About mkeintz

mkeintz

@ChrisNZ wrote: I'd have reported the other one as spam rather than merge. Why 2 different names? If there are two different names for this wholly identical post, then it may be that these are classmates asking about a homework task. If so, I think there is a benefit to merging.

mkeintz

@ChrisNZ wrote: Why do you post exactly the same question under a different name? https://communities.sas.com/t5/SAS-Programming/extract-additional-and-missing-records/m-p/935259 That post has now been merged with this one.

mkeintz

Questions: If a patient has one inpatient claim and one outpatient claim, do you keep both records for that patient? Or just the qualifying record? How do you identify inpatient vs outpatient claims? Do any of your patients have data spanning more than twelve months? If a patient has two years of data, and two qualifying records in the first twelve months, do you keep the entire history? Or maybe only records within twelve months of the nearest qualifying records? Are your data sorted, as in your example?

mkeintz

I assume your data are already sorted by ACCNO/TRANDATE. True, sorting by descending date (within each ACCNO) would allow you to take the first 20 of each ACCNO, a relatively straightforward process. But sorting is expensive, and you might need to re-sort the results by ascending date. So here is a two pass process (passing through each ACCNO twice, but in the original order: data transfer (drop=_:); set transactions (in=firstpass) transactions (in=secondpass) ; by accnow; _n1 + firstpass; _n2 + secondpass; if secondpass=1 and _n2>= _n1-20; run; Even though you pass through the entire dataset twice, it won't require twice the data-reading resources, because the SET statement interleaves the firstpass for a given ACCNO with the secondpass for the same ACCNO. Since the most recently read observations (from firstpass) will be cached in memory, they will therefore be available to the secondpass without rereading data from the storage device.

mkeintz

@PaigeMiller wrote: Seems pointless to me to re-arrange columns within a data set. What is the reason you want this? How would this help? I often do pay attention to establishing column order, for two reasons: To examine the data with a data set viewer In anticipation of using the double-dash technique for generating a variable list. Say I have a dataset with (from left to right), an ID, some demographic variables, some financial variables, some education variables. I often find myself using expressions like leftmost_demographic_variable -- rightmost_demographic_var to save a lot of typing where a list of variables would otherwise need to be explicitly typed.

mkeintz

@112211 wrote: Hello Quentin thanks for response , first thing is iam Senior clinical programmer and second thing is regarding my post that is real time problem iam facing that issue (i have framed the sample data by considering my realtime data ) So dont think any other way please. Any efforts solving the query its really helps me a lot .Thanks in advance You have one response offered by @Ksharp, but it apparently does not entirely meet your objective. And he was not entirely sure he complete undertood the tast. Could you please post a working DATA step with the results you expect from your sample input data? Help us help you.

mkeintz

I agree with @Quentin 's suggestion to at least mention you are attempting to understand a homework problem. Help us to help you own the question, not just answer it.

mkeintz

I see you already have a solution. Here's is an alternative that eschews DO groups: data want; set have; seq+mod(_n_,2); output; if mod(_n_,2)=0 then output; run;

mkeintz

@Heart_telling wrote: ... stuff deleted ... , but there has something wrong with my code . Could anyone kindly let me know how to fix this? Thank you very much. You provided your program code, but you gave us no clue as to how you know something is wrong. For instance, please provide: A copy of the sas log, with any notes or errors messages. A description of what your code produced, versus what you expected. BTW, you don't show a %MEND statement ending the macro definition, and you don't show code actually calling the macro. Help us help you.

mkeintz

@pavank wrote: data cities; input city $; datalines; chennai mumbai pune nodia ; run; data chennai mumbai pune noida; set cities nobs=nobs; do i = 1 to nobs; if city = 'chennai' then output chennai; else if city = 'mumbai' then output mumbai; else if city = 'pune' then output pune; else if city='nodia' then output noida; end;drop i; run; You could eliminate replace the collection of IF and ELSE IF statements by using "where" parameters in the output dataset names: data chennai (where=(city='chennai')) mumbai (where=(city='mumbai')) pune (where=(city='pune')) nodia (where=(city='nodia')) ; set cities nobs=nobs; do i=1 to nobs; output; end; drop i; run;

mkeintz

Instead of using the INDSNAME parameter, you could parse the log notes, which include the datasets with zero observations. All those notes have the text 'observations read from the data set', which can be used as the parameter of the INPUT @'character string' technique: data x_a ; x=1 ; a='abc' ; do i=1 to 4 ; output ; end ; data x_b ; x=1 ; a='defg' ; b='xyzzy' ; do i=1 to 4 ; output ; end ; data x_d ; y=1 ; b='xyz' ; a='pqrstuv' ; i=1; stop ; data x_f ; x='aaa' ; f='pqr' ; do i=1 to 4 ; output ; end ; data y1 y2 y3 y4 y9 y10 y11; qqq='rrr' ; do i=1 to 4 ; output ; end ; data x_1 x_empty x_3; x=1; output x_1; x=3; output x_3; output x_3; run; options obs=1; filename tmplog temp ; proc printto log=tmplog new; run; data _null_; set x_:(drop=_all_) y1-y4(drop=_all_) y9(drop=_all_) ; run ; proc printto; run; options obs=max; data datasets; infile tmplog dlm=' .'; input @'observations read from the data set' dslib :$8. dsname :$32. ; run;

mkeintz

If your data are sorted by ID, and you only have one non-missing DATE_DOSE1 value per ID, then a simple merge of the non-missing DATE_DOSE2 subset with the non-missing DATE_DOSE1 subset will allow you to use a subsetting IF statement: data dsin; input ID $ SESSION DATE_DOSE2 : mmddyy10. DATE_DOSE1 : mmddyy10.; format DATE_DOSE2 DATE_DOSE1 mmddyy10.; datalines; A 2 02/03/2021 . A 2 02/04/2021 . A 2 02/05/2021 . A 2 01/06/2020 . A 1 . 02/20/2020 B 2 04/16/2021 . B 2 04/17/2021 . B 2 04/18/2021 . B 1 . 02/21/2020 run; data want; merge dsin (where=(date_dose2^=.) drop=date_dose1) dsin (where=(date_dose1^=.) keep=id date_dose1); by id; if date_dose2>date_dose1; run;

mkeintz

If Your long dataset is already grouped (but not sorted) by CUSTID/VAR. You want to keep the dataset in original order Then: there is a single DATA step solution: data want2 (drop=v); array tmpval {6} _temporary_ (6*%sysfunc(constant(big))); /*First pass */ do v=1 by 1 until (last.var); set have2 ; by custid var notsorted; tmpval{v}=value; end; call sortn(of tmpval{*}); /*Second pass */ do until (last.var); set have2 ; by custid var notsorted; rank=whichn(value,of tmpval{*}); output; tmpval{rank}=constant('big'); end; run; This code accommodates up to 6 obs per custid/var. The initialization of the TMPVAL array to a set of %sysfunc(CONSTANT(BIG)) values helps with underpopulated (i.e. N<6) groups, because the sorting of values (between the first pass and second pass) keeps all observed values as the leftmost array elements (unlike initializing the array to missing values). That permits the WHICHN function to correctly determine the rank. That is also the reason why, once a rank is determined, the corresponding array element is reset to CONSTANT('BIG'), in preparation for the next group.

mkeintz

A minor efficiency note. You don't have to copy a data set to change its type to EST. Instead you can modify the metadata for COV1 using PROC DATASETS. proc datasets lib=work nodetails nolist; modify cov1 (type=EST); quit; :

mkeintz

@Fer_ wrote: Thanks for your reply. I want to get yearly std for each firm (permno) which has daily stock return. Which means for each firm in each year, I need to have one std at the end. That is what my second code block provides. Test it on a small sample of data to see for yourself.

Online Status	Online
Date Last Visited	yesterday