Databricks Machine Learning Associate Question 86
Select 2You are tasked with building a machine learning model to predict customer churn using a feature store table in Databricks. The feature store table is named customer_features and contains all the relevant features for the model. Which of the following steps are necessary to train the model using features from the feature store?
- A
Load the
customer_featurestable as a Spark DataFrame and use it directly for model training. - B
Use the Feature Store client to load features from the
customer_featurestable into a DataFrame. - C
Ensure the primary key is specified when retrieving features from the Feature Store.
- D
Register the trained model back to the Feature Store after training is complete.
- E
Manually write code to join features in the
customer_featurestable with labels before training the model.
Show answer and explanation
Correct answers: B, C
Explanation
When training a model using features from a Databricks Feature Store table, you must use the Feature Store client to retrieve the features. This ensures proper feature versioning and management. Additionally, specifying the primary key is critical to correctly join the features with the target labels. These steps are necessary to leverage the full capabilities of the Feature Store and maintain data integrity.
- A. Incorrect.
This option is incorrect because Feature Store tables are not directly used as Spark DataFrames without leveraging the Feature Store client. Loading the table this way would bypass important functionality such as feature versioning.
- B. Correct.
This option is correct because the Feature Store client in Databricks is specifically designed to retrieve and manage features from Feature Store tables for use in model training.
- C. Correct.
This option is correct because specifying the primary key is necessary for the Feature Store client to retrieve the correct features corresponding to the entity (e.g., customers in this case).
- D. Incorrect.
This option is incorrect because registering the trained model back to the Feature Store is optional, not a mandatory step for training a model using the Feature Store.
- E. Incorrect.
This option is incorrect because the Feature Store client automatically handles joining features with labels when you specify the primary key and label column, reducing manual effort.