Numerical issues in parameter estimation

74 views
Skip to first unread message

Deepa

unread,
Jun 13, 2022, 5:39:24 AM6/13/22
to COPASI User Forum
Dear All,

I'm using Evolution Strategy  method to solve a paramter estimation problem.I terminated the run after running the task for more than 2 hours. I checked the objective values and the objective value of one of the experiments was inf  and I find the following message


>EXCEPTION 2022-06-11T19:42:48<
  CTrajectoryMethod (6): Deterministic integration failed. LSODA reported:
  DLSODA-  At current T (=R1), MXSTEP (=I1) steps  
        taken on this call before reaching TOUT    
      In above message, I1 = '100000'
      In above message, R1 = '262842'
  DLSODA-  At current T (=R1), MXSTEP (=I1) steps  
        taken on this call before reaching TOUT    
      In above message, I1 = '100000'
      In above message, R1 = '262842'
   
 
  Please see result for indications of numerical instability.
>WARNING 2022-06-12T12:13:09<
  CFitting (12): The Fisher information matrix is singular. Therefore, the correlation matrix can not be calculated.

I'm not sure what `The Fisher information matrix is singular` means. I'm not sure if this could be due to issues in the model elements. Could someone please explain this a bit?


Thanks a lot,
Deepa

 


Frank Bergmann

unread,
Jun 13, 2022, 7:19:39 AM6/13/22
to COPASI User Forum
Hello Deepa, 

when running parameter estimation, it is quite common, to get into regions, where simulation becomes numerical difficult. During the time course simulation the integrator would have needed more steps, then the default '100000' specified in the Time-Course Task. If you change the value of 'Max Internal steps', this might help. Though you might also want to double check the bounds on the parameters to estimate, to see whether you could narrow them more to  avoid those regions. 

As for the Fisher information matrix, this is a metric, that basically tells you how much the provided data tells you about the parameters you are estimating. If it is singular, that means there are rows in there with zeros, which gives you an indication that there is little information for this parameter in the data you provided. 

best
Frank
Reply all
Reply to author
Forward
0 new messages