In rowanjh/stepFCBF: Fast Correlation Based Filter (FCBF) for Feature Selection: A 'tidymodels' Recipe Step Integration

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  out.width = "100%"
)

stepFCBF package

Introduction

Provides the step_fcbf function, which allows fast correlation based filter (FCBF) feature selection to be added as a recipe step in 'tidymodels'. FCBF is a feature selection method that aims to retain a set of features with high correlation to the outcome, and low correlation to other retained features. It uses the metric 'Symmetrical Uncertainty' to determine the magnitude of correlation.

The underlying calculations are conducted by the Bioconductor package FCBF, and the algorithm is described in Yu, L. and Liu, H.; Feature Selection for High-Dimensional Data: A Fast Correlation Based Filter Solution,Proc. 20th Intl. Conf. Mach. Learn. (ICML-2003), Washington DC, 2003.

Discretization

step_fcbf can handle both nominal and numeric features. However, the underlying FCBF algorithm can only handle nominal features, so numeric features first need to be discretized. step_fcbf function internally converts numeric features to binary nominal features, using a median split by default (other quantiles between 0-1 can be provided with the 'cutpoint = ' argument). Discretization is only used within the feature selection algorithm, once features have been selected the numeric version of the feature is retained for further processing and modeling.

Symmetric Uncertainty Threshold

FCBF requires a cut point to be provided for symmetrical uncertainty (between 0-1). Smaller thresholds of SU will result in more features being retained. Appropriate thresholds are data-dependent, so it is recommended that different values of SU be explored using a subset of the training set. e.g. see 'FCBF::su_plot()'

Installation

You can install the development version of stepFCBF from GitHub with:

# install.packages("devtools")
devtools::install_github("rowanjh/stepFCBF")

stepFCBF depends on the bioconductor package "FCBF". Run the below code or see the FCBF package website for more detailed installation instructions:

BiocManager::install("FCBF")

Examples

Basic example for including step_fcbf into a recipe

library(recipes)
library(stepFCBF)
data("iris")

my_recipe <- recipe(Species ~ ., data = iris) %>%
    step_fcbf(all_predictors(), min_su = 0.001)

prepped_recipe <- my_recipe %>% prep(iris)

# Original features
prepped_recipe$var_info

# Selected features
prepped_recipe$term_info

rowanjh/stepFCBF documentation built on April 8, 2023, 4:28 a.m.