Chapter 8 Feature engineering with recipes

Learning objectives:

• Define feature engineering.
• List reasons that feature engineering might be beneficial.
• Use the {recipes} package to create a simple feature engineering recipe.
• Use selectors from the {recipes} package to apply transformations to specific types of columns.
• List some advantages of using a recipe for feature engineering.
• Describe what happens when a recipe is prepared with recipes::prep().
• Use recipes::bake() to process a dataset.
• Recognize how to use recipes::step_unknown(), recipes::step_novel(), recipes::step_other() to prepare factor variables.
• Explain how recipes::step_dummy() encodes qualitative data in a numeric format.
• Recognize techniques for dealing with large numbers of categories, such as feature hashing or encoding using the {embed} package (as described in this talk by Alan Feder at rstudio::global(2021)).
• Recognize methods for encoding ordered factors.
• Use recipes::step_interact() to add interaction terms to a recipe.
• Understand why some steps might only be applicable to training data.
• Recognize the functions from {recipes} and {themis} that are only applied to training data by default.
• Recognize that {recipes} includes functions for creating spline terms, such as step_ns().
• Recognize that {recipes} includes functions for feature extraction, such as step_pca().
• Use themis::step_downsample() to downsample data.
• Recognize other row-sampling steps from the {recipes} package.
• Use recipes::step_mutate() and recipes::step_mutate_at() for general {dplyr}-like transformations.
• Recall that the {textrecipes} package exists for text-specific feature-engineering steps.
• Understand that the functions of the {recipes} package use training data for all preprocessing and feature engineering steps to prevent leakage.
• Use {recipes} to prepare data for traditional modeling functions.
• Use tidy() to examine a recipe and its steps.
• Refer to columns with roles other than "predictor" or "outcome".