SAS Syntax Automation

UrvishShah · Posted 12-12-2011 05:29 PM

Yes Sir you are right that in your code as the condition is satisfied, other will be disregarded

But in this case, we need all the IF's statement to execute

If i use IF-THEN/ELSE, only first time SAS will execute the statement as the condition is true and second ELSE statement won't execute

So collasping will not be done in all the regions

I may be wrong in SAS concepts but i got the output just because of you sir

Thanks again

Nice to Meet you

art297 · Posted 12-12-2011 05:35 PM

I disagree as only one statement can be true out of all of the conditions. However, it will work both ways, thus mark the question as answered and let's move on to the next question someone might have.

UrvishShah · Posted 12-12-2011 05:39 PM

Thanks for help

I learned so many new concepts from you

FriedEgg · Posted 12-12-2011 06:39 PM

Select statements should be faster than so many non-overlapping if or if/else methods. Give this a try. I did not spend time to make it pretty...

data _null_;

file "C:\changes_new.sas";

set data.collapse end=last;

by region26 var;

if _n_=1 then put 'select(region26);';

if first.region26 then put 'when(' region26 ') do;';

if first.var then put 'select(' var ');';

select(var);

when ('marriage') do;

select(group);

when (2,5) put 'when(2,5) ' var '=2;';

when (3,4) put 'when(3,4) ' var '=3;';

otherwise;

end;

when ('income') do;

select(group);

when (1,2) put 'when(1,2) ' var '=2;';

when (3,4) put 'when(3,4) ' var '=4;';

when (5,6) put 'when(5,6) ' var '=6;';

otherwise;

end;

otherwise;

end;

if last.var then put 'otherwise; end;';

if last.region26 then put 'end;';

if last then put 'otherwise; end;';

run;

I did some tests with expanding your provided data:

data coded;

input E3010 income hh_size region10 region26 minority_i hh_size_i income_i marriage_i own_rent_i E112_n marriage marriage_r own_rent minority total_i fadj f2 w0;

do i=1 to 100000;

output; *simulate more data;

end;

cards;

19000100000000 5 4 1 1 0 0 0 0 0 96 1 1 1 0 0 1.10534711 0.26676473 0.2948688

19000100000000 3 4 1 1 1 0 0 0 0 96 2 5 1 0 1 1.10534711 0.26676473 0.2948688

19000100000000 4 1 1 1 0 0 0 0 0 96 2 5 1 0 0 1.10534711 0.26676473 0.2948688

19000100000000 1 4 1 1 0 0 0 0 0 96 1 1 0 1 0 1.10534711 0.26676473 0.2948688

19000100000000 3 3 1 1 1 0 0 0 1 96 1 1 1 0 2 1.10534711 0.26676473 0.2948688

19000100000000 7 3 1 1 0 0 0 0 0 96 1 1 1 0 0 1.10534711 0.26676473 0.2948688

19000100000000 6 3 1 1 0 0 1 0 0 96 2 5 1 0 1 1.10534711 0.26676473 0.2948688

19000100000000 6 3 1 1 0 0 0 0 0 96 2 5 0 0 0 1.10534711 0.26676473 0.2948688

19000200000000 7 2 1 1 0 0 0 0 0 96 1 1 1 0 0 1.10534711 0.26676473 0.2948688

21005000000000 7 2 1 1 0 0 0 0 0 96 1 1 1 0 0 1.10534711 0.26676473 0.2948688

21005000000000 6 2 1 1 0 0 0 0 0 96 1 1 1 0 0 1.10534711 0.26676473 0.2948688

21005000000000 7 3 1 1 0 0 0 0 0 96 1 1 1 0 0 1.10534711 0.26676473 0.2948688

21005000000000 5 1 1 1 0 0 0 0 0 96 3 2 1 0 0 1.10534711 0.26676473 0.2948688

21005000000000 1 2 1 1 0 0 1 0 0 96 2 5 0 0 1 1.10534711 0.26676473 0.2948688

21005000000000 5 1 1 1 0 0 0 0 0 96 3 3 1 0 0 1.10534711 0.26676473 0.2948688

21005000000000 4 4 1 1 0 0 0 0 0 96 2 5 0 0 0 1.10534711 0.26676473 0.2948688

21005000000000 7 2 1 1 0 0 0 0 0 96 1 1 1 0 0 1.10534711 0.26676473 0.2948688

21005000000000 6 1 1 1 0 0 0 0 0 96 2 2 1 0 0 1.10534711 0.26676473 0.2948688

21005000000000 7 2 1 1 1 0 1 0 0 96 1 1 1 0 2 1.10534711 0.26676473 0.2948688

;

run;

Average performance over 5 runs:

Urvish Method 0.73 cpu time

Art Method 0.99 cpu time (I agree that Art's method should be more efficient, this is very strange to me. The efficiency would increase if the nested else's were in priority of likelihood... This is still strange to me.)

FE Method 0.60 cpu time (I win! )

art297 · Posted 12-12-2011 08:22 PM

Matt,

I sit corrected! While if then else CAN improve performance, given this particular set of sample data it won't unless one can optimize the constructs more efficiently than I did.

Art

art297 · Posted 12-12-2011 11:45 PM

Matt,

I disagree that select is faster than if then. Here is a more optimized version of my original code. In my tests, it runs neck and neck with your version:

data _null_;

file "c:\changes_newer.sas";

set collapse;

by region26 var;

if _n_ eq 1 then counter=0;

if count lt 10 then do;

if first.region26 then do;

counter+1;

if counter eq 1 then put "if region26 eq " @;

else put "else if region26 eq " @;

put region26 " then do;";

end;

if var eq 'income' then do;

if first.var then put "if income in (";

else put "else if income in (";

if group in (1, 2) then put "1, 2) then income=2;";

else if group in (3, 4) then put "3, 4) then income=4;";

else if group in (5, 6) then put "5, 6) then income=6;";

end;

if var eq 'marriage' then do;

if first.var then put "if marriage in (";

else put "else if marriage in (";

if group in (2, 5) then put "2, 5) then marriage=2;";

else if group in (3, 4) then put "3, 4) then marriage=3;";

end;

if last.region26 then put "end;";

run;

data want_newer;

set coded;

%include "c:\changes_newer.sas";

run;

FriedEgg · Posted 12-13-2011 12:09 AM

It is not meant as a blanket statement. In my experience it holds true dependent on number of nests and their order, etc. a well optimized if/then can definitely outperform a select statement. Everything with performance is always case based.

See this note from the select statement syntax document.

Comparisons

Use IF-THEN/ELSE statements for programs with few statements. Use subsetting IF statements without a THEN clause to continue processing only those observations or records that meet the condition that is specified in the IF clause.