recoverable_features_objs.md

Effect

Exposure of private data possible/ no such data field during production

Rules to avoid leakage

When try to hide the meaning of the feature if it principaly restorable provide as it is, don't trick yourself that is secure. Incorporate DS work to check if feature is truly non-restorable.

Incorporation stage

Dataset preparation

Locate in code

Feature processing before train dataset is taged as modeling ready.

Was met or loosely based on

kaggle "Optiver Realized Volatility Prediction" competition nyanpn comment