not a real MLflow function; autologging is enabled with `mlflow.autolog()` or framework-specific calls, and even without autologging, manual `mlflow.log_*` calls work immediately. Wrong - C: MLflow on Databricks stores data in a managed backend store, not the driver's local filesystem. Wrong - D: No external tracking server URI configuration is needed; Databricks sets this automatically for every notebook and job.

3. A Databricks AutoML run completes on a binary classification problem. The ML engineer inspects the generated experiment and finds several trial notebooks. What is the PRIMARY purpose of the 'data exploration notebook' that Databricks AutoML automatically generates alongside the trial notebooks?

  • A. It contains the final production-ready model code that can be deployed directly to Databricks Model Serving.
  • B. It provides an automated exploratory data analysis (EDA) report — including class distributions, missing value summaries, and feature correlations — to help the data scientist understand the dataset before or alongside modeling.✓ Correct
  • C. It defines the hyperparameter search space used by AutoML and allows the engineer to modify it before re-running the experiment.
  • D. It stores the MLflow experiment ID and run IDs for all trials so that the engineer can programmatically access results.
Explanation

Option B is correct: Databricks AutoML automatically generates a data exploration notebook that performs EDA — it shows class balance, missing value rates, feature statistics, and correlations — giving the data scientist insight into data quality and characteristics. Option A is wrong because the trial notebooks (not the data exploration notebook) contain model training code, and even those require further engineering before production deployment. Option C is wrong because the hyperparameter search space is defined internally by AutoML and is not exposed as an editable artifact in the data exploration notebook. Option D is wrong because MLflow experiment tracking metadata is accessible via the MLflow UI and API, not stored in the data exploration notebook.

4. A data scientist uses the following code to log metrics during model training: ```python with mlflow.start_run(): for epoch in range(10): loss = train_one_epoch() mlflow.log_metric('train_loss', loss, step=epoch) ``` After the run completes, what does the MLflow UI's metric chart for 'train_loss' display?

  • A. A time-series line chart showing the value of train_loss at each epoch (step 0 through 9), allowing the engineer to observe training convergence.✓ Correct
  • B. Only the final value of train_loss (step 9), because MLflow overwrites the metric value on each call to log_metric.
  • C. An error, because log_metric does not accept a 'step' argument and the code will raise a TypeError.
  • D. A histogram of all train_loss values logged across the 10 epochs.
Explanation

Option A is correct: mlflow.log_metric() accepts an optional 'step' parameter that allows multiple values for the same metric key to be recorded at different steps. The MLflow UI renders these as a time-series line chart, making it easy to visualize training convergence over epochs. Option B is incorrect; MLflow does not overwrite metric values when step is provided — each (metric_key, step) pair is stored as a distinct data point. Option C is incorrect; 'step' is a valid and officially documented parameter of mlflow.log_metric(). Option D is incorrect; the UI does not render metric history as a histogram — it uses a line chart with step on the x-axis and metric value on the y-axis.

5. A data scientist wants to compare several training runs' metrics visually and pick the best model. Where in Databricks do they do this?

  • A. The MLflow Experiments UI✓ Correct
  • B. The Delta Sharing page
  • C. The cluster metrics tab only
  • D. The Repos Git history
Explanation

The MLflow Experiments UI lists runs with parameters and metrics and supports comparison to select the best model. Delta Sharing (B), cluster metrics (C), and Git history (D) don't compare experiment runs.

95 more questions in this domain

Practice the full bank with instant grading, flashcards, and a timed mock exam.

Start practicing free