168

I have a utility script in Python:

#!/usr/bin/env python
import sys
unique_lines = []
duplicate_lines = []
for line in sys.stdin:
  if line in unique_lines:
    duplicate_lines.append(line)
  else:
    unique_lines.append(line)
    sys.stdout.write(line)
# optionally do something with duplicate_lines

This simple functionality (uniq without needing to sort first, stable ordering) must be available as a simple UNIX utility, mustn't it? Maybe a combination of filters in a pipe?

Reason for asking: needing this functionality on a system on which I cannot execute Python from anywhere.

Improve this question
3
  • 4
    Unrelated: you should really use a set rather than a list in that Python script; checking for membership in a list is a linear-time operation. Commented Jul 17, 2012 at 23:18
  • I removed "Python" from your tags and title since this really has nothing to do with Python. Commented Jul 17, 2012 at 23:20
  • 1
    if this had to be done in Python a better approach would involve using the uniq_everseen itertools recipe: docs.python.org/library/itertools.html#recipes Commented Jul 23, 2012 at 17:02

8 Answers 8

Reset to default
406

The UNIX Bash Scripting blog suggests:

awk '!x[$0]++'

This command is telling awk which lines to print. The variable $0 holds the entire contents of a line and square brackets are array access. So, for each line of the file, the node of the array x is incremented and the line printed if the content of that node was not (!) previously set.

Improve this answer
Sign up to request clarification or add additional context in comments.

16 Comments

For a short awk statement like this (no curly brackets involved), the command is simply telling awk which lines to print. The variable $0 holds the entire contents of a line and square brackets are array access. So, for each line of the file, we are incrementing a node of the array named x and printing the line if the content of that node was not (!) previously set.
Surely it would be less obfuscated to name that array e.g. seen instead of x, to avoid giving newbies the impression that awk syntax is line noise
Keep in mind that this will load the entire file into memory, so don't try this on a 3GB text file without lots of RAM to spare.
@Hitechcomputergeek This won't necessarily load the whole file into memory, only the unique lines. This of course could end up being the whole file though if all the lines are unique.
stackoverflow.com/a/1444448/44620 with a detailed description of how it works.
|
111

A late answer - I just ran into a duplicate of this - but perhaps worth adding...

The principle behind @1_CR's answer can be written more concisely, using cat -n instead of awk to add line numbers:

cat -n file_name | sort -uk2 | sort -n | cut -f2-
  • Use cat -n to prepend line numbers
  • Use sort -u remove duplicate data (-k2 says 'start at field 2 for sort key')
  • Use sort -n to sort by prepended number
  • Use cut to remove the line numbering (-f2- says 'select field 2 till end')
Improve this answer

8 Comments

Easy to understand, and this is often valuable. Any ideas of performance with big files against shortest Michael Hoffman's solution above?
More readable/maintainable. Needed the same but with a reverse sort to keep only the last occurrence of each unique value. Using both --reverse and --unique in the same sort command doesn't return the results one might expect. Apparently, sort does a premature optimization by 1st applying --unique on the input (in order to reduce processing in subsequent steps). This removes data needed for the --reverse step too early. To fix this, insert a sort --reverse -k2 as the 1st sort in the pipeline: cat -n file_name | sort -rk2 | sort -uk2 | sort -nk1 | cut -f2-
Took just 60 seconds for a 900MB+ text file with so many (randomly placed) duplicate lines that the result is only 39KB. Sufficiently fast.
"Pipe" version: cat file_name | cat -n | sort -uk2 | sort -nk1 | cut -f2-.
"Pipe" version for keeping last occurrence instead of first one: cat file_name | cat -n | sort -rk2 | sort -uk2 | sort -nk1 | cut -f2-.
|
10

To remove duplicate from 2 files : 

awk '!a[$0]++' file1.csv file2.csv
Improve this answer

Comments

8

uq

uq is a small tool written in Rust. It performs uniqueness filtering without having to sort the input first, therefore can apply on continuous stream.

There are two advantages of this tool over the top-voted awk solution and other shell-based solutions:

  1. uq remembers the occurence of lines using their hash values, so it doesn't use as much memory use when the lines are long.
  2. uq can keep the memory usage constant by setting a limit on the number of entries to store (when the limit is reached, there is a flag to control either to override or to die), while the awk solution could run into OOM when there are too many lines.
Improve this answer

1 Comment

Quite inconvenient and less portable, given awk already does this.
4

Michael Hoffman's solution above is short and sweet. For larger files, a Schwartzian transform approach involving the addition of an index field using awk followed by multiple rounds of sort and uniq involves less memory overhead. The following snippet works in bash

awk '{print(NR"\t"$0)}' file_name | sort -t$'\t' -k2,2 | uniq --skip-fields 1 | sort -k1,1 -t$'\t' | cut -f2 -d$'\t'
Improve this answer

1 Comment

this seems to be rather slow, though
1

Thanks 1_CR! I needed a "uniq -u" (remove duplicates entirely) rather than uniq (leave 1 copy of duplicates). The awk and perl solutions can't really be modified to do this, your's can! I may have also needed the lower memory use since I will be uniq'ing like 100,000,000 lines 8-). Just in case anyone else needs it, I just put a "-u" in the uniq portion of the command:

awk '{print(NR"\t"$0)}' file_name | sort -t$'\t' -k2,2 | uniq -u --skip-fields 1 | sort -k1,1 -t$'\t' | cut -f2 -d$'\t'
Improve this answer

Comments

0

I just wanted to remove all duplicates on following lines, not everywhere in the file. So I used:

awk '{
  if ($0 != PREVLINE) print $0;
  PREVLINE=$0;
}'
Improve this answer

1 Comment

doesn't uniq do just that...
-4

the uniq command works in an alias even http://man7.org/linux/man-pages/man1/uniq.1.html

Improve this answer

Comments

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.