Certified Machine Learning ProfessionalDemo

Question

A Machine Learning Engineer is developing a model with features derived from streaming clickstream data in Databricks. To maintain consistent feature values between training and real-time inference, the engineer must ensure features are computed in exactly the same manner in both environments. They must also avoid offline-online skew in feature computation.

Which approach meets these requirements?

A Use an MLflow PyFunc model to encapsulate and reuse the exact same feature computation code during both training and inference.
B Cache features in-memory and use custom scripts to merge them at inference.
C Implement separate preprocessing pipelines for training and inference, keep documentation synchronized.
D Use batch-computed features for training but real-time-computed features at inference to match production performance.

Explanation

An MLflow PyFunc model can package custom preprocessing and inference logic with the model. Reusing the same feature-computation code for both training and serving prevents independently implemented transformations from producing offline-online feature skew.

Learn more

Certified Machine Learning ProfessionalDemo

AstroTutor

Notes

Readiness

QuestionQ80

Community Discussion