Open ecsalomon opened 3 years ago
One potential implementation is to make scaling a top-level key in the config (with sklearn class and params) and to pass all learners to the scaler. Another option would be to make scaling an optional key for model grid.
More model types than LR should have scaled features, and scaling can aid in postmodeling analysis for all model types. Abstract the scaler that @shaycrk wrote and allow it to take a learner class (e.g.,
xgboost.XGBClassifier
orsklearn.linear_model.LogisticRegression
) and apply the same scaling at test and train.