Re: [Developers] your post about admb convergence

Showing 1-2 of 2 messages
Re: [Developers] your post about admb convergence Mark Maunder 10/21/12 9:40 PM

See below

 

From: Liz Brooks [mailto:liz.b...@noaa.gov]
Sent: Thursday, October 11, 2012 11:19 AM
To: Mark Maunder
Subject: Re: your post about admb convergence

 

hi mark,

thanks for the reply to my question.  it was motivated by a more generic question about whether one can simply look at the gradient and assert that a model has or hasn't converged.  even if the magnitude is not 1e+18 (as in the example online), but say rather that 98% of the parameters have a final gradient between 0.1 and 10, i would argue that one can conclude the model has not converged, particularly when the additional screen info indicates the model cannot improve and is exceeding the number of iterations, and the user-defined convergence criteria is 1e-5.  you may or may not want to weigh in on this more general question, but feel free to.

related to this point, i have a pretty brute force way of saving the final matrix of gradients by saving a screen dump in R and then parsing all of that saved text into a matrix.  who would i have to make a request to at 'ADMB Headquarters' to have the final gradient information printed to a text file? 

cheers, and thanks again,
liz

On Thu, Oct 11, 2012 at 1:35 PM, Mark Maunder <mmau...@iattc.org> wrote:

Liz,

 

That model has definitely not converged due to the high gradient. This type of very high gradient is usually caused by using the  += sign to assign something to the objective function without setting the objective function (or some other parameter, i.e. something used in the posfun function)  to zero at the start of the procedure section.

 

Regards,

 

Mark

 

From: Liz Brooks [mailto:liz.b...@noaa.gov]
Sent: Thursday, October 11, 2012 6:58 AM
To: Mark Maunder
Subject: your post about admb convergence

 

hi mark,

i was reading through some of the replies to questions on the ADMB website pages, and saw this one that both you and dave had replied to.  not sure if you remember it.  i noticed that the convergence criterion specified was 1 e-12, yet all of the gradients are enormous (5.7e+18 is the largest, 1e+6 is the smallest).  the minimization obviously stopped because it exceeded the number of function evaluations and wasn't making progress:
ic > imax  in fminim is answer attained ?

however, if one notes the magnitudes of the gradients, wouldn't you conclude that the model had not attained a minimum?  i've pasted the screed dump from the original question below (and the weblink).  i'm curious how the gradients can be so large yet, as i understand the thread, the parameter estimates were at the "true" solution values.  seems paradoxical.  do you have any insights? 

thanks
liz


https://groups.google.com/forum/?fromgroups=#!topic/admb-users/74AA16IWQNk
Sylvain Bonhommeau

25 variables; iteration 740; function evaluation 907
Function value  -2.7878808e+05; maximum gradient component mag   5.7352e+18
Var   Value    Gradient   |Var   Value    Gradient   |Var   Value    Gradient  
  1-36.23879  4.58537e+02 |  2  0.00190 -3.50573e+18 |  3  1.30460  2.70468e+16
  4  1.37629  7.65838e+15 |  5  0.01472  5.73518e+18 |  6  0.00288  1.15859e+18
  7 -0.03205  1.47477e+17 |  8 -0.52865 -4.26941e+14 |  9 -0.52865  1.04332e+15
 10 -1.47135 -6.57274e+14 | 11  0.12112 -1.61428e+16 | 12  0.16213 -2.80633e+16
 13  0.08573 -1.75030e+16 | 14 -0.00369 -1.96681e+16 | 15 -0.21428  7.12243e+15
 16-11.81483  1.82708e+14 | 17  4.18517  2.72772e+14 | 18  1.81483  3.84233e+13
 19  0.21166  3.25505e+13 | 20  0.09244  1.03861e+16 | 21  0.05250  1.08502e+15
 22  0.26054  2.77481e+15 | 23  0.78956 -9.50897e+13 | 24 210.1177  2.82176e+09
 25 227.8695 -9.76063e+06 |
  ic > imax  in fminim is answer attained ?
Var   Value    Gradient   |Var   Value    Gradient   |Var   Value    Gradient  
  1-36.23882  4.58537e+02 |  2  0.00190 -3.50573e+18 |  3  1.30460  2.70468e+16
  4  1.37629  7.65838e+15 |  5  0.01472  5.73518e+18 |  6  0.00288  1.15859e+18
  7 -0.03205  1.47477e+17 |  8 -0.52865 -4.26941e+14 |  9 -0.52865  1.04332e+15
 10 -1.47135 -6.57274e+14 | 11  0.12112 -1.61428e+16 | 12  0.16213 -2.80633e+16
 13  0.08573 -1.75030e+16 | 14 -0.00369 -1.96681e+16 | 15 -0.21428  7.12243e+15
 16-11.81483  1.82708e+14 | 17  4.18517  2.72772e+14 | 18  1.81483  3.84233e+13
 19  0.21166  3.25505e+13 | 20  0.09244  1.03861e+16 | 21  0.05250  1.08502e+15
 22  0.26054  2.77481e+15 | 23  0.78956 -9.50897e+13 | 24 210.1177  2.82176e+09
 25 227.8695 -9.76063e+06 |
Function minimizer not making progress ... is minimum attained?
Minimprove criterion =   0.0000e+00

 - final statistics:
25 variables; iteration 741; function evaluation 937
Function value  -2.7879e+05; maximum gradient component mag   5.7352e+18
Exit code = 1;  converg criter   1.0000e-12
Var   Value    Gradient   |Var   Value    Gradient   |Var   Value    Gradient  
  1-36.23882  4.58537e+02 |  2  0.00190 -3.50573e+18 |  3  1.30460  2.70468e+16
  4  1.37629  7.65838e+15 |  5  0.01472  5.73518e+18 |  6  0.00288  1.15859e+18
  7 -0.03205  1.47477e+17 |  8 -0.52865 -4.26941e+14 |  9 -0.52865  1.04332e+15
 10 -1.47135 -6.57274e+14 | 11  0.12112 -1.61428e+16 | 12  0.16213 -2.80633e+16
 13  0.08573 -1.75030e+16 | 14 -0.00369 -1.96681e+16 | 15 -0.21428  7.12243e+15
 16-11.81483  1.82708e+14 | 17  4.18517  2.72772e+14 | 18  1.81483  3.84233e+13
 19  0.21166  3.25505e+13 | 20  0.09244  1.03861e+16 | 21  0.05250  1.08502e+15
 22  0.26054  2.77481e+15 | 23  0.78956 -9.50897e+13 | 24 210.1177  2.82176e+09
 25 227.8695 -9.76063e+06 |

--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Liz Brooks, PhD
Operations Research Analyst
NOAA/NMFS
Northeast Fisheries Science Center
166 Water Street                       phone: 508.495.2238
Woods Hole, MA  02543             fax: 508.495.2393

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~




--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Liz Brooks, PhD
Operations Research Analyst
NOAA/NMFS
Northeast Fisheries Science Center
166 Water Street                       phone: 508.495.2238
Woods Hole, MA  02543             fax: 508.495.2393

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Re: [Developers] your post about admb convergence Ian Taylor 10/22/12 1:07 PM
I added the idea of a text output of gradients as Issue #109, but don't have time or skill to figure out how to do this myself.

_______________________________________________
Developers mailing list
Devel...@admb-project.org
http://lists.admb-project.org/mailman/listinfo/developers