cancel
Showing results for 
Search instead for 
Did you mean: 

challenger models support different features?

challenger models support different features?

We're looking at using challenger models and wondering if challengers can have different features to the champion/to each other? Or, do the features have to match exactly? tia!

Emily

Labels (1)
2 Replies

Challengers can have different features, however a superset of the data must be provided so that scoring can be run through a challenger.  Eg.

Deployed model uses length and width.

Challenger uses length, width, and height.

The data sent in to the deployed model must include length, width, and height.  The deployed model will simply ignore the additional feature.

When selecting a challenger model, consider the following: 1. It must have the same target type as the champion model 2. It does not need to be trained on the same feature list as the champion model, but must share some features. 3. However, to successfully replay predictions, the user must send union of all features required for champion and challengers 4. It does not have to be built from the same project as the champion model