estimate_ID_twoNN.R

estimate_ID_twoNN <- function(mat_data = NULL , dist_fun_twoNN = 'canberra', 
                              maxit = 11, 
                              perc_points = 0.9, verbose = 0, ncores = my_detectCores()){
  #mat_data e' una matrice con features sulle colonne
  ID_twoNN = NULL
  sd_twoNN = NULL
  #if (perc_points > 0.9999) maxit = 1
  cl <- makePSOCKcluster(min(ncores, detectCores()-1))
  registerDoParallel(cl)
  
  ID_twoNN = foreach(nit = 1:maxit, .combine='c', .packages = c("intRinsic"), .export=c("compute_distance") ) %dopar% {
  #for(nit in 1:maxit){
    cat('.')
    if (perc_points < 0.9999){
      sub_mat_data = mat_data[sample(nrow(mat_data), size = round(nrow(mat_data)*perc_points), replace = FALSE), ]  
    }else{
      sub_mat_data = mat_data
    }
    
    dm = compute_distance(sub_mat_data, dist_fun_twoNN)
    
    if (verbose == 1 & nit ==1) cat('distance mat with shape rxc = ', nrow(dm), ' x ', ncol(dm), '\n')
    id_twonn = twonn(dist_mat = dm, method = "mle")
    
    id_twonn_est = id_twonn$est[2]

    ID_twoNN = c(ID_twoNN, id_twonn_est)
  }  
  
  stopCluster(cl)
  
  sd_ID = sd(ID_twoNN)
  ID_est = mean(ID_twoNN)
 
  if (verbose == 1){ 
    cat('\nestimated ID on ', maxit, ' iterations on the ', round(perc_points*100), '% of points = \n', 
      ID_est, '[', -sd_ID,', ', sd_ID, '] \n')
  }
  return(list(id = ID_est, sd_id = sd_ID))
  
}