recipes-patternslisted

# Recipes Feature Engineering Patterns ## Overview Comprehensive patterns for feature engineering using the recipes package. Covers preprocessing steps for numeric, categorical, and text data while preventing information leakage. ## Recipe Fundamentals ### Basic Recipe Structure ```r library(recipes) # Initialize recipe with formula rec <- recipe(outcome ~ ., data = training_data) # Or with explicit roles rec <- recipe(training_data) |> update_role(outcome, new_role = "outcome") |> update_role(id_column, new_role = "ID") |> update_role(-outcome, -id_column, new_role = "predictor") ``` ### Selector Functions ```r # Type-based selectors all_predictors() all_outcomes() all_numeric_predictors() all_nominal_predictors() all_numeric() all_nominal() # Name-based selectors starts_with("prefix_") ends_with("_suffix") contains("pattern") matches("regex") one_of(c("var1", "var2")) ``` ## Numeric Preprocessing ### Normalization and Scaling ```r rec <- recipe(outcome ~ ., data = train) |> # Center and scale (z-score) step_normalize(all_numeric_predictors()) |> # Scale to [0, 1] step_range(all_numeric_predictors(), min = 0, max = 1) |> # Center only step_center(all_numeric_predictors()) |> # Scale only step_scale(all_numeric_predictors()) ``` ### Transformations for Normality ```r rec <- recipe(outcome ~ ., data = train) |> # Yeo-Johnson (handles zero and negative values) step_YeoJohnson(all_numeric_predictors()) |> # Box-Cox (positive values