Store only a subset of features in output_examples artifact.
Support inference on multiple models.
Motivation
A BulkInferrer TFX component is used to perform batch inference on unlabeled tf.Examples.
The generated output examples contains the original features and the prediction results.
Keeping all original features in the output is troubling when dealing with feature heavy models.
For most of the use cases we only require example identifiers and the predictions in the output.
In machine learning, it is a common practice to train multiple models using the same feature set to perform different tasks (sometimes same tasks).
It will be convenient to have a multimodel inference feature in bulk-inferrer. The component should take a list of models and produce predictions for all models.
RFC: Modify BulkInferrer TFX component
Objective
Modify BulkInferrer TFX component.
Changes :-
output_examplesartifact.Motivation
A BulkInferrer TFX component is used to perform batch inference on unlabeled tf.Examples. The generated output examples contains the original features and the prediction results. Keeping all original features in the output is troubling when dealing with feature heavy models. For most of the use cases we only require example identifiers and the predictions in the output.
In machine learning, it is a common practice to train multiple models using the same feature set to perform different tasks (sometimes same tasks). It will be convenient to have a multimodel inference feature in bulk-inferrer. The component should take a list of models and produce predictions for all models.