CODEWITHSUNDEEP
shellunixcommand-lineawk
Jadu Saikia
Jadu Saikia

Reputation: 21

Looking for a efficient split of file

Assuming this is my file:

$ cat file.txt 
A:1:i
B:2:ii
X:9:iv

With a for loop like this I can print all the fields separately and redirect to sub-file

$ for i in $(seq 1 3); do echo $i; awk -F ":" -v FL=$i '{print $FL}' file.txt > $i.out; done

So that:

$ cat 1.out 
A
B
X

$ cat 2.out 
1
2
9

$ cat 3.out 
i
ii
iv

Question: I have to perform this on nearly 70 columns and on file size of nearly 10 GB. It works, but slow. Can anyone suggest a better/efficient split to work on this big data set. Thanks. 

$ for i in $(seq 1 70); do echo $i; awk -F ":" -v FL=$i '{print $FL}' *.data > $i.out; done

Upvotes: 1

Views: 133

Answers (5)

chepner
chepner

Reputation: 531718

Here's a bash script that uses a feature I don't see often: asking bash to allocate a file descriptor for a file and storing the descriptor in a variable:

# Read the first line to get a count of the columns
IFS=: read -a columns < file.txt

# Open an output file for each column, saving the file descriptor in an array
for c in "${columns[@]}"; do
    exec {a}>$((++i)).txt
    fds+=( $a )
done

# Iterate through the iput, writing each column to the file opened for it
while IFS=: read -a fields; do
    for f in "${fields[@]}"; do
        printf "$f\n" >&${fds[++i]}
    done
done < file.txt

# Close the file descriptors
for fd in "${fds[@]}"; do
    exec {fd}>&-
done

Upvotes: 0

Thor
Thor

Reputation: 47169

With coreutils if you know that there are three columns:

< file.txt tee >(cut -d: -f1 > 1.out) >(cut -d: -f2 > 2.out) >(cut -d: -f3 > 3.out) > /dev/null

To make it more generic, here's one way to automate the command-line generation:

# Determine number of fields and generate tee argument
arg=""
i=1
while read; do 
  arg="$arg >(cut -d: -f$i > $((i++)).out)"
done < <(head -n1 file.txt | tr ':' '\n')

arg is now:

>(cut -d: -f1 > 1.out) >(cut -d: -f2 > 2.out) >(cut -d: -f3 > 3.out)

Save to a script file:

echo "< file.txt tee $arg > /dev/null" > script

And execute:

. ./script

Upvotes: 0

jedwards
jedwards

Reputation: 30230

Python version

#!/bin/env python

with open('file.txt', 'r') as ih:
    while True:
        line = ih.readline()
        if line == '': break
        for i,element in enumerate(line.strip().split(':')):
            outfile = "%d.out" % (i+1)
            with open(outfile, 'a') as oh:
                oh.write("%s\n" % element)

This might be a bit faster, as it only goes through the original file once. Note that it could be further optimized by leaving the output files open (as it is, I close each of them and re-open them for each write).

EDIT

For example, something like:

#!/bin/env python

handles = dict()

with open('file.txt', 'r') as ih:
    while True:
        line = ih.readline()
        if line == '': break
        for i,element in enumerate(line.strip().split(':')):
            outfile = "%d.out" % (i+1)

            if outfile not in handles:
                handles[outfile] = open(outfile, 'a');

            handles[outfile].write("%s\n" % element)

for k in handles:
    handles[k].close()

This leaves the handles open for the duration of the execution, then closes them all before continuing / ending.

Upvotes: 2

epsalon
epsalon

Reputation: 2294

In perl you can do: 

#!/usr/bin/perl -w
my $n = 3;
my @FILES;
for my $i (1..$n) {
  my $f;
  open ($f, "> $i.out") or die;
  push @FILES, $f;
}
while (<>) {
  chomp;
  @a = split(/:/);
  for my $i (0..$#a) {
    print $FILES[$i] $a[$i],"\n";
  }
}
close($f) for $f in @FILES;

Upvotes: 1

Steve
Steve

Reputation: 54502

This should be fairly quick considering what you are trying to do:

awk -F: '{ for (i=1; i<=NF; i++) print $i > i".out" }' file.txt

Upvotes: 6

Related Questions