sample-lines 0.0.2

Install

Install like so.

pip install sample-lines

How to

See the help for documentation.

sample-lines -h
usage: Randomly select lines from a file. [-h] [--sample-size N]
                                          [--method {simple-random,systematic}]
                                          [--repeat REPEAT]
                                          file

positional arguments:
  file

optional arguments:
  -h, --help            show this help message and exit
  --sample-size N, -n N
                        Number of lines to emit
  --method {simple-random,systematic}, -m {simple-random,systematic}
                        Sampling method
  --repeat REPEAT, -r REPEAT
                        Number of repetitions for systematic sampling

Samples are with replacement and weighted by line length. The probability of selecting a line is proportional to its line length. This allows us to sample very quickly, but it makes this approach appropriate only if your file has reasonably consistent line lengths or if you don’t care about short lines.

How fast

Consider this 1-gigabyte CSV file.

$ wc big-file.csv
 2388430 27673790 1071895374 big-file.csv

Running wc took three seconds.

time wc big-file.csv
 2388430 27673790 1071895374 big-file.csv

real    0m3.789s
user    0m3.560s
sys     0m0.190s

sample-lines is much faster. Here’s a simple random sample of 40 lines,

$ time sample-lines -n 40 -m simple-random big-file.csv > /dev/null

real    0m0.136s
user    0m0.113s
sys     0m0.018s

a systematic sample of 40 lines,

$ time sample-lines -n 40 -m systematic -r 4 big-file.csv > /dev/null

real    0m0.148s
user    0m0.122s
sys     0m0.019s

and repeated systematic sample, with 4 repeats and 10 lines each, for a total of 40 lines.

$ time sample-lines -n 10 -m systematic -r 4 big-file.csv > /dev/null

real    0m0.175s
user    0m0.140s
sys     0m0.025s

Most of the time in the above examples was spent loading Python and the various modules; printing the help takes almost as long as running the sample.

$ time sample-lines -h > /dev/null

real    0m0.157s
user    0m0.129s
sys     0m0.021s

So even a pretty big sample is still fast to run.

$ time sample-lines -n 2000 -m systematic -r 50 big-file.csv > /dev/null

real    0m2.695s
user    0m2.435s
sys     0m0.231s

Alternatives

Use sample if you want to sample from a stream.