
Last chance! 50% off unlimited learning
Sale ends in
Saves (serializes) an object to disk.
qsave(x, file,
preset = "high", algorithm = "zstd", compress_level = 4L,
shuffle_control = 15L, check_hash=TRUE, nthreads = 1)
the object to serialize.
the file name/path.
One of "fast", "high" (default), "high", "archive", "uncompressed" or "custom". See details.
Compression algorithm used: "lz4", "zstd", "lz4hc", "zstd_stream" or "uncompressed".
The compression level used (Default 1). For lz4, this number must be > 1 (higher is less compressed). For zstd, a number between -50 to 22 (higher is more compressed).
An integer setting the use of byte shuffle compression. A value between 0 and 15 (Default 15). See details.
Default TRUE, compute a hash which can be used to verify file integrity during serialization
Number of threads to use. Default 1.
The total number of bytes written to the file (returned invisibly)
This function serializes and compresses R objects using block compresion with the option of byte shuffling. There are lots of possible parameters. This function exposes three parameters related to compression level and byte shuffling.
`compress_level` - Higher values tend to have a better compression ratio, while lower values/negative values tend to be quicker. Due to the format of qs, there is very little benefit to compression levels > 5 or so.
`shuffle_control` - This sets which numerical R object types are subject to byte shuffling. Generally speaking, the more ordered/sequential an object is (e.g., `1:1e7`), the larger the potential benefit of byte shuffling. It is not uncommon to have several orders magnitude benefit to compression ratio or compression speed. The more random an object is (e.g., `rnorm(1e7)`), the less potential benefit there is, even negative benefit is possible. Integer vectors almost always benefit from byte shuffling whereas the results for numeric vectors are mixed. To control block shuffling, add +1 to the parameter for logical vectors, +2 for integer vectors, +4 for numeric vectors and/or +8 for complex vectors.
The `preset` parameter has several different combination of parameter sets that are performant over a large variety of data. The `algorithm` parameter, `compression_level` and `shuffle_control` parameters are ignored unless `preset` is "custom". "fast" preset: algorithm lz4, compress_level 100, shuffle_control 0. "balanced" preset: algorithm lz4, compress_level 1, shuffle_control 15. "high" preset: algorithm zstd, compress_level 4, shuffle_control 15. "archive" preset: algorithm zstd_stream, compress_level 14, shuffle_control 15. (zstd_stream is currently single threaded only)
# NOT RUN {
x <- data.frame(int = sample(1e3, replace=TRUE),
num = rnorm(1e3),
char = randomStrings(1e3), stringsAsFactors = FALSE)
myfile <- tempfile()
qsave(x, myfile)
x2 <- qread(myfile)
identical(x, x2) # returns true
# qs support multithreading
qsave(x, myfile, nthreads=2)
x2 <- qread(myfile, nthreads=2)
identical(x, x2) # returns true
# Other examples
z <- 1:1e7
myfile <- tempfile()
qsave(z, myfile)
z2 <- qread(myfile)
identical(z, z2) # returns true
w <- as.list(rnorm(1e6))
myfile <- tempfile()
qsave(w, myfile)
w2 <- qread(myfile)
identical(w, w2) # returns true
# }
Run the code above in your browser using DataLab