R/umap.R

In NumbersInternational/flipDimensionReduction: Algorithms for dimension reduction

##' \code{UMAP}
##' @description Produce a Uniform Manifold Approximation and Projection embedding.
##' @param data A code{\link{data.frame}} or numeric \code{\link{matrix}} containing the data to be
##' analyzed with each case as a row and each variable as a column.
##' @param n.neighbours A coefficient which determines the number of neighboring points used in local
##' approximations of manifold structured.
##' @param min.dist Controls how tightly the embedding is allowed compress points together.
##' @param seed Random seed.
##'
##' @importFrom reticulate import

#UMAP <- function(data, n.neighbours = 10, min.dist = 0.1, seed = 1066) {

#    # Convert to numeric matrix
#    data <- data.matrix(data)

#    output <- list(title = "UMAP", input.data = data)

#    # Import and use python module
#    umap <- import("umap")
#    output$embedding <- umap$UMAP(n_neighbors = as.integer(n.neighbours),
#                                  min_dist = min.dist,
#                                  random_state = as.integer(seed))$fit_transform(data)[, 1:2]

#    # Ensure first dimension has largest range
#    if (range(output$embedding)[2] > range(output$embedding)[1])
#        output$embedding[, c(1, 2)] <- output$embedding[, c(2, 1)]

#    output$input.is.distance <- FALSE
#    class(output) <- c(class(output), "UMAP")
#    return(output)
#}