[ADMB Users] Understanding the -dd output
mbrooks at ufl.edu
Thu Apr 26 21:08:19 PDT 2012
The thing to check for in this output is that the analytical and finite diff derivatives match.
Since yours match, there seems to be no problem.
Occasionally one will be NA and then you know there's a problem with using that parameter.
NSF IGERT Fellow
University of Florida
mbrooks at ufl.edu
On 26 Apr 2012, at 6:40 PM, Edgar Gonzalez wrote:
> After painstakingly working out my code and having it running, I've been running it with different pin's and having lots of nan's (ca. 90%) and when that doesn't happen the ADMB ends up at the limits of the parameters. Since I need parameter confidence intervals that's not good news.
> I found on the internet that maybe the derivatives where not been correctly calculated and that by using -dd I could see if that's the case.
> However, I don't understand the output:
> Edgar$ ./mipdacmds6 -dd 0
> maxind = 7 maxg = 5878.4
> Enter index (1 ... 10) of derivative to check. To check all derivatives, enter 0: To quit enter -1: 0
> Checking all derivatives. Press X to terminate checking.
> Enter step size (to quit derivative checker, enter 0): 1.e-6
> If you want the derivatives approximated in order
> of decreasing magnitude enter 1
> Else enter 0
> X Function Analytical Finite Diff; Index
> -7.13029e-03 4.21599e+04 5.87840e+03 5.87840e+03 ; 7
> -2.99721e-01 4.21599e+04 4.65079e+03 4.65079e+03 ; 5
> 3.16081e-01 4.21599e+04 3.47460e+03 3.47460e+03 ; 2
> 3.20906e-01 4.21599e+04 3.21187e+03 3.21187e+03 ; 8
> -1.55476e-01 4.21599e+04 2.80267e+03 2.80267e+03 ; 10
> 1.17485e-01 4.21599e+04 1.26323e+03 1.26323e+03 ; 3
> 3.65032e-02 4.21599e+04 -1.20355e+03 -1.20355e+03 ; 9
> 2.57356e-01 4.21599e+04 8.90272e+02 8.90272e+02 ; 6
> 1.34974e-02 4.21599e+04 8.19416e+02 8.19416e+02 ; 1
> -1.05949e-01 4.21599e+04 5.14226e+02 5.14225e+02 ; 4
> Could anyone tell me if there's a problem there with the derivatives?
> Thanks in advance,
> Edgar J. González
> Users mailing list
> Users at admb-project.org
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Users