Dear Marco,
I'm trying to remove global variable objective_function_penalty_base that is used by mode_compute 4 and 5.
Doing that, I remarked things that I don't understand concerning mode_compute 5 in master:
1) in ./matlab/optimization/dynare_minimize_objective.m, newrat() is called on line 247 with the 6th input argument set to zero. My guess is that it should be newflagrat, otherwise I don't understand what newflgrat is used for. In any case, the current behavior and results seem different from what we had in the stable version
2) looking at the code in ./matlab/optimization/newrat.m, I understand that the final value of the hessian is computed at the end of this function according to user's options. I don't understand the reason for the code in ./matlab/dynare_estimation_1. lines 262-288
3) Is it necessary for newrat() to reset options_.analytic_derivation and have ./matlab/dynare_estimation_1.m lines 241-242?
My impression is that we have changed the code without fully understanding how newrat.m is working.
Best
Dear Michel, thank you I will look into these issues. The code is rather cryptic and badly documented, which makes changes and updates difficult... I also did some changes at some point. let you know best Marco
On 09/20/15, Michel Juillard michel.juillard@mjui.fr wrote:
Dear Marco,
I'm trying to remove global variable objective_function_penalty_base that is used by mode_compute 4 and 5.
Doing that, I remarked things that I don't understand concerning mode_compute 5 in master:
- in ./matlab/optimization/dynare_minimize_objective.m, newrat() is
called on line 247 with the 6th input argument set to zero. My guess is that it should be newflagrat, otherwise I don't understand what newflgrat is used for. In any case, the current behavior and results seem different from what we had in the stable version
- looking at the code in
./matlab/optimization/newrat.m, I understand that the final value of the hessian is computed at the end of this function according to user's options. I don't understand the reason for the code in ./matlab/dynare_estimation_1. lines 262-288
- Is it necessary for newrat() to reset options_.analytic_derivation
and have ./matlab/dynare_estimation_1.m lines 241-242?
My impression is that we have changed the code without fully understanding how newrat.m is working.
Best
Michel Juillard _______________________________________________ Dev mailing list Dev@dynare.org https://www.dynare.org/cgi-bin/mailman/listinfo/dev
Thanks Marco
Michel
Marco Ratto writes:
Dear Michel, thank you I will look into these issues. The code is rather cryptic and badly documented, which makes changes and updates difficult... I also did some changes at some point. let you know best Marco
On 09/20/15, Michel Juillard michel.juillard@mjui.fr wrote:
Dear Marco,
I'm trying to remove global variable objective_function_penalty_base that is used by mode_compute 4 and 5.
Doing that, I remarked things that I don't understand concerning mode_compute 5 in master:
- in ./matlab/optimization/dynare_minimize_objective.m, newrat() is
called on line 247 with the 6th input argument set to zero. My guess is that it should be newflagrat, otherwise I don't understand what newflgrat is used for. In any case, the current behavior and results seem different from what we had in the stable version
- looking at the code in
./matlab/optimization/newrat.m, I understand that the final value of the hessian is computed at the end of this function according to user's options. I don't understand the reason for the code in ./matlab/dynare_estimation_1. lines 262-288
- Is it necessary for newrat() to reset options_.analytic_derivation
and have ./matlab/dynare_estimation_1.m lines 241-242?
My impression is that we have changed the code without fully understanding how newrat.m is working.
Best
Michel Juillard _______________________________________________ Dev mailing list Dev@dynare.org https://www.dynare.org/cgi-bin/mailman/listinfo/dev
Dev mailing list Dev@dynare.org https://www.dynare.org/cgi-bin/mailman/listinfo/dev
-
Marco Ratto -
Michel Juillard