R/1a_ds_points.R

In dabrowskia/dspace: Dividing Space - Data Driven Segementation of Spatial Data

#' ds_points
#'
#' Creates a vector of community assignment based on neighboring points. It is based
#' on a topological network structure where points represent nodes and the edges
#' are the degree of similarity between those nodes. Communities are created using fast greedy
#' algorithm that maximizes their modularity.
#'
#' @param x point or polygon shapefile data;
#' @param k number of communities;
#' @param n.neigh number of neighbors considered in the k-nearest neighbor
#'   algorithm that builds topology network
#' @param data attributes of the spatial data frame to calculate similarity or
#'   distance measure;
#' @param similarity.measure Character or function to declare distance method 
#'   transformed into similarity measure. If method is
#'   character, method must be "mahalanobis" or "euclidean", "maximum",
#'   "manhattan", "canberra", "binary" or "minkowski". If method is one of
#'   "euclidean", "maximum", "manhattan", "canberra", "binary" or "minkowski",
#'   see dist for details, because this function is used to compute the distance.
#'   If method is set to "mahalanobis", the mahalanobis distance is computed between
#'   neighbor points. If method is a function, this function is used to compute
#'   the distance.
#' @param style style can take values “W”, “B”, “C”, “U”, “minmax” and “S” (see spdep::nb2listw)
#' @param plot should the neighborhood be plotted
#' @param explain logical. If TRUE a machine learning (randomForest 
#' using 5 fold cross validation) model is being constructed based 
#' on the data provided for regionalization. The accuracy of this model
#' explains how much of the regionalization can be attributed to the data
#' and how much to the spatial distribution.
#'  
#' @return vector of numbers representing regions to which each element belongs to


ds_points <- function(x,
                      k = 2,
                      data = -grep(names(x), pattern = '^geom'),
                      similarity.measure = "euclidean",
                      style = "B",
                      n.neigh = 8,
                      plot = TRUE,
                      explain = TRUE)
{
  #First step is to prepare the points for further analysis 
  res <- prepare_points(
    x = x,
    n.neigh = n.neigh,
    plot = plot
  )
  #Building network/graph representation out of neighbor representation
  fg.graph <- build_graph(
      x = res[["x"]],
      x.nb = res[["x.nb"]],
      data = data,
      similarity.measure = similarity.measure,
      style = style
  )
  #Dividing the points based on their graph representation using fast greedy algorithm
  classes <- part_communities(fg.graph = fg.graph[["fg"]], k = k)
  #Calculating the accuracy based on random forest algorithm for evaluating suitability of the partition
  if (explain == TRUE)
  {
    data <- names(x)[data]
    data.to.accu <-
      sf::st_drop_geometry(res[["x"]]) %>%
      dplyr::select(data) %>%
      dplyr::mutate(class = classes)
    accu <- ds_accuracy(data.to.accu = data.to.accu)
    print(paste(accu*100, 
                'percent of the regionalization process', 
                'can be attributed to the data itself while', 
                'the rest is due to spatial location (neghborhoods)'))
  }
  classes
}