Remove duplicates from text file based on second text file

Question

How can I remove all lines from a text file (main.txt) by checking a second textfile (removethese.txt). What is an efficient approach if files are greater than 10-100mb. [Using mac]

Example:

main.txt
3
1
2
5

Remove these lines

removethese.txt
3
2
9

Output:

output.txt
1
5

Example Lines (these are the actual lines I'm working with - order does not matter):

ChIJW3p7Xz8YyIkRBD_TjKGJRS0
ChIJ08x-0kMayIkR5CcrF-xT6ZA
ChIJIxbjOykFyIkRzugZZ6tio1U
ChIJiaF4aOoEyIkR2c9WYapWDxM
ChIJ39HoPKDix4kRcfdIrxIVrqs
ChIJk5nEV8cHyIkRIhmxieR5ak8
ChIJs9INbrcfyIkRf0zLkA1NJEg
ChIJRycysg0cyIkRArqaCTwZ-E8
ChIJC8haxlUDyIkRfSfJOqwe698
ChIJxRVp80zpcEARAVmzvlCwA24
ChIJw8_LAaEEyIkR68nb8cpalSU
ChIJs35yqObit4kR05F4CXSHd_8
ChIJoRmgSdwGyIkRvLbhOE7xAHQ
ChIJaTtWBAWyVogRcpPDYK42-Nc
ChIJTUjGAqunVogR90Kc8hriW8c
ChIJN7P2NF8eVIgRwXdZeCjL5EQ
ChIJizGc0lsbVIgRDlIs85M5dBs
ChIJc8h6ZqccVIgR7u5aefJxjjc
ChIJ6YMOvOeYVogRjjCMCL6oQco
ChIJ54HcCsaeVogRIy9___RGZ6o
ChIJif92qn2YVogR87n0-9R5tLA
ChIJ0T5e1YaYVogRifrl7S_oeM8
ChIJwWGce4eYVogRcrfC5pvzNd4

Answer 1

There are two standard ways to do this:

With grep:

grep -vxFf removethese main

This uses:

• -v to invert the match.
• -x match whole line, to prevent, for example, he to match lines like hello or highway to hell.
• -F to use fixed strings, so that the parameter is taken as it is, not interpreted as a regular expression.
• -f to get the patterns from another file. In this case, from removethese.

With awk:

$ awk 'FNR==NR {a[$0];next} !($0 in a)' removethese main
1
5

Like this we store every line in removethese in an array a[]. Then, we read the main file and just print those lines that are not present in the array.

Answer 2

With grep:

grep -vxFf removethese.txt main.txt >output.txt

With fgrep:

fgrep -vxf removethese.txt main.txt >output.txt

fgrep is deprecated. fgrep --help says:

Invocation as 'fgrep' is deprecated; use 'grep -F' instead.

With awk (from @fedorqui):

awk 'FNR==NR {a[$0];next} !($0 in a)' removethese.txt main.txt >output.txt

With sed:

sed "s=^=/^=;s=$=$/d=" removethese.txt | sed -f- main.txt >output.txt

This will fail if removethese.txt contains special chars. For that you can do:

sed 's/[^^]/[&]/g; s/\^/\\^/g' removethese.txt >newremovethese.txt

and use this newremovethese.txt in the sed command. But this is not worth the effort, it's too slow compared to the other methods.

---

Test performed on the above methods:

The sed method takes too much time and not worth testing.

Files Used:

removethese.txt : Size: 15191908 (15MB)     Blocks: 29672   Lines: 100233
main.txt : Size: 27640864 (27.6MB)      Blocks: 53992   Lines: 180034

Commands:
grep -vxFf | fgrep -vxf | awk

Taken Time:
0m7.966s | 0m7.823s | 0m0.237s
0m7.877s | 0m7.889s | 0m0.241s
0m7.971s | 0m7.844s | 0m0.234s
0m7.864s | 0m7.840s | 0m0.251s
0m7.798s | 0m7.672s | 0m0.238s
0m7.793s | 0m8.013s | 0m0.241s

AVG
0m7.8782s | 0m7.8468s | 0m0.2403s

This test result implies that fgrep is a little bit faster than grep.

The awk method (from @fedorqui) passes the test with flying colors (0.2403 seconds only !!!).

Test Environment:

HP ProBook 440 G1 Laptop
8GB RAM
2.5GHz processor with turbo boost upto 3.1GHz
RAM being used: 2.1GB
Swap being used: 588MB
RAM being used when the grep/fgrep command is run: 3.5GB
RAM being used when the awk command is run: 2.2GB or less
Swap being used when the commands are run: 588MB (No change)

Test Result:

Use the awk method.

Answer 3

Here are a lot of the simple and effective solutions I've found: http://www.catonmat.net/blog/set-operations-in-unix-shell-simplified/

You need to use one of Set Complement bash commands. 100MB files can be solved within seconds or minutes.

Set Membership

$ grep -xc 'element' set    # outputs 1 if element is in set
                            # outputs >1 if set is a multi-set
                            # outputs 0 if element is not in set

$ grep -xq 'element' set    # returns 0 (true)  if element is in set
                            # returns 1 (false) if element is not in set

$ awk '$0 == "element" { s=1; exit } END { exit !s }' set
# returns 0 if element is in set, 1 otherwise.

$ awk -v e='element' '$0 == e { s=1; exit } END { exit !s }'

Set Equality

$ diff -q <(sort set1) <(sort set2) # returns 0 if set1 is equal to set2
                                    # returns 1 if set1 != set2

$ diff -q <(sort set1 | uniq) <(sort set2 | uniq)
# collapses multi-sets into sets and does the same as previous

$ awk '{ if (!($0 in a)) c++; a[$0] } END{ exit !(c==NR/2) }' set1 set2
# returns 0 if set1 == set2
# returns 1 if set1 != set2

$ awk '{ a[$0] } END{ exit !(length(a)==NR/2) }' set1 set2
# same as previous, requires >= gnu awk 3.1.5

Set Cardinality

$ wc -l set | cut -d' ' -f1    # outputs number of elements in set

$ wc -l < set

$ awk 'END { print NR }' set

Subset Test

$ comm -23 <(sort subset | uniq) <(sort set | uniq) | head -1
# outputs something if subset is not a subset of set
# does not putput anything if subset is a subset of set

$ awk 'NR==FNR { a[$0]; next } { if !($0 in a) exit 1 }' set subset
# returns 0 if subset is a subset of set
# returns 1 if subset is not a subset of set

Set Union

$ cat set1 set2     # outputs union of set1 and set2
                    # assumes they are disjoint

$ awk 1 set1 set2   # ditto

$ cat set1 set2 ... setn   # union over n sets

$ cat set1 set2 | sort -u  # same, but assumes they are not disjoint

$ sort set1 set2 | uniq

# sort -u set1 set2

$ awk '!a[$0]++'           # ditto

Set Intersection

$ comm -12 <(sort set1) <(sort set2)  # outputs insersect of set1 and set2

$ grep -xF -f set1 set2

$ sort set1 set2 | uniq -d

$ join <(sort -n A) <(sort -n B)

$ awk 'NR==FNR { a[$0]; next } $0 in a' set1 set2

Set Complement

$ comm -23 <(sort set1) <(sort set2)
# outputs elements in set1 that are not in set2

$ grep -vxF -f set2 set1           # ditto

$ sort set2 set2 set1 | uniq -u    # ditto

$ awk 'NR==FNR { a[$0]; next } !($0 in a)' set2 set1

Set Symmetric Difference

$ comm -3 <(sort set1) <(sort set2) | sed 's/\t//g'
# outputs elements that are in set1 or in set2 but not both

$ comm -3 <(sort set1) <(sort set2) | tr -d '\t'

$ sort set1 set2 | uniq -u

$ cat <(grep -vxF -f set1 set2) <(grep -vxF -f set2 set1)

$ grep -vxF -f set1 set2; grep -vxF -f set2 set1

$ awk 'NR==FNR { a[$0]; next } $0 in a { delete a[$0]; next } 1;
       END { for (b in a) print b }' set1 set2

Power Set

$ p() { [ $# -eq 0 ] && echo || (shift; p "$@") |
        while read r ; do echo -e "$1 $r\n$r"; done }
$ p `cat set`

# no nice awk solution, you are welcome to email me one:
# [email protected]

Set Cartesian Product

$ while read a; do while read b; do echo "$a, $b"; done < set1; done < set2

$ awk 'NR==FNR { a[$0]; next } { for (i in a) print i, $0 }' set1 set2

Disjoint Set Test

$ comm -12 <(sort set1) <(sort set2)  # does not output anything if disjoint

$ awk '++seen[$0] == 2 { exit 1 }' set1 set2 # returns 0 if disjoint
                                         # returns 1 if not

Empty Set Test

$ wc -l < set            # outputs 0  if the set is empty
                         # outputs >0 if the set is not empty

$ awk '{ exit 1 }' set   # returns 0 if set is empty, 1 otherwise

Minimum

$ head -1 <(sort set)    # outputs the minimum element in the set

$ awk 'NR == 1 { min = $0 } $0 < min { min = $0 } END { print min }'

Maximum

$ tail -1 <(sort set)    # outputs the maximum element in the set

$ awk '$0 > max { max = $0 } END { print max }'

Answer 4

I like @fedorqui's use of awk for setups where one has enough memory to fit all the "remove these" lines: a concise expression of an in-memory approach.

But for a scenario where the size of the lines to remove is large relative to current memory, and reading that data into an in-memory data structure is an invitation to fail or thrash, consider an ancient approach: sort/join

sort main.txt > main_sorted.txt
sort removethese.txt > removethese_sorted.txt

join -t '' -v 1 main_sorted.txt removethese_sorted.txt > output.txt

Notes:

• this does not preserve the order from main.txt: lines in output.txt will be sorted
• it requires enough disk to be present to let sort do its thing (temp files), and store same-size sorted versions of the input files
• having join's -v option do just what we want here - print "unpairable" from file 1, drop matches - is a bit of serendipity
• it does not directly address locales, collating, keys, etc. - it relies on defaults of sort and join (-t with an empty argument) to match sort order, which happen to work on my current machine