magick_image_read_perf.R

library(magrittr)
# You will also nead
# stringr, dplyr, purrr, pryr, tidyr, magick
# Install magic with devtools::install('ropensci/magick') CRAN copy won't work.


# Shell code to generate some random images
# mkdir -p ./data/RandomNoise
# (cd ./data/RandomNoise && for i in `seq 512 64 4096`; do convert -size ${i}x${i} xc: +noise Random random-${i}_$i.png; done)

# Read and Prepare data ----

d <- './data/RandomNoise'
num_iter <- 10 # Number of Iterations

files <- list.files(d)
files.sizes <- purrr::map_dbl(files,function(x) file.size(sprintf("%s/%s",d,x)))

files.df <- data.frame(path=stringr::str_c(d,'/',files),
                       name=files, size=files.sizes,
                       stringsAsFactors = F) 
files.df %<>%
  dplyr::arrange(size)

# Utility Functions ----

getObjectSize <- function(obj) {
  c <- class(obj)
  if(c=='magick-image') {
    pryr::object_size(obj[[1]])
  } else {
    pryr::object_size(obj)
  }
}

getFileReadPerf <- function(file, readFunc) {
  gc()
  pryr_mem_used_before_read <- pryr::mem_used()
  t <- system.time(obj <- readFunc(file))
  #obj_class = class(obj)
  reading_function <- deparse(substitute(readFunc))
  pryr_obj_size <- getObjectSize(obj)
  pryr_mem_used_after_read <- pryr::mem_used()
  rm(obj)
  gc()
  
  data.frame(
    reading_function = reading_function,
    #obj_class = obj_class,
    User.Time= as.numeric(t[[1]]),
    System.Time= as.numeric(t[[2]]),
    Elapsed.Time= as.numeric(t[[3]]),
    Object.Size = as.numeric(pryr_obj_size),
    Mem.Increase = as.numeric(pryr_mem_used_after_read) -
      as.numeric(pryr_mem_used_before_read),
    stringsAsFactors = F
    )
}

# calculate perf numbers ----

# Calculate read performance for each file 
# We replicate the procedure `num_iter` times.
magick_image_read.perf <- purrr::map(replicate(num_iter,files.df$path),
                                      function(x) {
                                        getFileReadPerf(x,magick::image_read)})

# Combine the list into a single dataframe and add iteration and filesize columns
magick_image_read.perf.df <- dplyr::bind_rows(magick_image_read.perf) %>%
  dplyr::mutate(File.Size=as.vector(replicate(num_iter,files.df$size)),
                Iteration=rep(1:10,each=length(files))) %>%
  dplyr::select(reading_function, Iteration,
         File.Size, Object.Size, Mem.Increase,
         User.Time, System.Time, Elapsed.Time )

# This dataframe stores the means over 10 iterations
magick_image_read.perf.mean.df <- magick_image_read.perf.df %>%
  dplyr::group_by(File.Size, reading_function) %>%
  dplyr::summarise_each(dplyr::funs(mean), Object.Size, Mem.Increase,
                        User.Time, System.Time, Elapsed.Time) %>% 
  dplyr::ungroup() %>% dplyr::arrange(File.Size)

# This dataframe stores the means over 10 iterations
magick_image_read.perf.median.df <- magick_image_read.perf.df %>%
  dplyr::group_by(File.Size, reading_function) %>%
  dplyr::summarise_each(dplyr::funs(median), Object.Size, Mem.Increase,
                        User.Time, System.Time, Elapsed.Time) %>% 
  dplyr::ungroup() %>% dplyr::arrange(File.Size)