Databricks-ML-Pro Model Development Practice Test 8 — 15 Questions

Question 1

What is the primary purpose of registering a model in the MLflow Model Registry?

Accepted Answer

To provide a structured workflow for versioning, stage management, and model governance.. The MLflow Model Registry provides a centralized hub for managing the model lifecycle, including versioning, stage transitions (e.g., Staging to Production), and lineage tracking. This is foundational for MLOps because it provides a single source of truth for all stakeholders, enabling controlled deployments, auditability, and the ability to easily revert to previous model versions if performance regressions are detected in production.

Answer

To increase the training speed of the machine learning model.

Answer

To automatically retrain the model when data drift is detected.

Answer

To convert Python code into high-performance C++ code.

Question 2

When designing a model training pipeline, which TWO features of Unity Catalog best support compliance and model governance?

Accepted Answer

Column-level access control to restrict sensitive data visibility during feature engineering.. Unity Catalog acts as a centralized governance layer for all data and AI assets. By providing fine-grained access control and end-to-end lineage, it ensures that only authorized users can access sensitive training data and that the origin of every model can be traced back to the original source data, which is essential for meeting regulatory requirements and maintaining corporate security standards.

Answer

Automated hyperparameter grid search for all registered datasets.

Answer

Built-in model serving endpoints for real-time inference.

Answer

Automatic translation of SQL queries into Python code.

Question 3

An ML engineer wants to ensure that their model training pipeline is robust against data quality issues. Which approach, if integrated into the pipeline, most effectively detects skewed or missing values before training begins?

Accepted Answer

Using DLT Expectations to define and enforce constraints on input data.. Integrating data quality checks, such as those provided by Delta Live Tables (DLT) expectations or Great Expectations, is crucial for MLOps. By enforcing schema and statistical constraints at the ingest and preparation stages, the pipeline can fail early if data quality falls below standards, preventing the training of 'garbage-in-garbage-out' models and saving compute costs while maintaining model reliability in production.

Answer

Manually inspecting the first 100 rows of the dataset using a notebook cell.

Answer

Relying on the model to handle missing values by using imputation during training.

Answer

Increasing the size of the training cluster to process more data.

Question 4

What is the best way to handle secrets (like API keys for external feature sources) within a Databricks notebook during model development?

Accepted Answer

Use the Databricks Secrets API to retrieve credentials at runtime.. Using the Databricks Secrets API is the only secure way to manage credentials. By referencing keys through the `dbutils.secrets.get()` function, developers ensure that sensitive information is never hardcoded or stored in clear text within the version-controlled code. This practice prevents unauthorized access to external systems and is a mandatory security requirement for any enterprise environment dealing with sensitive or paid data APIs.

Answer

Hardcode the keys in a configuration Python script and import it into the notebook.

Answer

Store the secrets in an environment variable on the cluster config.

Answer

Prompt the user to enter the secret manually when the notebook runs.

Question 5

Which of the following is an advantage of using Databricks AutoML compared to building a custom Scikit-Learn training loop?

Accepted Answer

It generates the training notebook for the trial, allowing for further refinement and transparency.. Databricks AutoML significantly accelerates the development lifecycle by automating tedious tasks like data cleaning, feature engineering, and model selection. It provides a baseline of high-performance models while simultaneously producing the source code for the best-performing model. This allows teams to iterate rapidly, then customize the generated code, combining the speed of automation with the flexibility of manual tuning for complex production requirements.

Answer

AutoML models are guaranteed to have higher accuracy than custom models.

Answer

It prevents the need for any feature engineering by automatically creating all required variables.

Answer

It eliminates the need for monitoring the model after deployment.

Question 6

A data scientist is training a machine learning model on Databricks using MLflow. They need to track hyperparameter tuning experiments while ensuring that each iteration is uniquely identifiable and reproducible. Which feature should they use to group related runs within a single experiment?

Accepted Answer

MLflow Nested Runs via the 'parent_run_id' parameter.. MLflow nested runs are the industry-standard approach for hierarchical organization of hyperparameter tuning tasks. By creating a parent run for the overall optimization process and child runs for individual parameter sets, data scientists gain clear visibility into model performance metrics across the entire search space, simplifying comparative analysis and facilitating efficient model selection during the development lifecycle.

Answer

MLflow Tags to append metadata for each run iteration.

Answer

MLflow Experiment IDs for every parameter variation.

Answer

MLflow Model Registry versions to track iterative progress.

Question 7

When developing a machine learning model on Databricks, what is the primary benefit of using Feature Store over standard Delta Lake tables for feature management?

Accepted Answer

It ensures feature consistency between training and inference.. Databricks Feature Store provides a unified interface for feature engineering, storage, and retrieval. Its primary advantage is consistency, ensuring that the same feature transformation logic is applied during both training and inference. This eliminates "training-serving skew," a common issue where discrepancies between development and production pipelines lead to degraded model performance in real-world scenarios.

Answer

It provides built-in GPU acceleration for all training jobs.

Answer

It automates the elimination of data leakage during training.

Answer

It automatically converts all data types to tensors.

Question 8

A data scientist is using Databricks AutoML to solve a classification problem. After the run completes, they want to modify the feature engineering logic for the best-performing model. Which artifact should they retrieve from the AutoML run?

Accepted Answer

The AutoML generated 'best trial' notebook.. Databricks AutoML generates a 'data exploration' notebook and a 'best trial' notebook for every experiment. The notebook contains the full Python code used for feature engineering, model training, and hyperparameter tuning. Accessing this notebook allows data scientists to treat AutoML results as a baseline, enabling them to refine feature engineering steps or adjust model parameters for further performance improvements.

Answer

The raw MLflow model artifact stored in S3/DBFS.

Answer

The MLflow experiment metrics file in JSON format.

Answer

The database schema definition for the input features.

Question 9

A data scientist is building a model that requires custom preprocessing logic that is not available in standard libraries. They need to ensure this logic is bundled with the model for inference. What is the recommended approach to encapsulate this custom logic?

Accepted Answer

Implement a custom class inheriting from mlflow.pyfunc.PythonModel.. Using the MLflow PyFunc (Python Function) flavor is the standard way to package arbitrary logic with a model. It allows developers to define a custom wrapper that includes preprocessing, prediction, and post-processing steps. When the model is logged, the custom class and its environment dependencies are saved, ensuring that the exact same logic is executed during inference, regardless of the deployment target.

Answer

Save the preprocessing logic in a separate Python script.

Answer

Use a Spark UDF for preprocessing during inference.

Answer

Hardcode the preprocessing logic directly into the SQL query.

Question 10

A data scientist is training a deep learning model on Databricks. They observe that the training process is significantly slower than expected. Upon inspection, they find that data loading from DBFS is the bottleneck. What is the most effective way to improve data loading speed for deep learning training on Databricks?

Accepted Answer

Use the Petastorm library to read data directly from Delta tables.. Using Petastorm or the standard 'tf.data' dataset API with Delta Lake significantly improves throughput for deep learning models. By converting data into a highly efficient, partitioned format that can be streamed directly to GPU memory, developers bypass the latency overhead associated with reading small files from DBFS, enabling the model to utilize the full processing power of the allocated cluster.

Answer

Move all data to local disk on the driver node.

Answer

Convert the dataset into a single large CSV file.

Answer

Increase the learning rate to reduce training time.

Question 11

A data scientist is developing a scikit-learn model on Databricks and wants to track the full lineage of the training data, including the exact Delta table version used. They are using MLflow Tracking with a Unity Catalog-enabled workspace. Which approach best captures this lineage as part of the MLflow run?

Accepted Answer

Log the Delta table version as a tag using `mlflow.set_tag` and record the table name in the run's tags.. The correct approach is to explicitly log the Delta table version and table name as tags on the MLflow run. This creates a clear, queryable link between the model and the exact data snapshot, supporting reproducibility and audit requirements. Neither autologging nor Unity Catalog lineage alone captures the specific version used in a run, and saving transaction logs as artifacts is inefficient and not queryable.

Answer

Enable MLflow autologging for scikit-learn, which automatically records the Delta table version used during training.

Answer

Use `mlflow.log_artifact` to save a copy of the Delta table's transaction log as a JSON file.

Answer

Rely on Unity Catalog's automatic lineage tracking, which captures data-to-model relationships without additional logging.

Question 12

A data scientist is using MLflow to track experiments on Databricks. They want to compare multiple runs and identify the one with the lowest validation RMSE. Which MLflow UI feature allows them to sort and filter runs by a specific metric?

Accepted Answer

The Runs table with column sorting and metric filters.. The MLflow UI's Runs table is designed for comparing runs. It allows sorting by metrics such as RMSE and filtering runs based on metric values or tags. This makes it straightforward to identify the best run. Other tabs or the Model Registry serve different purposes.

Answer

The Notebook's `mlflow.search_runs` output.

Answer

The Model Registry's version list.

Answer

The Artifacts tab within a run.

Question 13

A machine learning engineer is developing a custom MLflow Python model that requires a pre-processing step using a scikit-learn pipeline. They want to log the model such that it can be served with the pipeline included. Which approach should they take?

Accepted Answer

Create a Python class that inherits from `mlflow.pyfunc.PythonModel`, include the pipeline in the `predict` method, and log with `mlflow.pyfunc.log_model`.. For custom models that include pre-processing and a scikit-learn pipeline, the recommended approach is to create a custom `mlflow.pyfunc.PythonModel` subclass. The `predict` method can invoke the pipeline and any additional logic. Logging with `mlflow.pyfunc.log_model` packages the model and its dependencies for serving. This ensures the entire pipeline is included.

Answer

Log the pipeline as a separate model and use MLflow's multi-model serving to chain them.

Answer

Log the scikit-learn pipeline directly with `mlflow.sklearn.log_model`.

Answer

Use `mlflow.sklearn.log_model` and pass the pipeline as an artifact, then load it manually in a custom serving script.

Question 14

A machine learning engineer is building a feature engineering pipeline in Databricks using Feature Store. They need to ensure that the same feature computation logic is used for both training and batch scoring, and that features are automatically refreshed. Which approach should they take?

Accepted Answer

Define a feature computation function using the Feature Store client, register it, and create a Feature Store table with a scheduled refresh.. Defining a feature computation function and registering it with Feature Store ensures that training and scoring use identical logic. Creating a Feature Store table with scheduled refresh automates feature updates. This combination provides consistency, reproducibility, and freshness. The other options rely on manual steps or lack the necessary integration with Feature Store, so they do not guarantee consistent feature computation and automatic refresh.

Answer

Create a Feature Store table using a PySpark UDF that reads from a Delta table and schedule a job to refresh it.

Answer

Write a Python function that computes features and call it manually in both training and scoring notebooks.

Answer

Use MLflow to log the feature engineering code as an artifact and manually apply it during scoring.

Question 15

A data scientist is using MLflow on Databricks to tune a scikit-learn GradientBoostingRegressor with Hyperopt. They configure fmin with max_evals=50, but notice that runs appear in the experiment without parameters or metrics logged, and the best model cannot be reproduced. They want to ensure every trial is fully tracked. Which change should they make?

Accepted Answer

Wrap the objective function's training and evaluation code in `with mlflow.start_run():` (or use `mlflow.start_run(nested=True)` inside the objective) and log params/metrics explicitly.. Hyperopt's fmin does not manage MLflow runs; each trial must explicitly start a run to log parameters and metrics. Wrapping the objective in mlflow.start_run ensures isolated tracking per trial, which is essential for reproducibility and comparison. Other options confuse search breadth, autologging, or distributed execution with run lifecycle management.

Answer

Use `SparkTrials` instead of `Trials` so that each trial is logged as a separate MLflow run.

Answer

Call `mlflow.autolog()` before Hyperopt and set `nested=True` in fmin.

Answer

Increase `max_evals` to at least 200 so that MLflow has enough data to create runs automatically.