Tune hyperparameters for LoCoMDS

Usage

tune_locoMDS(obj, data, ks = NULL, make_plots = TRUE)

Arguments

obj: Output of locoMDS().
data: Original data used to evaluate LCMC.
ks: Vector of neighborhood sizes to evaluate LCMC. If NULL, defaults to a range of sizes between 1 and 70% of the number of samples in data.
make_plots: Logical indicating whether to create a plot of LCMC values.

Value

A list with two components:

lcmc: A data frame containing the LCMC values for each neighborhood size and hyperparameter combination.
plots: A ggplot object showing the LCMC values across neighborhood sizes and hyperparameters.

Examples

data(iris)
# remove duplicates so that tSNE can run
X <- dplyr::distinct(iris[, 1:4])

# fit various dimension reduction methods
pca_scores <- prcomp(X, center = TRUE, scale = TRUE)$x
tsne_scores <- Rtsne::Rtsne(X, dims = 2, perplexity = 30, verbose = FALSE)$Y
umap_scores <- umap::umap(X, n_components = 2, verbose = FALSE)$layout
dr_list <- list(
  pca = pca_scores,
  tsne = tsne_scores,
  umap = umap_scores
)

# fit LoCoMDS with multiple possible hyperparameters
locomds_fits_all <- locoMDS(
  embed_list = dr_list, ndim = 2, tau = c(0.01, 0.1), percentile = c(0.5, 0.8)
)

# tune LoCoMDS hyperparameters
locomds_tuning <- tune_locoMDS(locomds_fits_all, data = X)