Supervised and unsupervised uniform manifold approximation and projection (UMAP)Source:
step_umap() creates a specification of a recipe step that will project a
set of features into a smaller space.
step_umap( recipe, ..., role = "predictor", trained = FALSE, outcome = NULL, neighbors = 15, num_comp = 2, min_dist = 0.01, metric = "euclidean", learn_rate = 1, epochs = NULL, options = list(verbose = FALSE, n_threads = 1), seed = sample(10^5, 2), prefix = "UMAP", keep_original_cols = FALSE, retain = deprecated(), object = NULL, skip = FALSE, id = rand_id("umap") )
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
selections()for more details.
For model terms created by this step, what analysis role should they be assigned? By default, the new columns created by this step from the original variables will be used as predictors in a model.
A logical to indicate if the quantities for preprocessing have been estimated.
A call to
varsto specify which variable is used as the outcome in the encoding process (if any).
An integer for the number of nearest neighbors used to construct the target simplicial set. If
neighborsis greater than the number of data points, the smaller value is used.
An integer for the number of UMAP components. If
num_compis greater than the number of selected columns minus one, the smaller value is used.
The effective minimum distance between embedded points.
Character, type of distance metric to use to find nearest neighbors. See
uwot::umap()for more details. Default to
Positive number of the learning rate for the optimization process.
Number of iterations for the neighbor optimization. See
uwot::umap()for more details.
A list of options to pass to
uwot::umap(). The arguments
learning_rateshould not be passed here. By default,
Two integers to control the random numbers used by the numerical methods. The default pulls from the main session's stream of numbers and will give reproducible results if the seed is set prior to calling
A character string for the prefix of the resulting new variables. See notes below.
A logical to keep the original variables in the output. Defaults to
keep_original_colsinstead to specify whether the original predictors should be retained along with the new embedding variables.
An object that defines the encoding. This is
NULLuntil the step is trained by
A logical. Should the step be skipped when the recipe is baked by
bake()? While all operations are baked when
prep()is run, some operations may not be able to be conducted on new data (e.g. processing the outcome variable(s)). Care should be taken when using
skip = TRUEas it may affect the computations for subsequent operations.
A character string that is unique to this step to identify it.
An updated version of
recipe with the new step added to the
sequence of any existing operations.
UMAP, short for Uniform Manifold Approximation and Projection, is a nonlinear dimension reduction technique that finds local, low-dimensional representations of the data. It can be run unsupervised or supervised with different types of outcome data (e.g. numeric, factor, etc).
num_comp controls the number of components that will be retained
(the original variables that are used to derive the components are removed from
the data). The new components will have names that begin with
prefix and a
sequence of numbers. The variable names are padded with zeros. For example, if
num_comp < 10, their names will be
num_comp = 101,
the names would be
tidy() this step, a tibble with columns
(the selectors or variables selected) is returned.
This step has 5 tuning parameters:
num_comp: # Components (type: integer, default: 2)
neighbors: # Nearest Neighbors (type: integer, default: 15)
min_dist: Min Distance between Points (type: double, default: 0.01)
learn_rate: Learning Rate (type: double, default: 1)
epochs: # Epochs (type: integer, default: NULL)
This recipe step may require native serialization when saving for use in another R session. To learn more about serialization for prepped recipes, see the bundle package.
McInnes, L., & Healy, J. (2018). UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction. https://arxiv.org/abs/1802.03426.
"How UMAP Works" https://umap-learn.readthedocs.io/en/latest/how_umap_works.html
library(recipes) library(ggplot2) split <- seq.int(1, 150, by = 9) tr <- iris[-split, ] te <- iris[split, ] set.seed(11) supervised <- recipe(Species ~ ., data = tr) %>% step_center(all_predictors()) %>% step_scale(all_predictors()) %>% step_umap(all_predictors(), outcome = vars(Species), num_comp = 2) %>% prep(training = tr) theme_set(theme_bw()) bake(supervised, new_data = te, Species, starts_with("umap")) %>% ggplot(aes(x = UMAP1, y = UMAP2, col = Species)) + geom_point(alpha = .5)