Databricks Machine Learning Associate Question 88
Single answerYou are tasked with training a machine learning model to predict customer churn using features stored in a Databricks Feature Store table named customer_churn_features. What is the correct sequence of steps to properly load features from the feature store and use them for model training?
- A
Use the Feature Store client to load the feature table as a Pandas DataFrame, split the data into training and test sets, and train your model.
- B
Use the Feature Store client to create a training dataset by specifying the feature table and labels, retrieve the features as a Spark DataFrame, and train your model.
- C
Directly query the feature table using PySpark SQL, convert it to a Pandas DataFrame, and train your model.
- D
Write a SQL query to retrieve the feature table, join it with the labels table, and train your model.
Show answer and explanation
Correct answer: B
Explanation
The Databricks Feature Store client is designed to manage feature metadata, ensure consistent feature versioning, and provide seamless integration with machine learning workflows. By using the Feature Store client to create a training dataset, you can ensure reproducibility, governance, and compatibility with Databricks ML pipelines. Direct SQL queries or manual DataFrame manipulations bypass these critical features, making them unsuitable for leveraging the full capabilities of the Feature Store.
- A. Incorrect.
This is incorrect because loading the feature table as a Pandas DataFrame without using the Feature Store client does not ensure the correct feature versioning or tracking of feature usage.
- B. Correct.
This is correct because using the Feature Store client ensures proper feature versioning, retrieval of features as a training dataset, and integration with machine learning pipelines.
- C. Incorrect.
This is incorrect because querying the feature table directly without the Feature Store client bypasses the feature store's versioning and tracking capabilities, which are essential for reproducibility and governance.
- D. Incorrect.
This is incorrect because while joining tables via SQL is possible, it does not leverage the Feature Store client, which is necessary for managing feature metadata, versioning, and tracking.