5

I'm in a directory with a couple thousand files, but the files I want to filter all have the following syntax: *.imputed.*_info

I want to use awk to filter out the records in each file where the 5th column of data has a value > 0.50 and I was able to do that with: awk '{if($5 >= .5) {print}}' filename.

That too worked. I then tried to loop through all 500 or so files and concatenate records from each that match this criteria.

I tried the following but I am not getting the syntax right.

touch snplist.txt
for name in *.imputed.*_info; do
    snps="awk '{if($5 >= .5) {print}}' $name"
    cat snplist.txt "$snps" > snplist.txt
done
14

Your code overwrites the output file in each iteration. You also do not actually call awk.

What you want to do is something like

awk '$5 >= 0.5' ./*.imputed.*_info >snplist.txt

This would call awk with all your files at once, and it would go through them one by one, in the order that the shell expands the globbing pattern. If the 5th column of any line in a file is greater or equal to 0.5, that line would be outputted (into snplist.txt). This works since the default action, if no action ({...} block) is associated with a condition, is to output the current line.

In cases where you have a large number of files (many thousands), this may generate an "Argument list too long" error. In that case, you may want to loop:

for filename in ./*.imputed.*_info; do
    awk '$5 >= 0.5' "$filename"
done >snplist.txt

Note that the result of awk does not need to be stored in a variable. Here, it's just outputted and the loop (and therefore all commands inside the loop) is redirected into snplist.txt.

For many thousands of files, this would be quite slow since awk would need to be invoked for each of them individually.

To speed things up, in the cases where you have too many files for a single invocation of awk, you may consider using xargs like so:

printf '%s\0' ./*.imputed.*_info | xargs -0 awk '$5 >= 0.5' >snplist.txt

This would create a list of filenames with printf and pass them off to xargs as a nul-terminated list. The xargs utility would take these and start awk with as many of them as possible at once, in batches. The output of the whole pipeline would be redirected to snplist.txt.

This xargs alternative is assuming that you are using a Unix, like Linux, which has an xargs command that implements the non-standard -0 option to read nul-terminated input. It also assumes that you are using a shell, like bash, that has a built-in printf utility (ksh, the default shell on OpenBSD, would not work here as it has no such built-in utility).


For the zsh shell (i.e. not bash):

autoload -U zargs
zargs -- ./*.imputed.*_info -- awk '$5 >= 0.5' >snplist.txt

This uses zargs, which is basically a reimplementation of xargs as a loadable zsh shell function. See zargs --help (after loading the function) and the zshcontrib(1) manual for further information about that.

  • If the arg list is too long for awk, surely it is too long for printf too. – Paul_Pedant Dec 10 '19 at 23:44
  • 1
    @Paul_Pedant:  No.  awk is a program.  printf is a shell builtin, and is therefore exempt from the "Argument list too long" error. – Scott Dec 11 '19 at 2:37
  • @Scott Correct, except when printf is not available as a shell built-in, as with OpenBSD's ksh shell, as I mentioned. – Kusalananda Dec 12 '19 at 5:09
5

Just do this :

awk '$5 >= .5' *.imputed.*_info > snplist.txt
  • 1
    You can move the if (condition) into awk's condition part, and then the {print} can be omitted as it's the default action if none is specified. awk '$5 >= .5' *.imputed.*_info. But then you're repeating Kusalananda's answer.. – roaima Dec 9 '19 at 22:53
3

I have a habit of using find for this kind of thing.

find . -type f -name "*.imputed.*_info" -exec awk '$5 >= 0.5' {} \; > ./snplist.txt
  • Worth mentioning that the {} + is equivalent to xargs, vs. {} \; being slow like a shell for loop that runs awk separately on each file. – Peter Cordes Dec 10 '19 at 23:06

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.