BookmarkSubscribeRSS Feed
Pinegulf
Calcite | Level 5

Hello,

 

Stack overflow could not help me so and referred me here. 

ref: http://stackoverflow.com/questions/42804051/obtaining-minimum-maximum-output-of-a-macro-function

 

I want to find to find extreme values produced by a macro as a function of input parameters:

For example purposes here is some data:

data Input_data;
    input X Y;
    cards;
    10 15
    20 18
    30 27
    33 41
    ;
run; 

The following is not the actual formula (as the minimum for this is easily found analytically.) and I want computational method. For examples sake I have semi-empirical rule, which takes three parameters in:

%macro Function(const1, const2 const3);
    data output;
        set input;
        X_eff1=((X > &const1.)*(X - &const1.)**2);
        X_eff2=((X > &const2.)*(X - &const2.)**2);
        X_eff3=((X > &const3.)*(X - &const3.)**2);
        Residual= Y - (1.3*X_eff1 - 2.7*X_eff2+ 3.1*X_eff3);
    run;
%mend;

I want the find const1, const2 const3 which produce the minimum value for variable 'Residual'.  For sake of argument, lets say minimum of sum(abs(residual)).

 Can SAS do this? Few options that are to be considered:

 

a) Find global minimum

b) Find local minimum within boundary conditions, for example: const1[0,1] const2[10,17], const3[20, 22]

I could generate huge table and brute force feed it to the function. However, this is not feasible if the number of input parameter rise.

I could scratch program something?

I could do this in numpy (fmin from cipy.optimize comes to mind)or R if it proves hard for SAS.

Any thoughts on how to solve it?

10 REPLIES 10
RW9
Diamond | Level 26 RW9
Diamond | Level 26

Sounds like a proc iml problem to me all those matrices.  I will however move this over to the stats section.

Astounding
PROC Star

You could program it yourself in this fashion:

 

data want;

set have;

min_residual = 999;

do const1 = 0 to 1 by 0.01;

   do const2 = 10 to 17 by 0.01;

      do const3 = 20 to 22 by 0.01;

         * Calculations using no macro variables, only const1 const2 and const3;

         if abs(residual) < min_residual then do;

            min_residual = abs(residual);

            min_const1 = const1;

            min_const2 = const2;

            min_const3 = const3;

         end;

      end;

   end;

end;

run;

 

You can adjust the fineness of the grid, depending on how many constants you have, how many observations you have, and how much time  you have to wait for the program to complete.  Also, you may want to add another variable to track whether the minimum residual is positive or negative.

Pinegulf
Calcite | Level 5
The brute force approach works in the case of the example.
However, I'm working with rather large datasets (about 1gb) and the actual number of variables may be larger.
Rick_SAS
SAS Super FREQ

The standard ways to solve an optimization problem (constrained or unconstrained) in SAS are:

 

1. Use one of the NLP routines in SAS/IML. For an overview and many links, see the article "Ten tips before you run an optimization" or see the SAS/IML documentation for many examples.

 

2. PROC OPTMODEL in SAS/OR software provides a simpler syntax and a wide variety of solvers.

 

Do you have a license for either SAS/IML and/or SAS/OR software?

Pinegulf
Calcite | Level 5
>Do you have a license for either SAS/IML and/or SAS/OR software?
Sorry, no.
Reeza
Super User

Moved this thread to mathematical optimization, which only helps if you have the module though. 

Do you know what's included in your SAS license? 

Pinegulf
Calcite | Level 5

These are the packages I hae access to:

 

PROC SETINIT; run;

-Base SAS Software
---SAS/STAT
---SAS/GRAPH
---SAS/ETS
---SAS/OR
---SAS/Secure 168-bit
---SAS/Secure Windows
---SAS Enterprise Guide
---OR OPT
---OR PRS
---OR IVS
---OR LSO
---SAS/ACCESS Interface to PC Files
---SAS/ACCESS Interface to OLE DB
---SAS Workspace Server for Local Access
---High Performance Suite

Reeza
Super User

Then look at PROC OPTMODEL as per Ricks suggestion above. 

user24feb
Barite | Level 11
I haven't fully understood what you are up to, but maybe take a look at proc model (http://support.sas.com/documentation/cdl/en/etsug/63939/HTML/default/viewer.htm#etsug_model_sect011....) or proc optmodel.
Ksharp
Super User
http://blogs.sas.com/content/iml/2017/04/12/quadratic-optimization-sas.html

sas-innovate-2024.png

Join us for SAS Innovate April 16-19 at the Aria in Las Vegas. Bring the team and save big with our group pricing for a limited time only.

Pre-conference courses and tutorials are filling up fast and are always a sellout. Register today to reserve your seat.

 

Register now!

Multiple Linear Regression in SAS

Learn how to run multiple linear regression models with and without interactions, presented by SAS user Alex Chaplin.

Find more tutorials on the SAS Users YouTube channel.

Discussion stats
  • 10 replies
  • 1593 views
  • 0 likes
  • 7 in conversation