Hi,

I've hit an issue with the .ext file generated by NM7.1.2.  It appears that if 
the estimation fails (message below), the .ext file doesn't get the header with 
the parameter values:

Estimation failure message:

0PROGRAM TERMINATED BY OBJ
 ERROR IN OBJ2   WITH INDIVIDUAL       1   ID= 1.00000000000000E+00
 CONDITIONAL CONTRIBUTION OR IT'S DERIVATIVE IS INFINITE
 MESSAGE ISSUED FROM ESTIMATION STEP
 AT INITIAL OBJ. FUNCTION EVALUATION
 Elapsed estimation time in seconds:     0.03

 #TERM:
 #TERE:

.ext file contents:

  -1000000000  7.77542E+01  6.38155E-01  1.00000E-01  2.23246E+00  1.00000E+00  
1.00000E-02   1.34078079299425971E+154

It would be helpful to at minimum give the header and preferably give 
estimation failure a new iteration # so that the file would look like:

TABLE NO.     1: First Order Conditional Estimation with Interaction: Goal 
Function=MINIMUM VALUE OF OBJECTIVE FUNCTION
 ITERATION    THETA1       THETA2       THETA3       THETA4       SIGMA(1,1)   
OMEGA(1,1)   OBJ
  -1000001001  7.77542E+01  6.38155E-01  1.00000E-01  2.23246E+00  1.00000E+00  
1.00000E-02   1.34078079299425971E+154

That would make it easier to detect estimation failure-perhaps different 
reasons for estimation failure could be given different values within a range.  
E.g.:


*         -1000001000 - Other estimation failure

*         -1000001001 - PROGRAM TERMINATED BY OBJ

*         -1000001002 - ...

That way detection of automatic errors could be found by searching for 
iteration numbers between -1000001000 and -1000002000, and more sophisticated 
programs could handle some or all reasons for termination differently.

Thanks,

Bill

Reply via email to