topic Re: count occurences of specific words in a separate table in SAS Programming

count occurences of specific words in a separate table

alex_philby — Tue, 09 Jan 2018 10:10:16 GMT

Hi everyone,

I have two tables. Table 1 looks like this :

cat

dog

lion

hamster

bear

guinea pig

And table 2 looks like this :

my dog is stupid

that cat likes my dog

fat hamster

bear are dangerous

lions are dangerous

lions eat antelopes

I like cats and lions

giraffes are cute

What I'd like as a result is to have a count of table 1 words occurences in table 2. To make myself clear, something like this :

words of table 1	occurrences in table 2
cat	2
dog	2
lion	3
hamster	1
bear	1
guinea pig	0

You will notice that plural forms of words, like "lions" for instance, are counted. In fact, we're looking for the chain of character "lion" in table 2, not the exact word.

I've been trying to split the strings in table 2 into words with the scan function, and running some proc freq and merges, but I'm not going anywhere.

Don't hesitate to ask any question. English is not my mother language so maybe I'm not being very clear.

Re: count occurences of specific words in a separate table

PeterClemmensen — Tue, 09 Jan 2018 10:12:12 GMT

What if the word "dog" appears twice in the same line?

Re: count occurences of specific words in a separate table

alex_philby — Tue, 09 Jan 2018 10:17:06 GMT

Sorry, it would count as a single occurence 🙂

Re: count occurences of specific words in a separate table

PeterClemmensen — Tue, 09 Jan 2018 11:07:57 GMT

An SQL approach

data have1;
input animals$50.;
datalines;
cat
dog
lion
hamster
bear
guinea pig
;

data have2;
input sentence$200.;
infile datalines truncover;
datalines;
my dog is stupid
that cat likes my dog
fat hamster
bear are dangerous
lions are dangerous
lions eat antelopes
I like cats and lions
giraffes are cute
;

proc sql;
	create table want as
	select animals
		  ,sum(ifn(find(sentence, strip(animals))>0, 1, 0)) as occurences
	from have1, have2
	group by animals
	order by calculated occurences desc;
quit;

Re: count occurences of specific words in a separate table

alex_philby — Tue, 09 Jan 2018 11:43:35 GMT

Thanks a lot. It works with the small data I've given as an example. Sadly, when I use it on the tables I'm working with (table 1 contains 1.7 millions observations and table 2, 7k observations), I get a "sort execution failure" error. I'm now trying to deal with this...I've found it could be a memory issue.

Re: count occurences of specific words in a separate table

andreas_lds — Tue, 09 Jan 2018 12:52:15 GMT

Using a hash-object and an iterator:

data have1;
input animal $50.;

occurences = 0;

datalines;
cat
dog
lion
hamster
bear
guinea pig
;
run;

data have2;
input sentence $200.;
infile datalines truncover;

datalines;
my dog is stupid
that cat likes my dog
fat hamster
bear are dangerous
lions are dangerous
lions eat antelopes
I like cats and lions
giraffes are cute
;
run;

data _null_;
   if 0 then set work.have1;

   set work.have2 end= jobDone;

   if _n_ = 1 then do;
      declare hash h(dataset: 'work.have1', multidata: 'yes');
      declare hiter iter('h');
      h.defineKey('animal');
      h.defineData('animal', 'occurences');
      h.defineDone();
   end;

   rc = iter.first();

   do while (rc = 0);
      if find(sentence, animal, 'it') then do;
         occurences = occurences + 1;
         h.replace();         
      end;

      rc = iter.next();
   end;

   if jobdone then do;
      h.output(dataset: 'work.want');
   end;
run;

proc sort data=work.want;
   by descending occurences;
run;

Without the not-exact-match requirement something like the 5th example in http://support.sas.com/documentation/cdl/en/lecompobjref/69740/HTML/default/viewer.htm#p00ilfw5pzcjvtn1nfya9863fozd.htm could be created - faster and less code.

Re: count occurences of specific words in a separate table

mkeintz — Tue, 09 Jan 2018 13:36:06 GMT

This program does not attempt a Cartesian join in memory, so it may be more efficient:

Editted modification. I overlooked the fact that you wanted frequencies in descending order, with just 2 columns. Here's a revised program using the same approach:

data have1;
input animals $50.;
datalines;
cat
dog
lion
hamster
bear
guinea pig
;

data have2;
input sentence $200.;
infile datalines truncover;
datalines;
my dog is stupid
that cat likes my dog
fat hamster
bear are dangerous
lions are dangerous
lions eat antelopes
I like cats and lions
giraffes are cute
;
proc sql noprint;
  select max(length(trim(animals))) into :maxlen from have1;
  select quote(trim(animals))       into :val_list separated by "," from have1;
quit;
%put _user_;

data want (keep=animals freq);
  array n_   {&sqlobs} ;
  array vals {&sqlobs} $&maxlen (&val_list);

  set have2 end=eoh;
  do w=1 to &sqlobs;
    n_{w} + (find(sentence, strip(vals{w}))>0);
  end;

  if eoh;
  do i=1 to &sqlobs;
    ix=whichn(max(of n_{*}),of n_{*});
    animals=vals{ix};
    freq=n_{ix};
    output;
    n_{ix}=.;
  end; 
run;

Re: count occurences of specific words in a separate table

Ksharp — Tue, 09 Jan 2018 13:54:32 GMT

I don't understand

"In fact, we're looking for the chain of character "lion" in table 2, not the exact word."

data have1;
input animals$50.;
datalines;
cat
dog
lions
hamster
bear
guinea pig
;

data have2;
input sentence$200.;
infile datalines truncover;
datalines;
my dog is stupid
that cat likes my dog
fat hamster
bear are dangerous
lions are dangerous
lions eat antelopes
I like cats and lions
giraffes are cute
;

data _null_;
 if _n_=1 then do;
   if 0 then set have1;
   declare hash h(dataset:'have1');
   declare hiter hi('h');
   h.definekey('animals');
   h.definedone();

   declare hash hh();
   hh.definekey('animals');
   hh.definedata('animals','n');
   hh.definedone();
 end;
set have2 end=last;
do while(hi.next()=0);
 if find(sentence,strip(animals)) then do;
  if hh.find()=0 then do;n=n+1;hh.replace(); end;
   else do;n=1;hh.add();end;
 end;
end;

if last then hh.output(dataset:'want');
run;

Re: count occurences of specific words in a separate table

alex_philby — Wed, 10 Jan 2018 13:27:35 GMT

It works with small data tables, but my SAS session still crashes when I try to run it on the 1mil+ observations tables.

Re: count occurences of specific words in a separate table

alex_philby — Wed, 10 Jan 2018 13:30:33 GMT

This one too is working with small data tables, but makes my SAS session crash when I try to run it on the 1mil+ observations tables. Sorry everyone, I guess it's just a memory problem. I can run simple data step on the mentionned tables, but your programs makes my session crash, for some reason.