Databricks Machine Learning Professional Question 71
Select 3You are designing a custom machine learning model class in Databricks to predict customer churn. The model requires data preprocessing steps such as normalizing numerical features and encoding categorical variables. Additionally, the model needs access to metadata about the preprocessing logic to ensure consistent predictions during deployment. What are the benefits of including preprocessing logic and context within the custom model class?
- A
It ensures the consistency of preprocessing steps during both training and inference.
- B
It improves the scalability of the model when deployed on a cluster.
- C
It allows the model to include preprocessing metadata for better explainability.
- D
It completely eliminates the need for a separate feature store.
- E
It simplifies the deployment process by bundling both the model and preprocessing logic together.
Show answer and explanation
Correct answers: A, C, E
Explanation
Including preprocessing logic and context within custom model classes ensures that the same transformations are applied consistently during training and inference, increases transparency by embedding preprocessing metadata, and simplifies the deployment process by combining both preprocessing and prediction capabilities into a single entity. This approach minimizes errors during deployment and makes the model more robust.
- A. Correct.
Including preprocessing logic ensures that the same transformations applied during training are consistently applied during inference, reducing the risk of mismatched data formats.
- B. Incorrect.
While preprocessing logic is important, it does not inherently improve scalability. Scalability depends on other factors like architecture and infrastructure.
- C. Correct.
Including preprocessing metadata allows for easier understanding of model behavior and ensures transparency in the feature engineering process.
- D. Incorrect.
While preprocessing logic can reduce dependency on external systems, it does not eliminate the need for a feature store, which serves as a centralized repository for feature management.
- E. Correct.
Bundling preprocessing logic within the model simplifies deployment by ensuring the model is self-contained and can handle raw inputs directly.