Multi-threaded csv file loading

If you have a point layer with many attributes and objects in .csv, it is a good idea to use the multi-threaded loading available in data.table::fread() or vroom::vroom(). This is much faster than loading geopackage or .csv through GDAL.

Code

library("sf")
library("data.table") # library("vroom")
set.seed(1)

n = 500000
df = data.frame(x = rnorm(n), y = rnorm(n),
                col_1 = sample(c(TRUE, FALSE), n, replace = TRUE), # logical
                col_2 = sample(letters, n, replace = TRUE),        # character
                col_3 = runif(n),                                  # double
                col_4 = sample(1:100, n, replace = TRUE))          # integer
sf = st_as_sf(df, coords = c("x", "y"))

csv = tempfile(fileext = ".csv")
gpkg = tempfile(fileext = ".gpkg")
write.csv(df, csv, row.names = FALSE)
write_sf(sf, gpkg)

## sf + geopackage
system.time({
  t1 = read_sf(gpkg)
})

##    user  system elapsed 
##    1.65    0.06    1.73

## sf + csv
system.time({
  t2 = read_sf(csv, options = c("AUTODETECT_TYPE=YES",
                                "X_POSSIBLE_NAMES=x",
                                "Y_POSSIBLE_NAMES=y",
                                "KEEP_GEOM_COLUMNS=NO"))
})

##    user  system elapsed 
##    3.84    2.39    6.30

## data.table::fread + csv
system.time({
  t3 = fread(csv, nThread = 4)
  t3 = st_as_sf(t3, coords = c("x", "y"))
})

##    user  system elapsed 
##    0.21    0.03    0.14