step_corr: High Correlation Filter

View source: R/corr.R

step_corrR Documentation

High Correlation Filter

Description

step_corr() creates a specification of a recipe step that will potentially remove variables that have large absolute correlations with other variables.

Usage

step_corr(
  recipe,
  ...,
  role = NA,
  trained = FALSE,
  threshold = 0.9,
  use = "pairwise.complete.obs",
  method = "pearson",
  removals = NULL,
  skip = FALSE,
  id = rand_id("corr")
)

Arguments

recipe

A recipe object. The step will be added to the sequence of operations for this recipe.

...

One or more selector functions to choose variables for this step. See selections() for more details.

role

Not used by this step since no new variables are created.

trained

A logical to indicate if the quantities for preprocessing have been estimated.

threshold

A value for the threshold of absolute correlation values. The step will try to remove the minimum number of columns so that all the resulting absolute correlations are less than this value.

use

A character string for the use argument to the stats::cor() function.

method

A character string for the method argument to the stats::cor() function.

removals

A character string that contains the names of columns that should be removed. These values are not determined until prep() is called.

skip

A logical. Should the step be skipped when the recipe is baked by bake()? While all operations are baked when prep() is run, some operations may not be able to be conducted on new data (e.g. processing the outcome variable(s)). Care should be taken when using skip = TRUE as it may affect the computations for subsequent operations.

id

A character string that is unique to this step to identify it.

Details

This step can potentially remove columns from the data set. This may cause issues for subsequent steps in your recipe if the missing columns are specifically referenced by name. To avoid this, see the advice in the Tips for saving recipes and filtering columns section of selections.

This step attempts to remove variables to keep the largest absolute correlation between the variables less than threshold.

When a column has a single unique value, that column will be excluded from the correlation analysis. Also, if the data set has sporadic missing values (and an inappropriate value of use is chosen), some columns will also be excluded from the filter.

The arguments use and method don't take effect if case weights are used in the recipe.

Value

An updated version of recipe with the new step added to the sequence of any existing operations.

Tidying

When you tidy() this step, a tibble with column terms (the columns that will be removed) is returned.

Tuning Parameters

This step has 1 tuning parameters:

  • threshold: Threshold (type: double, default: 0.9)

Case weights

This step performs an unsupervised operation that can utilize case weights. As a result, case weights are only used with frequency weights. For more information, see the documentation in case_weights and the examples on tidymodels.org.

Author(s)

Original R code for filtering algorithm by Dong Li, modified by Max Kuhn. Contributions by Reynald Lescarbeau (for original in caret package). Max Kuhn for the step function.

See Also

Other variable filter steps: step_filter_missing(), step_lincomb(), step_nzv(), step_rm(), step_select(), step_zv()

Examples


data(biomass, package = "modeldata")

set.seed(3535)
biomass$duplicate <- biomass$carbon + rnorm(nrow(biomass))

biomass_tr <- biomass[biomass$dataset == "Training", ]
biomass_te <- biomass[biomass$dataset == "Testing", ]

rec <- recipe(
  HHV ~ carbon + hydrogen + oxygen + nitrogen + sulfur + duplicate,
  data = biomass_tr
)

corr_filter <- rec %>%
  step_corr(all_numeric_predictors(), threshold = .5)

filter_obj <- prep(corr_filter, training = biomass_tr)

filtered_te <- bake(filter_obj, biomass_te)
round(abs(cor(biomass_tr[, c(3:7, 9)])), 2)
round(abs(cor(filtered_te)), 2)

tidy(corr_filter, number = 1)
tidy(filter_obj, number = 1)


recipes documentation built on Aug. 26, 2023, 1:08 a.m.