Table of Highly Correlated Feature Pairs

Lists every pair of numeric features whose absolute correlation is at least `cutoff`, with optional labels and an executive caption ("X of the Y features (Z correlated"). Where [drop_correlated_columns()] says which columns to remove, this table is for the reader of the report.

Usage

tbl_correlated_pairs(
  data,
  features = NULL,
  cutoff = 0.65,
  method = c("pearson", "spearman"),
  metadata = NULL,
  format = c("kable", "gt", "tibble"),
  caption = TRUE,
  digits = 3
)

Arguments

data

A data frame.

features

Numeric columns to check; `NULL` uses all numeric columns.

cutoff

Absolute correlation threshold in (0, 1]; a pair at exactly the cutoff counts.

method

`"pearson"` or `"spearman"` (pairwise complete observations). Pairs with an undefined correlation (a constant column, fewer than 3 shared rows) are not listed.

metadata

Optional data frame with columns `columns` (feature names) and `labels`. Features without a label keep their name.

format

`"kable"` (needs `knitr`), `"gt"` (needs `gt`) or `"tibble"`.

caption

Build the caption (shown as the table caption for `"kable"` and `"gt"`, and always returned as the `caption` attribute of the tibble form).

digits

Decimals of the correlation in the formatted tables.

Value

For `"tibble"`, a tibble with `feature_1`, `feature_2`, `label_1`, `label_2` and `correlation`, sorted by decreasing absolute correlation, with attributes `caption`, `n_features` and `n_involved`. For the other formats the formatted table.

See also

[drop_correlated_columns()], [plot_target_correlation_heatmap()]

Examples

tbl_correlated_pairs(mtcars, cutoff = 0.85, format = "tibble")
#> # A tibble: 4 × 5
#>   feature_1 feature_2 label_1 label_2 correlation
#>   <chr>     <chr>     <chr>   <chr>         <dbl>
#> 1 cyl       disp      cyl     disp          0.902
#> 2 disp      wt        disp    wt            0.888
#> 3 mpg       wt        mpg     wt           -0.868
#> 4 mpg       cyl       mpg     cyl          -0.852