A team applies normalization and vocabulary lookups to features during training, then reimplements them in the serving service. Predictions drift because the two copies diverge. What key advantage does TensorFlow Transform offer over preprocessing written separately for training and serving?
Show answer and explanation
Answer: C
TensorFlow Transform records preprocessing as a graph during its analysis phase and exports it together with the model. At serving time the same graph runs on raw inputs, so training and inference cannot diverge through two separate implementations. Speed is not the main benefit, and TFT does not choose transforms or replace feature design.
Why the other options are wrong
- A. Speed is not the central reason to use TFT. A faster but separately maintained implementation would still allow training and serving to diverge.
- B. TFT applies the transforms you specify; it does not pick them. Choosing appropriate transformations remains an engineering decision.
- D. Feature design still requires domain knowledge. TFT helps apply features consistently, but it does not remove the need to decide what features to build.