Covariate transformations available in the recipe

Which transformation a covariate needs is a modeling choice, so it is config rather than code. The original applied log(abs(x)) to a hardcoded trio of chlorophyll, integrated chlorophyll, and bathymetry (original/buildZoopModel.R:134-136).

Usage

covariate_transforms()

Value

a named list, one entry per transform, each with label, description, folds_sign, undefined_at_zero, and either fn (applied directly) or step (a recipes step function)

Details

Two properties decide what a transform can safely be given:

• folds_sign marks the transforms that take abs(x) first, because they are undefined for negative input. That is right for a magnitude stored with a sign convention — bathymetry is negative depth — and wrong for a genuinely signed covariate, where it maps -2 and 2 onto the same value. Derived covariates make the second case common: a temporal gradient, a vertical gradient, and the current components are all signed. A transform warns when the column it is given actually holds both signs.

• undefined_at_zero marks log and log10, which return -Inf there. Zeros are ordinary in chlorophyll and in any derived integral that starts at zero, so this is checked against the data rather than trusted.

boxcox and yeojohnson estimate their parameter from the data instead of applying a fixed function, so they are recipe steps rather than a function of x. Box-Cox requires strictly positive input; Yeo-Johnson does not, which makes it the one estimated option that suits a signed covariate.

References

Box GEP, Cox DR (1964). An analysis of transformations. Journal of the Royal Statistical Society: Series B 26(2), 211-252. doi:10.1111/j.2517-6161.1964.tb00553.x — boxcox

Yeo I-K, Johnson RA (2000). A new family of power transformations to improve normality or symmetry. Biometrika 87(4), 954-959. doi:10.1093/biomet/87.4.954 — yeojohnson

Field JG, Clarke KR, Warwick RM (1982). A practical strategy for analysing multispecies distribution patterns. Marine Ecology Progress Series 8, 37-52. doi:10.3354/meps008037 — the fourth root as the plankton standard

Examples

names(covariate_transforms())
#> [1] "log1p"       "log"         "log10"       "sqrt"        "fourth_root"
#> [6] "boxcox"      "yeojohnson" 
covariate_transforms()$log1p$description
#> [1] "The default, and what `log_transform` has always meant here. Defined at zero, unlike log and log10, which is why it is the one to reach for on chlorophyll and on integrals that start at zero."