ggcorr does not support "non syntactically valid" column names.

[trekonom]

When answering this question on SO I stumbled over an issue with ggcorr when column names are not syntactically valid variable names. In that case the column names are converted to syntactically valid names and all "special" symbols are replaced by dots.

A minimal reproducible example of the issue:

ex_db <- structure(list(
  SAPS = c(11L, 14L, 14L, 15L, 14L, 13L, 14L, 13L, 12L, 15L),
  `e'` = c(119, 62, 74, 75, 111, 66, 102, 71, 100, 108),
  `E/e'` = c(50, 111, 82, 68, 78, 105, 60, 91, 61, 49)
), class = "data.frame", row.names = c(NA, -10L))

library(GGally)
#> Loading required package: ggplot2
#> Registered S3 method overwritten by 'GGally':
#>   method from   
#>   +.gg   ggplot2

ggcorr(ex_db)

^{Created on 2023-05-11 with reprex v2.0.2}

IMHO the issue is that when converting the correlation matrix to a dataframe using data.frame() the default check.names=TRUE is used:

ggally/R/ggcorr.R

Line 220 in 4d8fa81

m = data.frame(m * lower.tri(m))

A "hacky" workaround would be to manipulate the ggplot object returned by ggcorr and to replace the diagLabel column containing the labels with the original column names.

Note: This requires to identify the correct geom_text layer which adds the labels stored in the diagLabel column, i.e. the one with mapping: label = ~diagLabel .

p <- ggcorr(ex_db)

p$layers[[2]]$data[c("diagLabel")] <- names(ex_db)

p

^{Created on 2023-05-11 with reprex v2.0.2}

[92amartins]

Thanks for bringing the issue!

I'm not a big fan of the "hacky" way.

Alternatively, can we just let the user specify check.names=FALSE and propagate the arg to the data.frame function?

[trekonom]

Haha. Not a fan of hacky approaches either. Was just meant as a workaround for people who can't wait until the issue is fixed.

From my understanding it would probably be safe to simply set check.names=FALSE when converting the correlation matrix to a data.frame. But I haven't checked whether this causes some issues downstream.