Handling Missing Data in Inputs

by: Tony Lawrence

Handling Missing Data in Inputs
Missing data can be very annoying to a programmer. In fact, it is so annoying that very often we'll write separate programs to clean up data and eliminate unpleasant conditions so that the main program doesn't have to deal with it. Here, I'll show some examples of the kind of problems we see.

Let's take a comman data format, a TAB delimited file. A simplistic Perl program to read such a file might be:

#!/usr/bin/perl

while (<>) {

 #split on tab into @x array

 @x=split /\t/;

 #print first three elements

 print "$x[0]\t$x[1]\t$x[2]\n";

}

An equivalent shell script might be

IFS="(tab here)"

while read a b c d

do

echo "$a        $b      $c"

done

The Perl script works, but the shell script doesn't. Here's the output if the imput file looks like this:

$ cat t;hexdump -c t

1       2       3       4

1               3       4

        2       3       4

1       2               4

                3       4

0000000   1  \t   2  \t   3  \t   4  \n   1  \t  \t   3  \t   4  \n  \t

0000010   2  \t   3  \t   4  \n   1  \t   2  \t  \t   4  \n  \t  \t   3

0000020  \t   4  \n                                                    

0000023

The Perl script produces

1       2       3

1               3

        2       3

1       2

                3

but the shell script messes up:

1       2        3

1       3        4

2       3        4

1       2        4

3       4

If this were a problem with Perl, we'd handle it like this:

#!/usr/bin/perl

while (<>) {

 # make sure there is at least one space between adjacent tabs

 s/\t\t/\t \t/g;

 #split on tab into @x array

 @x=split /\t/;

 #print first three elements

 print "$x[0]\t$x[1]\t$x[2]\n";

}

But things can be worse. For example, if we are processing what was once a report format, we may have no delimiters, just empty space. We might see something like this:

Date          Customer             Phone           Terms     Balance



09/04/04      ABCD Corp.                           PPD          0.00

09/04/04      Abba Corp.          555-5555         Net 30     985.00

You can't process that with delimiters, but you can use unpack:

#!/usr/bin/perl

while(<>) {

@x=unpack("A8A6A20A17A9A12",$_);

print "$x[0]:$x[2]:$x[3]:$x[4]:$x[5]\n";

}

Which will produce:

Date:Customer: Phone:Terms: Balance

::::

09/04/04:ABCD Corp.::PPD:    0.00

09/04/04:Abba Corp.:555-5555:Net 30:  985.00

Comma separated value files can be annoying if they also contain commas within quoted fields. You can't use split because of that. There are at least two ways to handle that: either use the Text::Parsewords module:

#!/usr/bin/perl

use Text::ParseWords;

while(<>) {

 @x=quotewords(",",0,$_);

 foreach (@x) {

  print " $_";

 }

print "\n";

}

Or (assuming the data is regular enough), replace commas not inside quotes with a different delimiter and then split it. I think ParseWords is easier.

But sometimes none of that is going to work either. I'm working on a project right now where the input data can have up to three fields, but any of the three can be missing and there are no delimiters and no spacing. The only way to determine what we have is to know that the field one, if present, is alpha, field two is a whole integer, and field three will always have decimal points. So

ABC  982.00

8

15.45

means that I have 1 and 3 on line 1, only 2 on line 2, and only 3 on line 3. It's actually much worse than this; there are other fields, some of which are always present and some which are not, and it is quite a challenge to normalize this stuff to be able to massage the data. The way to handle it is to do splits on / /, and then determine what we got. So it's something like this:

#!/usr/bin/perl

while(<>) {

s/\s+/ /g;

@x=split / /;

foreach (@x) {

  .. determine what we have based on previous field(s) seen and content

}