How to minimize function per group of observations with Proc Optmodel?

xyz3 — Mon, 11 Jun 2018 11:33:42 GMT

Hi, I am trying to minimize a function (that I created with proc fcmp) per group of observations that are defined by a category column. I tried using this example (http://support.sas.com/kb/42/332.html) for by group processing with proc optmodel but I'm not sure if the way I'm indexing the observations is correct.

This is my code:

options cmplib=work.funcs;
%let byvar=pat_ID;

proc optmodel cdigits=10 fdigits=10 pdigits=9;

	set OBS;
	str pat_ID {OBS};
	num UW_acc{OBS};
	num EP_incr{OBS};
	num Time{OBS};

	read data work.pat_data into OBS=[_n_] pat_ID=pat_ID;
	read data work.pat_data into OBS=[_n_] UW_acc=UW_pattern_acc;
	read data work.pat_data into OBS=[_n_] EP_incr=EP_pattern_incr;
	read data work.pat_data into OBS=[_n_] Time=Year;

	/* Collect values of BY variable into an index set */
	set BYSET = setof {i in OBS} &byvar.[i];
	put BYSET=;
	str by;

	/* variables */
	set OBS_BY = {i in OBS: &byvar.[i] = by};

	var a >=&LOW_A. <= &UP_A. init &INITA.;
	var b >=&LOW_B. <= &UP_B. init &INITB.;
	
	minimize FuncVal = pat_opt(a, b, UW_acc[OBS_BY], EP_incr[OBS_BY], Time[OBS_PY]);

	/*variables to store result*/
	num a_sol {BYSET};
	num b_sol {BYSET};

	do by = BYSET;
		put by=;
		solve with nlp / multistart=(maxstarts=&MAXSTART.) seed=&SEED. maxiter=&MAXITER. opttol=&OPTTOL. feastol=&FEASTOL.;
		print a b;
		a_sol[by] = a.sol;
		b_sol[by] = b.sol;
	end;

	create data work.solver_results from [by] a=a_sol b=b_sol;

run; 

proc print data=work.solver_results;
run;

quit;

TITLE;

When running the code I get "subscript 1 may not be a set" as an error for this line for each variable uw_acc, ep_incr and time.

minimize FuncVal = pat_opt(a, b, UW_acc[OBS_BY], EP_incr[OBS_BY], Time[OBS_PY]);

Let me know, if you need more info!

Thanks!

(I'm using SAS Version 9.4)

Re: How to minimize function per group of observations with Proc Optmodel?

RobPratt — Mon, 11 Jun 2018 14:20:20 GMT

The argument to UW_acc should be an element of the set OBS. but you are passing OBS_BY, which is instead a subset of OBS. Same story for EP_incr and Time. What does your pat_opt function expect as input?

By the way, you can combine the READ DATA statements as follows:

read data work.pat_data into OBS=[_n_] pat_ID=pat_ID UW_acc=UW_pattern_acc EP_incr=EP_pattern_incr Time=Year;

topic Re: How to minimize function per group of observations with Proc Optmodel? in Mathematical Optimization, Discrete-Event Simulation, and OR

How to minimize function per group of observations with Proc Optmodel?

Re: How to minimize function per group of observations with Proc Optmodel?