Hi,
I'm using optmum to estimate a highly non-linear model by Indirect Inference. Optmum stops after one iteration indicating normal convergence with gradients that are far from zero. Anybody knows what that mean? Is this an identification problem? Thanks for any help!
3 Answers
0
Could you confirm that the return code was zero? There are two possibilities. First, that it failed to compute the objective function at the first iteration. In this case it returns with an error code. Later failures Optmum is able to recover from, but not the first one.
The second possibility is that there is a serious scaling problem and that that large gradient looks small in context.
If it is the second case you will need to scale your data. Mixing small numbers with large numbers in a computer can be catastrophic.
If it is the first case, using the debugger, go to your objective function to see where the calculations go wrong.
0
Thank you for the answer. There return code is 0 and there is no error message. All the variables in the model are in the 0-1 range. The model requires smoothing so I tried to run it for different values of the smoothing parameter. For some values of the smoothing parameter the programs does up to 5 iterations but then stops indicating normal convergence with gradients that are between 0 and 0.1.
0
If you could send a copy of your command file and any other required files compressed into a zip file to [email protected], and I'll take a look at it.
Your Answer
3 Answers
Could you confirm that the return code was zero? There are two possibilities. First, that it failed to compute the objective function at the first iteration. In this case it returns with an error code. Later failures Optmum is able to recover from, but not the first one.
The second possibility is that there is a serious scaling problem and that that large gradient looks small in context.
If it is the second case you will need to scale your data. Mixing small numbers with large numbers in a computer can be catastrophic.
If it is the first case, using the debugger, go to your objective function to see where the calculations go wrong.
Thank you for the answer. There return code is 0 and there is no error message. All the variables in the model are in the 0-1 range. The model requires smoothing so I tried to run it for different values of the smoothing parameter. For some values of the smoothing parameter the programs does up to 5 iterations but then stops indicating normal convergence with gradients that are between 0 and 0.1.
If you could send a copy of your command file and any other required files compressed into a zip file to [email protected], and I'll take a look at it.