Retain both records if group is different otherwise keep the last

[Stalk]

This is extension to my previous question.

I would like to output last. if the group value is same.

If the group value is different I want to retain both records.(keep both Bold records below)

data to_clean;
infile cards dlm='|' truncover ;
input subDate :mmddyy10. unitName :$100. ADDR1 :$100. group $4.;
format subDate yymmdd10.;
cards;
11/21/2020|BAIRD HUDSON ENTERPRISES|106 E MAIN|TNF
10/30/2020|BAIRD HUDSON ENTERPRISES|106 E MAIN STREET|TNF
10/30/2020|BIG HORN ENTERPRISE|146 S. BENT STREET|SNF
10/30/2020|BIG HORN ENTERPRISE|641 WARREN STREET|SNF
11/5/2020|BROOKDALE |tt|ALF
10/29/2020|BROOKDALE|2401 COUGAR AVENUE|ALF
10/30/2020|ELMCROFT|1551 SUGARLAND DRIVE|ALF
11/2/2020|ELMCROFT|1551 SUGARLAND DRIVE DRIVE|SNF
11/21/2020|GREEN HOUSE LIVING|2311 SHIRLEY|SNF
10/29/2020|GREEN HOUSE LIVING|2311 SHIRLEY COVE|ALF
11/21/2020|MISSION AT THE VILLA|1445 UINTA|ALF
11/2/2020|MISSION AT THE VILLA|1445 UINTA DRIVE|ALF
;

proc sort data=to_clean; by unitName subDate; run;

proc sql;
create table to_clean_new as
select * from to_clean
order by unitName, subDate, group;
quit;

data to_clean_2;
set to_clean_new;
by unitName;
length goodAddr $100;
retain goodAddr;
if first.unitName then goodAddr = addr1;
else if complev(trim(goodAddr), trim(addr1), "IL:") > 2 then goodAddr = addr1;
run;

proc sort data=to_clean_2; by unitName goodAddr subDate group; run;

data want;
set to_clean_2;
by unitName goodAddr;
if last.goodAddr;
run;

proc print noobs data=want; run;

****

Thank you

[novinosrin]

Hi @Stalk  I'm not sure of your expected output. However, rather than looking at your code, I'm taking a guess with your bold lines and your description-

"I would like to output last. if the group value is same.

If the group value is different I want to retain both records.(keep both Bold records below)"

data to_clean;
infile cards dlm='|' truncover ;
input subDate :mmddyy10. unitName :$100. ADDR1 :$100. group $4.;
format subDate yymmdd10.;
cards;
11/21/2020|BAIRD HUDSON ENTERPRISES|106 E MAIN|TNF
10/30/2020|BAIRD HUDSON ENTERPRISES|106 E MAIN STREET|TNF
10/30/2020|BIG HORN ENTERPRISE|146 S. BENT STREET|SNF
10/30/2020|BIG HORN ENTERPRISE|641 WARREN STREET|SNF
11/5/2020|BROOKDALE |tt|ALF
10/29/2020|BROOKDALE|2401 COUGAR AVENUE|ALF
10/30/2020|ELMCROFT|1551 SUGARLAND DRIVE|ALF
11/2/2020|ELMCROFT|1551 SUGARLAND DRIVE DRIVE|SNF
11/21/2020|GREEN HOUSE LIVING|2311 SHIRLEY|SNF
10/29/2020|GREEN HOUSE LIVING|2311 SHIRLEY COVE|ALF
11/21/2020|MISSION AT THE VILLA|1445 UINTA|ALF
11/2/2020|MISSION AT THE VILLA|1445 UINTA DRIVE|ALF
;

data want;
 do _n_=1 by 1 until(last.unitname);
  set to_clean;
  by unitname group notsorted;
  if first.group then n=sum(n,1);
 end;
 if n=1 then output;
 else do _n_=1 to _n_;
  set to_clean;
  output;
 end;
 drop n;
run;

[Stalk]

Records with different group are not captured in the output dataset. From the data above I would like to get 1,3,4,5, 7,8,9,10 observations( eliminate 2 , 6 and 11).

[mkeintz · Accepted solution]

I've restated your objective to "keep the last observation (highest subdate) for each unitname/group combination).  If that's correct, then this application of proc summary and a subsequent use of a hash object will do, which prunes the dataset while preserving the original order of the data (in case that's important).  It requires only 2 passes through the data:

data to_clean;
infile cards dlm='|' truncover ;
input subDate :mmddyy10. unitName :$100. ADDR1 :$100. group $4.;
format subDate yymmdd10.;
cards;
11/21/2020|BAIRD HUDSON ENTERPRISES|106 E MAIN|TNF
10/30/2020|BAIRD HUDSON ENTERPRISES|106 E MAIN STREET|TNF
10/30/2020|BIG HORN ENTERPRISE|146 S. BENT STREET|SNF
10/30/2020|BIG HORN ENTERPRISE|641 WARREN STREET|SNF
11/5/2020|BROOKDALE |tt|ALF
10/29/2020|BROOKDALE|2401 COUGAR AVENUE|ALF
10/30/2020|ELMCROFT|1551 SUGARLAND DRIVE|ALF
11/2/2020|ELMCROFT|1551 SUGARLAND DRIVE DRIVE|SNF
11/21/2020|GREEN HOUSE LIVING|2311 SHIRLEY|SNF
10/29/2020|GREEN HOUSE LIVING|2311 SHIRLEY COVE|ALF
11/21/2020|MISSION AT THE VILLA|1445 UINTA|ALF
11/2/2020|MISSION AT THE VILLA|1445 UINTA DRIVE|ALF
;

proc summary data=to_clean nway;
  class unitname group;
  var subdate;
  output out=need (drop=_:) max=subdate;
run;

data want;
  set to_clean;
  if _n_=1 then do;
    declare hash h (dataset:'need');
	  h.definekey(all:'Y');
	  h.definedone();
  end;
  if h.check()=0;
run;

The proc summary (because of the NWAY option) outputs a dataset with one observation for each unitname/group combination.  It records the maximum value of the analysis variable subdate for each combo, thereby providing the desired unitname/group/subdate values.   

Stick that dataset in the hash object h, and check each incoming obs from to_clean against h, and keep only those whose key values (unitname,group,subdate) are found in h (i.e. "if h.check()=0").

[Stalk]

Thank you mkeintz for suggesting Hash object. I really don't understand how hash object works so avoid using that. But I got my program working for the desired results with simple fixxes to sort..

proc sort data=to_clean_2; by unitName goodAddr group descending subDate ; run;

data new;
set to_clean_2;
by unitname goodaddr group;
if first.group then output;
run;