Small helpers that return one loading, one variance percentage or a short ranked list from a PCA fit, so a Quarto or R Markdown report can write “ `r pca_loading(fit, "age", 1)` “ instead of keeping local accessor functions.
Usage
pca_loading(pca_mod, feature, pc)
pca_variance_pct(pca_mod, pc, cumulative = FALSE)
pca_top_features(pca_mod, pc, n = 5L, by = c("abs", "positive", "negative"))Arguments
- pca_mod
A `prcomp` object or a [run_pca()] result.
- feature
Name of one original variable (a row name of the rotation matrix). An unknown name is an error that lists the available names.
- pc
Component number or name (`"PC2"`). `pca_loading()` takes exactly one; `pca_variance_pct()` accepts several.
- cumulative
If `TRUE`, the percentage of variance explained by components 1 to `pc` together (each value of `pc` gives the cumulative share up to it).
- n
Number of features to return (default 5; fewer if the fit has fewer variables).
- by
`"abs"` (default) ranks by the size of the loading regardless of sign; `"positive"` and `"negative"` rank only the loadings of that sign, largest first. Ties keep the order of the variables in the fit.
Value
`pca_loading()`: one number. `pca_variance_pct()`: a numeric vector, one value per `pc`. `pca_top_features()`: a tibble with `feature`, `loading` (signed), `abs_loading` and `rank`.
Supported objects
A [stats::prcomp()] object, or the result of [run_pca()] (which carries its `prcomp` fit). Anything else is an error. [stats::princomp()] fits are not supported.
Components
`pc` is a component number (`1`, `2`, ...) or a name (`"PC1"`). A component that does not exist in the fit is an error, never `NA`. For `prcomp(..., rank. = k)` the variance percentages still use all components of the full fit (`sdev` is complete), so they are shares of the total variance.
Scale
Variance percentages are on a 0 to 100 scale (`12.3` means 12.3 percent), like `variance_pct` in [run_pca()]. Loadings are the elements of `rotation`: signed, and unit-length per component. The sign of a component is arbitrary (a refit can flip it), so describe direction relative to the other loadings of the same component.
Examples
fit <- run_pca(mtcars[, 1:5], scale = TRUE)
pca_loading(fit, "mpg", 1)
#> [1] -0.4624556
pca_variance_pct(fit, 1)
#> [1] 80.75879
pca_variance_pct(fit, 2, cumulative = TRUE)
#> [1] 92.00349
pca_top_features(fit, 1, n = 3)
#> # A tibble: 3 × 4
#> feature loading abs_loading rank
#> <chr> <dbl> <dbl> <int>
#> 1 cyl 0.477 0.477 1
#> 2 disp 0.473 0.473 2
#> 3 mpg -0.462 0.462 3