Databricks Machine Learning Associate Question 438
Select 3You are tasked with training a machine learning model in Databricks using features stored in a feature store table. Which of the following steps are necessary to successfully utilize the feature store for training your model?
- A
Use the
FeatureStoreClientto load the required features from the feature store table. - B
Export the feature store table to a local CSV file before training the model.
- C
Join the feature store table with your raw dataset to create a training dataset using the feature store utilities.
- D
Log the model to the feature store after training to enable feature lineage and tracking.
- E
Use a Spark SQL query to directly access the feature store table for model training.
Show answer and explanation
Correct answers: A, C, D
Explanation
To train a model using features from a Databricks Feature Store table, you must use the FeatureStoreClient to access the features and join them with your raw dataset to create a training dataset. After training, logging the model to the feature store ensures proper tracking and lineage. Exporting the table or using Spark SQL directly bypasses key feature store functionalities, which is not recommended.
- A. Correct.
This is correct. The
FeatureStoreClientis the primary interface for interacting with the Databricks Feature Store and is used to load features for model training. - B. Incorrect.
This is incorrect. Exporting the feature store table to a local CSV file is unnecessary and goes against best practices, as the feature store is directly accessible within Databricks.
- C. Correct.
This is correct. Joining the feature store table with your raw dataset is a required step to create a training dataset enriched with the features you need.
- D. Correct.
This is correct. Logging the model to the feature store ensures that feature lineage and tracking are maintained, which is a key benefit of using the feature store.
- E. Incorrect.
This is incorrect. While you can query the feature store table using Spark SQL, doing so bypasses important feature store utilities such as versioning, feature lineage, and tracking.