Position normalizer

Definition

A position normalizer is a feature available to a machine learning algorithm at the learning stage, but not at the stage where it is to be used.

An example of a position normalizer feature is for ranking problems. The feature of "what rank was the item shown at" is a feature available when training, but not when using the model, because the whole purpose of the model is to rank the items, so the rank is not pre-determined (there are interesting ways to overcome this, but these are complex, and treating the rank feature as a position normalizer is usually adequate).