bash: how to split a text file in two unequal parts

Humbro · 11-22-2007, 05:06 AM

Hi,

While googling around to figure this out, I came across this thread, and thought this seemed like a friendly forum, so here's a follow-up:

I am splitting a series of text files in 10% and 90% sized parts by line count (for testing and training of a machine learning program).

The best way I have found to do this, is to use split to create ten 10% segments, and then use cat to splice nine of them together.

It works, but it seems unnecessarily roundabout. Anyone got a better idea?

BTW: I tried using head and tail instead. head worked fine, it prints the first n lines of the file as specified, but I can't get tail to print the last n lines of the file.

Thanks in advance

bigearsbilly · 11-22-2007, 05:21 AM

????

last n lines...

tail -n

syg00 · 11-22-2007, 05:35 AM

Ain't no way to do it without reading the file - maybe more than once.

Given the way (sequential) disk reads are cached, it'd be almost better to re-read the file, and split on a re-run, after all the maths is done.
Saves trashing storage by trying to do everything in memory - especially if you can't (accurately) predict the filesize. For "one-offs", sed is as easy as anything, else I'd use perl - I'm sure those that can spell awk could do it just as well.

bigearsbilly · 11-22-2007, 07:02 AM

near as dammit?

Code:

#!/usr/bin/perl


open TEN,    "> 10_percent.lis" or die "Ooops:$!";
open NINETY, "> 90_percent.lis" or die "Ooops:$!";


while (<>) {

    if ($count++ % 10) {
        print TEN;
    } else {
        print NINETY;
    }
}

Code:

billy %wc -l 1
     100 1
billy %1.pl 1
billy %wc -l *percent.lis
      90 10_percent.lis
      10 90_percent.lis
     100 total

edit Oops!
wrong way round, but you get the idea

jschiwal · 11-22-2007, 07:18 AM

This will read the file twice. First to get the number of lines, and the second time to split the file.

Code:

length=$(wc -l file)
top=$(($length/10))

sed -n "1,${top}w topfile" file  
sed -n "$(($top+1)),\$w bottomfile" file

You could also use
head -n $top file >topfile
tail -n $(($length-$top)) file >bottomfile

makyo · 11-22-2007, 11:20 PM

Hi.

If you had really long files, and if you would settle for approximations, you could estimate the number of lines in a file. The stat function is very cheap. Once you have one datum, namely the size of the file in bytes, you calculate the number of lines based on, say, 60 characters per line average.

Here is an output from such a perl script running on two files, one the first chapter of Moby Dick, second a 1 MB file of an artificial mixture of dictionary words and nonsense-created words. The script writes two files, one "ten", the other "ninety".

This allows one to make a single pass over the file ... cheers, makyo

Code:

% ./s1

 ( Lines read: 206 )
 length of /home/makyo/not-backed-up/data/moby11.txt is 12440 bytes.
 number of lines in /home/makyo/not-backed-up/data/moby11.txt estimated 207
 Comparison to wc:
 File /home/makyo/not-backed-up/data/moby11.txt has 206 lines.
 File ten has 20 lines.
 Percent = 9.700

 ( Lines read: 16569 )
 length of /home/makyo/not-backed-up/data/one-mb is 1013995 bytes.
 number of lines in /home/makyo/not-backed-up/data/one-mb estimated 16899
 Comparison to wc:
 File /home/makyo/not-backed-up/data/one-mb has 16569 lines.
 File ten has 1689 lines.
 Percent = 10.100

bigearsbilly · 11-23-2007, 03:20 AM

well, if yo get the size in bytes you can use
tail -c

Humbro · 11-23-2007, 07:15 AM

Thanks a bunch for your answers. I feel kinda stupid now, since *tail* is suddenly working like it should. Can't really say what I did wrong the last time...

So I guess that will do the trick. I'll have a look at the more advanced solutions you've posted too, though. Some of them are bound to be more optimal, I suppose. And in any case instructive