R package {bigreadr}

# remotes::install_github("privefl/bigreadr")
library(bigreadr)

# Create a temporary file of ~141 MB (just as an example)
csv <- fwrite2(iris[rep(seq_len(nrow(iris)), 1e4), rep(1:5, 4)], tempfile())
format(file.size(csv), big.mark = ",")

## Splitting lines (1)
# Read (by parts) all data -> using `fread` would be faster
nlines(csv)  ## 1M5 lines -> split every 500,000
big_iris1 <- big_fread1(csv, every_nlines = 5e5)
# Read and subset (by parts)
big_iris1_setosa <- big_fread1(csv, every_nlines = 5e5, .transform = function(df) {
  dplyr::filter(df, Species == "setosa")
})

## Splitting columns (2)
big_iris2 <- big_fread2(csv, nb_parts = 3)
# Read and subset (by parts)
species_setosa <- (fread2(csv, select = 5)[[1]] == "setosa")
big_iris2_setosa <- big_fread2(csv, nb_parts = 3, .transform = function(df) {
  dplyr::filter(df, species_setosa)
})

## Verification
identical(big_iris1_setosa, dplyr::filter(big_iris1, Species == "setosa"))
identical(big_iris2, big_iris1)
identical(big_iris2_setosa, big_iris1_setosa)