route the sas iml log into an external file

[Hanyu]

Hi, I want to route the log generated by SAS IML to an external file. I have a loop to execute and I want to route the log of each loop to a different file. I have used proc printto procedure in combination with submit and endsubmit statement to execute the procedure inside SAS IML. However the result I got is empty log file.

The code looks like this:

 proc iml;

do i=1 to 7;

submit i;

proc printto log="D:\SAS generated files\log &i.log" new;

run;

endsubmit;

end;

quit;

[Rick_SAS]

Could you give us some context and tell us what you are trying to accomplish? In other words, what are you trying to do statistically/numerically that you think will become easier if you can redirect the Log?

[Hanyu]

Hi Rick. I have a long time series of bond price and I am trying to compute
the yield to maturity by mininizing the distance between the discounted
coupon value and par value and the quoted dirty bond price. Therefore I
need to do a lot of optimisation for each bond price. Some of the
optimisation cannot converge probably because I cannot manually try
different starting values; There are too many bond prices. I want to
analyze the log file each time I try a different starting value for the
entire series to see which starting value generate the least amount of no
convergence.
##- Please type your reply above this line. Simple formatting, no
attachments. -##

[Rick_SAS · Accepted solution]

OK. I assume that you are using a built-in NLP routine (such as NLPNRA) to compute each optimization.  The first argument to an NLP function is a return code (rc). So your call looks like this:

call nlpnra(rc, result, "ObjectiveFunc", InitGuess, options);

When the function returns, the value of the return code will be positive if the optimization converged and negative if the optimization did not converge.  Therefore you can save the value of each return code and examine them later, together with the initial guess.  Here's some pseudocode to get you started:

proc iml;
...
convergence = j(100,1,.);
initialGuess = j(100, numParams);
do i = 1 to 100;
   /* set i_th guess from file or randomly or systematically */
   initGuess = T( randfun(NumParams, "Normal") );
   call nlpnra(rc, result, "ObjectiveFunc", InitGuess, options);
   /* save the initial guess and the return code */
   convergence[i] = rc;
   initialGuess[i,] = initGuess;
end;

/* now analyze relationship between convergence and initial guess */

You might be interested in reading this article about how to choose a good starting guess for an optimization.