Pankaj,

Here is some code to help you get started. Note I changed the structure of the control file per Howard's directions.

/* test data */ data w ; do obs = 1 to 100 ; num_b = 300 * ranuni ( 14231 ) ; output ; end ; run ;

/* get mean */ proc sql noprint ; select mean(num_b) into :mean_b from w ; run ;

/* make format data */ data fmtdata ; retain mean_b &mean_b ; input Fmtname $ Start $ end $ label $ hlo $ ; if upcase(fmtname) = "NUM_BFMT" then label = put ( mean_b , best12. ) ; cards ; Num_Bfmt . . . . Num_Bfmt low 100 . l Num_Bfmt 300 high . h ;

/* make the format */ proc format cntlin = fmtdata fmtlib ; run ;

/* prove of the pudding */ data q ; set w ; newb = input(put(num_b,num_bfmt12.6), best12.) ; run ;

IanWhitlock@westat.com

-----Original Message----- From: Pankaj Singal [mailto:pankaj.singal@WELLSFARGO.COM] Sent: Friday, August 29, 2003 10:54 AM To: SAS-L@LISTSERV.UGA.EDU Subject: Re: proc format from dataset for multiple variables.

When I tried to implement the process for one numeric and character variable, the character variable runs fine, but I am facing problems with the numeric variables. How to go about the same.

This is what I intend to do: (I have numeric variable num_b that can have legal values b/w 100-300. and if in my input dataset the value for num_b is in the range ([100-300]), we do not change anything but if it is not, we change as per the following.

Control_file Fmtname Start Label Num_Bfmt . mean(num_b) Num_Bfmt low-100 mean(num_b) nu_Bfmt 300-high mean(num_b)

How do I make a caro that calculates the mean and put it in the control_file. Also, I am not sure how to incorporate the range that is valid. i.e. if the num_b = 105, then my variable in "test" dataset should be 105. How to get that in without using an if statement in data step.

Any suggestions, would be greatly appreciated.

Thanks

Howard_Schreier@ITA.DOC.GOV (Howard Schreier) wrote in message news:<200308282045.h7SKjid31677@listserv.cc.uga.edu>... > Seems like a good approach. > > I suggest that you you make your control dataset conform as much as > possible to the structure and naming conventions of a SAS PROC FORMAT > CNTLIN dataset. > > On Wed, 27 Aug 2003 09:23:35 -0700, Pankaj Singal > <pankaj.singal@WELLSFARGO.COM> wrote: > > >I have a huge dataset(sample) (>100 variables/fields eg, A B C D E F > >...A1..A75...). I am trying to create new dateset (test) with all > >valid entries for various fields in "sample" dataset. I created a > >separate dataset (control_dataset) that has 5 columns, which will > >help me create the proc formats to create the "test" dataset with > >valid entries. > > > >Control_dataset > > > >Var_name old_value Operator new_value Description > >A "1" EQ "WALK" "1 changes to Walk" > >A "2" EQ "RUN" "2 changes to RUN" > >A other else "INVALID" "Invalid entry" > >B [100,300] EQ [100,300] "valid range 100-300" > >B other else mean "invalid value replaced by mean" > >. > >. > >. > >?.. > >>>>This table lists each valid entry for the variable. E.g. A can > >>>>have > >valid values for 1 or 2, if it is not 1 or 2 then the entry is > >invalid. > > > >>>>Other includes invalid entries for that variable and assuming that > >it will pick up the missing values too. > > > >There are both character and numeric fields that needs to be > >validated and put in "test" dataset. I need to get the new_value in > >the variable and also the description field associated with the same. > >(There are around 75 fields that needs to be validated). > > > >I am trying to create the proc formats from control_dataset to create > >"test" dataset. that would have the > > > >I am not sure whether this is an efficient idea or not, but we need > >to implement the same. > > > >Please help.