About joehinson

joehinson · ‎12-06-2012

A hash-transpose. Not so elegant; way too many hard-coding of variables! Also, dates were handled as character to simplify matters. The hash variables could have been handled with macro variables, through a prior Proc SQL. data _null_; if(1=2)then set have; length var1_200101-var1_200105 var2_200101-var2_200105 $2 ; declare hash varmon (ordered:"a"); varmon.defineKey("id"); varmon.defineData ("id","Var1_200101","Var1_200102","Var1_200103","Var1_200104","Var1_200105", "Var2_200101","Var2_200102","Var2_200103","Var2_200104","Var2_200105"); varmon.defineDone(); call missing (of _all_); array w{5,2} $ _temporary_; do until(done); set have end=done; rcf=varmon.find(key:id); w(1,1)=Var1_200101;w(2,1)=Var1_200102;w(3,1)=Var1_200103;w(4,1)=Var1_200104;w(5,1)=Var1_200105; w(1,2)=Var2_200101;w(2,2)=Var2_200102;w(3,2)=Var2_200103;w(4,2)=Var2_200104;w(5,2)=Var2_200105; mx=whichc(month,"200101","200102","200103","200104","200105"); w(mx,1)=var1; w(mx,2)=var2; rcr=varmon.replace(key:id,data:id,data:w(1,1),data:w(2,1),data:w(3,1),data:w(4,1),data:w(5,1), data:w(1,2),data:w(2,2),data:w(3,2),data:w(4,2),data:w(5,2)); call missing(of _all_); end; varmon.output(dataset:"want"); stop; run;

joehinson · ‎11-28-2012

Thanks, DN. I chose to create labels within the data step because they were needed in the hash-of-hashes processing. I'm very much aware of those useful features of Proc Transpose. As a matter of fact, if I were not hard pressed for time, I wouldn't have needed Proc Transpose at all. Regards, Joe

joehinson · ‎11-28-2012

....and here is another approach that also works for any number of columns. (I did this mainly as an academic exercise, testing the use of data access functions (eg. getvarc()), and the hash-of-hashes for group processing): data have; infile datalines truncover; input (Id Question responce1-responce7) ($); datalines; fgg45 S1 YES fgg45 S2 NO NO NO NO NO NO fgg45 G5 YES YES YES YES YES YES YES fgg45 G6 NO gg453 S1 YES YES gg453 S2 NO NO NO NO NO NO NO gg453 G5 YES YES YES gg453 G6 NO NO NO NO ; run; data _null_; if _N_=1 then do; length Responce $3 colx $12; if(1=2)then set have; declare hash final (ordered:"a",multidata:"Y"); final.defineKey("id","colx"); final.defineData("id","colx","responce"); final.defineDone(); declare hash ids (ordered:"a"); ids.defineKey("id"); ids.defineData("id", "objQues","hobjQues"); ids.defineDone(); declare hiter hids("ids"); declare hash objQues; declare hiter hobjQues; declare hash objResp; declare hiter hobjResp; call missing (of _all_); dsid=open("have"); put dsid=; nrows=attrn(dsid,"NOBS");put nrows=; ncols=attrn(dsid,"NVARS");put ncols=; end; do y=1 to nrows; rc=fetchobs(dsid,y); id=getvarc(dsid,1);question=getvarc (dsid,2); if ids.find() ne 0 then do; objQues=_new_ hash(ordered:"a"); objQues.defineKey("id","Question"); objQues.defineData("Question","objResp","hobjResp"); objQues.defineDone(); hobjQues=_new_ hiter("objQues"); rci=ids.replace(); end; if objQues.find() ne 0 then do; objResp=_new_ hash(ordered:"a",multidata:"Y"); objResp.defineKey("id","Question"); objResp.defineData("id","Question","colx","Responce"); objResp.defineDone(); hobjResp=_new_ hiter("objResp"); rcq=objQues.replace(); do n=3 to ncols; cx=varname(dsid,n); dx=compress(cx,"0123456789","k"); colx=strip(Question)||"_"||"X"||dx; responce=getvarc(dsid,n); rcr=objResp.replace(); rcf=final.replace(); end; end; end; final.output(dataset:"finaldata"); stop; run; proc transpose data=finaldata out=want;; by id; id colx; var responce; run;

joehinson · ‎11-27-2012

Wow! Many thanks, Tom. Will keep that in mind. Kind regards, Joe

joehinson · ‎11-27-2012

Here is another approach, although not quite sure how efficient that would be: data _null_; if(1=2) then set have; declare hash cols(ordered:"a"); cols.defineKey("period"); cols.defineData("period","colA", "colB", "colC"); cols.defineDone(); do until(done); set have end=done; x1=colA; x2=colB; x3=colC; rc1=cols.find(); y1=max(x1, colA); y2=max(x2, colB); y3=max(x3, colC); rc2=cols.replace(key:period,data:period,data:y1,data:y2,data:y3); end; cols.output(dataset:"want"); stop; run; RESULTS: Period ColA ColB ColC 1 1 1 2 3 2 2 1 2 . 3 3 5 . 8

joehinson · ‎11-23-2012

Hello My Friend Haikuo, How could you possibly fail to notice? (You need a new optician or opthamologist. LOL!) It is indeed HOH (just a 2-level nesting). ctag and hictag are objects within the info hash table; declare hash info(ordered:"a"); info.defineKey("id"); info.defineData("id","swap","ctag", "hictag"); info.defineDone(); declare hiter hinfo("info"); declare hash ctag; declare hiter hictag; ----- ----- if info.find() ne 0 then do; ctag=_new_ hash(ordered:"a",multidata:"y"); hictag=_new_ hiter("ctag"); ctag.defineKey("id"); ctag.defineData("seq","id","varb","varc"); ctag.defineDone(); rca=info.replace(); end; rcb=ctag.replace();

joehinson · ‎11-23-2012

Hello Bhpinder, Certainly, the temporary array approach is the fastest. However, as you well know, every approach has its limitations, more or less. For regular arrays, you need to know the size in advance; and you dare not mix variable types (characters and numeric). Hashes ("associative arrays") have their limitations too, aside from being memory hogs and tending to take more programming lines at times. The problem is simply a case of group processing as well as the ability to see the data as a matrix. Thus, to some extent, hashes can fit the bill. Here is a one-pass-data hash solution: (I added a sequencing variable to the input data, just to preserve the original order)! data have; infile datalines truncover; input ID (varb varc) ($); seq=_N_; datalines; 1 a db 1 b db 1 c db 2 a 2 a 3 a db 3 a db 3 b db 3 c db 4 b db 5 b db 5 b db 5 a db 6 b 6 b db 6 c db 6 a 6 b 6 b db 6 c ; data want(keep=seq ID varb varc); length swap 8; if(1=2)then set have; if _N_=1 then do; array sw (2) $ _temporary_; declare hash info(ordered:"a"); info.defineKey("id"); info.defineData("id","swap","ctag", "hictag"); info.defineDone(); declare hiter hinfo("info"); declare hash ctag; declare hiter hictag; end; do until(done); set have end=done; by id notsorted; if info.find() ne 0 then do; ctag=_new_ hash(ordered:"a",multidata:"y"); hictag=_new_ hiter("ctag"); ctag.defineKey("id"); ctag.defineData("seq","id","varb","varc"); ctag.defineDone(); rca=info.replace(); end; rcb=ctag.replace(); swap=((last.id) * (varb="c")) ; info.replace(); end; hinfo.first(); do until (hinfo.next() ne 0); hictag.first(); do until (hictag.next() ne 0); sw(1)=varb; sw(2)="c"; varb=sw(swap+1); output; call missing( varb, varc); end; end; stop; run;

joehinson · ‎11-15-2012

if (id=101) or (id=104) or (id=106);

joehinson · ‎10-26-2012

Hello Sven, About the input dataset. You have initially: set SASUSER.HEJ; What do you have in SASUSER.HEJ to begin with? Do you want to generate a single random number in sasuser.hej or a set of random numbers in that dataset?

joehinson · ‎10-19-2012

You may also try the PARMBUFF option with subsequent SYSPBUFF parsing. Please see http://www.lexjansen.com/pharmasug/2006/technicaltechniques/tt28.pdf

joehinson · ‎10-19-2012

Sorry, this is four years late !!! (At that time, as a neuroscientist, I was rather busy with neuronal signals). Today, with SAS 9.3 and DOSUBL function, you can run your PROC IML code within the DATA step containing active hash objects. From my experience with other PROCs and DOSUBL, the hash objects can interact with the PROCs and vice versa, within the same DATAstep.

joehinson · ‎10-04-2012

x4=(x1|x2|x3 ) is the same as x4=(x1 OR x2 OR x3) which is TRUE and therefore 1 (because it is an assignment statement and will always be true). Furthermore, any number OR'ed with another number would give 1, except (0 OR 0) which would give 0. So, (34 OR 987765)=1

joehinson · ‎09-25-2012

It looks like the certification program at Philadelphia University would suit you quite well: http://www.philau.edu/scps/CPE/Certificates/SAS.html

joehinson · ‎09-14-2012

Andre, in fact I barely scratched the surface. You would be amazed what RWI can do! See the paper written by SAS Institute's Dan O'Connor (the father of RWI): http://support.sas.com/rnd/base/datastep/dsobject/Power_to_show_paper.pdf

joehinson · ‎09-11-2012

Hello Andre, make sure the ODS Report Writing Interface (RWI) code (the part you showed) is part of the SAME data step that created the hiter object HICOD. Once you leave that Data Step containing declare hiter hicod("cod"); the object is lost forever. The error mssage suggests that declaration statement was not seen. You must have started another Data Step to run the RWI. RWI uses the hiter object from the previous hash processing Data Step to access data. Please let me know if that was the case. Also, the 9.3 version should not matter. Actually the code should work even in 9.2 except that you might get a lousy PDF. But absolutely no errors!

Online Status	Offline
Date Last Visited	‎03-23-2017 12:34 PM

Re: transpose - help

Re: transpose - help

Re: transpose - help

Re: transpose - help

Re: Check for identical values.

Re: Identifying observations Found in Set B but not found in Set A

Re: create 2 new vars a new and old from 2 rows

Re: how to flag consecutive values

Re: How to replace the value..

Re: Give one number for matching Records

Re: Splitting a database

Re: select on unique character codes from a column and generate a repo...

Re: Identifying observations Found in Set B but not found in Set A

Re: subset a numeric variable

Re: Give one number for matching Records

Re: Proc Transpose Alternative

Re: Proc Transpose Issue

Re: Proc Transpose Issue

Re: How can I combine 2 columns into one?

Re: How can I combine 2 columns into one?

Re: Populating values

Re: Populating values

Re: subset a numeric variable

Re: Randomly assigning cases from a data set to different groups repea...

Re: Repeat a macro

Re: Using hash objects in IML?

Re: what does x4 do?

Re: How to prepare as a Statistical Programming

Re: select on unique character codes from a column and generate a repo...

Re: select on unique character codes from a column and generate a repo...