Package index • featR

Working with results

Every selection function except fs_pca() and fs_svd() returns an fs_result. These functions print it, summarize it, and extract the selected names.

selected(): Extract the selected features from a featR result
print(<fs_result>): Print a featR result
summary(<fs_result>): Summarize a featR result

Filters

Score each feature without fitting a predictive model. They are fast and work on wide data, but they judge one feature at a time, so they miss features that only help in combination.

fs_supervised(): Supervised filter that scores each feature against a target
fs_unsupervised(): Unsupervised filter that scores each feature without a target
fs_chi(): Chi-squared tests for categorical features
fs_infogain(): Rank features by information gain
fs_correlation(): Find and prune highly correlated variables

Regularization and embedded importance

Fit one model and keep the features it uses: nonzero coefficients for the lasso and elastic net, variable importance for random forests, and the terms of the final MARS model.

fs_lasso(): Lasso feature selection with cross-validation
fs_elastic(): Elastic-net feature selection with tuned alpha and lambda
fs_randomforest(): Random forest importance with a held-out evaluation
fs_mars(): MARS (earth) feature selection

Wrappers

Refit a model on many candidate subsets and keep the best one. They are the slowest family, and their choice is tied to the model they refit.

fs_recursivefeature(): Recursive feature elimination with held-out evaluation
fs_svm(): Train and evaluate an SVM, with optional SVM-RFE feature selection
fs_boruta(): All-relevant feature selection with Boruta
fs_stepwise(): Stepwise linear-regression feature selection by AIC
fs_bayes(): Bayesian predictor-subset selection with brms and LOO

Dimensionality reduction

Replace the original columns with new components instead of keeping a subset. These return a list of their own rather than an fs_result.

fs_pca(): Principal component analysis with tidy results and optional plotting
fs_svd(): Singular value decomposition with optional scaling and truncation

Package

featR featR-package: featR: A Unified Toolkit for Feature Selection