Premature conversion of factors to numeric

[goergen95]

I think that the conversion of factor variables to numeric before applying aoa() causes unintended treatment of categorical variables as numeric. In fact, conversion to dummy variables is intended to happen within aoa() which should not trigger if converted to numeric early.
(CC: @Nowosad)

CAST/R/aoa.R

Lines 187 to 188 in 716a035

	if (any(is.factor(newdata)))
	newdata[[which(is.factor(newdata))]] <- as.numeric(newdata[[which(is.factor(newdata))]])

[goergen95]

Interestingly, is.factor() delivers a different result depending whether or not terra is attached in the current session:

v <- rep(c(1,2,3), 3)
v <- factor(v, labels = c("a", "b", "c"))
r <- terra::rast(matrix(v, ncol=3))
is.factor(r)
#> [1] FALSE
library(terra)
#> terra 1.9.1
is.factor(r)
#> [1] TRUE

^{Created on 2026-03-26 with reprex v2.1.1}

[Nowosad]

I think that the first call is base::is.factor and the second call is terra::is.factor..

[goergen95]

So this delivers different results depending on if the user loaded terra or not since CAST never loads it. I guess that is the reason why this slipped CI, since this code path is never triggerd in automated runs. I'll open an PR removing this code since categorical variable handling should anyway be handled inside aoa().