R/gbm.subset.R

In gbm.auto: Automated Boosted Regression Tree Modelling and Mapping Suite

#' Subset gbm.auto input datasets to 2 groups using the partial deviance plots
#'
#' Set your working directory to the output folder of a gbm.auto/gbm.loop run.
#' This function returns the variable value corresponding to the 0 value on the
#' lineplots, which should be the optimal place to split the dataset into 2
#' subsets, low and high, IF the relationship doesn't cross 0 more than once.
#' Function is similarly useful to quickly get the 0-point value in these cases,
#'  i.e. where values below are detrimental, values above beneficial (check
#'  plots though)
#'
#' loop varnames are BinLineLoop_VAR.csv & GausLineLoop_VAR.csv
#' normal varnames are Bin_Best_line_VAR.csv & Gaus_Best_line_VAR.csv
#'
#' Just use average between the last negative & first positive point
#' unless any points fall on zero

#' @param x Vector of variable names.
#' @param fams Vector of statistical data distribution family names to be modelled by gbm.
#' @param loop Is the folder a gbm.loop output?
#'
#' @return a list of breakpoint values which datasets can be subsetted using.
#' @export
#' @importFrom utils read.csv
#' @author Simon Dedman, \email{simondedman@@gmail.com}
#' @examples
#' \donttest{
#' # Not run: requires completed gbm.auto run.
#' # having run gbm.auto (with linesfiles=TRUE), set working directory there
#' data(samples)
#' gbm.subset(x = names(samples[c(4:8, 10)]), fams = c("Bin", "Gaus"))
#' }
#'
gbm.subset <- function(x, #Vector of variable names.
                       fams = c("Bin", "Gaus"), # Vector of family names modelled by gbm.
                       loop = FALSE) { #is the folder a gbm.loop output?
  subsetsplits <- list() #create blank list object
  for (j in fams) { #loop through families
    for (i in x) { #loop through variable names' files
      if (loop) {if (file.exists(paste0(j, "LineLoop_", i, ".csv"))) { #if file exists
        tmp <- read.csv(paste0(j, "LineLoop_", i, ".csv")) #read in csv file
        tmp <- tmp[, c(1, length(tmp) - 2)]} #keep only X & averageY
      } else {#if not loop
        if (file.exists(paste0(j, "_Best_line_", i, ".csv"))) {#if file exists
          tmp <- read.csv(paste0(j, "_Best_line_", i, ".csv"))}} #read in csv file
      if (exists("tmp")) { #if csv file was read (x names used in gbm may not have generated files)
        if (!is.na(match(0, sign(tmp[, 2])))) { # if there's an exact 0 value in the Y column,
          subsetsplits[[i]] <- tmp[match(0, sign(tmp[, 2])), 1] #set the corresponding X value as a list item named i
        } else {# if there isn't (normal)
          row1 <- which(diff(sign(tmp[, 2])) != 0) #gives the last row before crossing the Y=0 intercept
          row2 <- row1 + 1 #first row after
          subsetsplits[[i]] <- mean(c(tmp[row1, 1], #set list value i as average of the points
                                      tmp[row2, 1])) #before & after intercept crossing
        } #close 'point on 0 line' ifelse
        rm(tmp) #remove tmp if it was created
      } #close if exists tmp
    } #close var.names loop
  } #close fam.names loop
  return(subsetsplits) #return object
} #close function