`step_famd()` creates a *specification* of a recipe step that converts a mix of numeric and categorical variables into principal components with Factor Analysis of Mixed Data ([FactoMineR::FAMD()]). It is the mixed-data counterpart of [recipes::step_pca()]: numeric variables are standardized and categorical variables are weighted, so each variable contributes on a comparable scale.
Usage
step_famd(
recipe,
...,
role = "predictor",
trained = FALSE,
num_comp = 2,
threshold = NA,
options = list(),
res = NULL,
columns = NULL,
levels = NULL,
prefix = "FAMD",
keep_original_cols = FALSE,
skip = FALSE,
id = recipes::rand_id("famd")
)
# S3 method for class 'step_famd'
tidy(x, type = "coef", ...)Arguments
- recipe
A recipe object. The step will be added to the sequence of operations for this recipe.
- ...
One or more selector functions to choose variables for this step. See [recipes::selections()].
- role
Role of the new component columns (default `"predictor"`).
- trained
A logical indicating whether the step has been trained.
- num_comp
Number of components to keep (default 2); ignored when `threshold` is set. `0` leaves the data unchanged.
- threshold
Fraction of total variance, in (0, 1], that the kept components should cover. Overrides `num_comp` when not `NA`.
- options
A named list of further arguments to [FactoMineR::FAMD()].
- res
The fitted FAMD object, once trained.
- columns
The selected column names, once trained.
- levels
The factor levels of the categorical columns, once trained.
- prefix
Prefix of the new column names (default `"FAMD"`), numbered by [recipes::names0()]. It differs from [recipes::step_pca()]'s `"PC"`, so both steps can be used in one recipe.
- keep_original_cols
A logical to keep the original variables in the output. Defaults to `FALSE`.
- skip
A logical. Should the step be skipped when the recipe is baked by [recipes::bake()]?
- id
A character string that is unique to this step.
- x
A `step_famd` object.
- type
For `tidy()`: `"coef"` (variable contributions) or `"variance"`.
Details
The selected variables must include at least one numeric and one categorical (factor, character, or logical) variable; use [recipes::step_pca()] for all-numeric data. Character and logical variables are treated as factors, with levels learned by `prep()`. Variables of any other type, such as dates, are an error in `prep()`: convert them first, e.g. with [recipes::step_date()] or [recipes::step_mutate()]. A variable that was numeric in training must be numeric in the data given to `bake()`. Missing values, and categories that `bake()` meets but training did not, are errors: impute or collapse levels first, e.g. with [recipes::step_impute_mode()] or [recipes::step_other()].
FAMD has at most \(\min(n - 1, p + \sum_j (L_j - 1))\) dimensions, for \(n\) rows, \(p\) numeric variables, and categorical variables with \(L_j\) levels, which can be more than the number of variables. `num_comp` is capped there. With `threshold`, the step keeps the fewest components whose cumulative share of variance reaches `threshold`, choosing among all dimensions.
Frequency weights ([hardhat::frequency_weights()]) are passed to FAMD as row weights. Like other unsupervised recipe steps, such as [recipes::step_pca()], it ignores importance weights.
# Tidying
`tidy()` with `type = "coef"` returns each variable's percentage contribution to each component (`value`): FactoMineR's per-variable summary, which is comparable between numeric and categorical variables. With `type = "variance"`, it returns each component's variance, cumulative variance, percent variance, and cumulative percent variance, as for [recipes::step_pca()].
References
Pagès J (2004). Analyse factorielle de données mixtes. *Revue de Statistique Appliquée*, 52(4), 93-111.
Examples
if (requireNamespace("FactoMineR", quietly = TRUE)) {
library(recipes)
rec <- recipe(~ ., data = iris) |>
step_famd(all_predictors(), num_comp = 3)
prepped <- prep(rec)
head(bake(prepped, new_data = NULL))
tidy(prepped, number = 1, type = "variance")
}
#> Loading required package: dplyr
#>
#> Attaching package: ‘dplyr’
#> The following objects are masked from ‘package:stats’:
#>
#> filter, lag
#> The following objects are masked from ‘package:base’:
#>
#> intersect, setdiff, setequal, union
#>
#> Attaching package: ‘recipes’
#> The following object is masked from ‘package:stats’:
#>
#> step
#> Warning: did not converge--results might be invalid!; try increasing maxit or work
#> # A tibble: 12 × 4
#> terms value component id
#> <chr> <dbl> <int> <chr>
#> 1 variance 3.87 1 famd_iouKO
#> 2 variance 1.34 2 famd_iouKO
#> 3 variance 0.559 3 famd_iouKO
#> 4 cumulative variance 3.87 1 famd_iouKO
#> 5 cumulative variance 5.21 2 famd_iouKO
#> 6 cumulative variance 5.77 3 famd_iouKO
#> 7 percent variance 64.5 1 famd_iouKO
#> 8 percent variance 22.4 2 famd_iouKO
#> 9 percent variance 9.32 3 famd_iouKO
#> 10 cumulative percent variance 64.5 1 famd_iouKO
#> 11 cumulative percent variance 86.9 2 famd_iouKO
#> 12 cumulative percent variance 96.2 3 famd_iouKO