R/c_H2OKMeans.R

In egenn/rtemis: Machine Learning and Visualization

# c_H2OKMeans.R
# ::rtemis::
# 2017 E.D. Gennatas www.lambdamd.org

#' K-Means Clustering with H2O
#'
#' Perfomr [K-Means clustering](https://en.wikipedia.org/wiki/K-means_clustering) using 
#' `h2o::h2o.kmeans`
#' 
#' Check out the H2O Flow at `[ip]:[port]`, Default IP:port is "localhost:54321"
#' e.g. if running on localhost, point your web browser to `localhost:54321`
#' For additional information, see help on `h2o::h2o.kmeans`
#' 
#' @inheritParams c_KMeans
#' @param estimate.k Logical: if TRUE, estimate k up to a maximum set by the `k` argument
#' @param nfolds Integer: Number of cross-validation folds
#' @param max.iterations Integer: Maximum number of iterations
#' @param ip Character: IP address of H2O server. Default = "localhost"
#' @param port Integer: Port number of H2O server. Default = 54321
#' @param seed Integer: Seed for H2O's random number generator. Default = -1 (time-based ranodm number)
#' @param init Character: Initialization mode: "Furthest", "Random", "PlusPlus", "User".
#' Default = "Furthest"
#' @param categorical.encoding Character: How to encode categorical variables: "AUTO", "Enum", "OneHotInternal",
#' "OneHotExplicit", "Binary", "Eigen", "LabelEncoder", "SortByResponse", "EnumLimited".
#' Default = "AUTO"
#' @param n.cores Integer: Number of cores to use
#' @param ... Additional arguments to pass to `h2p::h2o.kmeans`
#' 
#' @return `rtMod` object
#' @author E.D. Gennatas
#' @family Clustering
#' @export

c_H2OKMeans <- function(x, x.test = NULL,
                        k = 2,
                        estimate.k = FALSE,
                        nfolds = 0,
                        max.iterations = 10,
                        ip = "localhost",
                        port = 54321,
                        n.cores = rtCores,
                        seed = -1,
                        init = c("Furthest", "Random", "PlusPlus", "User"),
                        categorical.encoding = c("AUTO", "Enum", "OneHotInternal", "OneHotExplicit", "Binary",
                        "Eigen", "LabelEncoder", "SortByResponse", "EnumLimited"),
                        verbose = TRUE, ...) {

  # Intro ----
  start.time <- intro(verbose = verbose)
  clust.name <- "H2OKMeans"

  # Data ----
  if (is.null(colnames(x))) colnames(x) <- paste0("Feature_", seq_len(NCOL(x)))
  x <- as.data.frame(x)
  xnames <- colnames(x)

  # Dependencies ----
  dependency_check("h2o")

  # Arguments ----
  init <- match.arg(init)
  categorical.encoding <- match.arg(categorical.encoding)

  # Data ----
  # h2o Frames
  if (verbose) msg2("Connecting to H2O server...")
  h2o::h2o.init(ip = ip, port = port, nthreads = n.cores)
  if (verbose) msg2("Creating H2O frames...")
  df.train <- h2o::as.h2o(data.frame(x), "df_train")
  if (!is.null(x.test)) {
    df.test <- h2o::as.h2o(data.frame(x.test), "df_test")
  } else {
    df.test <- NULL
  }

  # H2OKMEANS ----
  if (verbose) msg2("Performing K-Means Clustering using H2O with k = ", k, "...", sep = "")
  clust <- h2o::h2o.kmeans(training_frame = df.train,
                           model_id = paste0("rtemis.H2OKMEANS.", format(Sys.time(), "%b%d.%H:%M:%S.%Y")),
                           k = k,
                           estimate_k = estimate.k,
                           nfolds = nfolds,
                           max_iterations = max.iterations,
                           seed = seed,
                           init = init,
                           categorical_encoding = categorical.encoding, ...)

  # Clusters ----
  clusters.train <- as.data.frame(h2o::h2o.predict(clust, df.train))[, 1] + 1
  if (!is.null(x.test)) {
    clusters.test <- as.data.frame(h2o::h2o.predict(clust, df.test))[, 1] + 1
  } else {
    clusters.test <- NULL
  }

  # Outro ----
  extra <- list(centroids = clust@model$centers)
  cl <- rtClust$new(clust.name = clust.name,
                    k = k,
                    xnames = xnames,
                    clust = clust,
                    clusters.train = clusters.train,
                    clusters.test = clusters.test,
                    parameters = list(k = k,
                                      estimate.k = estimate.k,
                                      nfolds = nfolds,
                                      max.iterations = max.iterations,
                                      seed = seed,
                                      init = init,
                                      categorical.encoding = categorical.encoding),
                    extra = extra)
  outro(start.time, verbose = verbose)
  cl

} # rtemis::c_H2OKMeans