Re: [DynareDev] Fw: [Trac] #60: Fix memory leaks in k_order_perturbation DLL
Le lundi 30 novembre 2009 à 17:50 +0000, G. Perendia a écrit :
Is there any more info on this issue?
Please see my comment associated to the #60 issue on the trac system. Please answer there if you want to continue the discussion. -- Sébastien Villemot
Hi Sebastien There appears to be a slight problem with the current implementation of k_order_pert.m. It is called from the top of dr1.m and dr1.m then returns after running it but it then does not appear to provide for the log-linear adjustment of ghx and ghu usually performed further down the line in dr1.m when if the model is log-linear - though I may be missing something. There is also a C++ log-linear adjustment built in the C++ version of the dr1 for C++ DsgeLikelihood and that code may be integrated into k_order_perturbation.cpp too to run adjustment within C++ when if needed. Best regards George
Hi George, you are right. Thanks for pointing it out. Best Michel G. Perendia wrote:
Hi Sebastien
There appears to be a slight problem with the current implementation of k_order_pert.m.
It is called from the top of dr1.m and dr1.m then returns after running it but it then does not appear to provide for the log-linear adjustment of ghx and ghu usually performed further down the line in dr1.m when if the model is log-linear - though I may be missing something.
There is also a C++ log-linear adjustment built in the C++ version of the dr1 for C++ DsgeLikelihood and that code may be integrated into k_order_perturbation.cpp too to run adjustment within C++ when if needed.
Best regards
George
_______________________________________________ Dev mailing list Dev@dynare.org http://www.dynare.org/cgi-bin/mailman/listinfo/dev
I have copied the log-linear correction from dr1 to k_ord_pert.m as a quick fix and can upload that once I test it if that is ok. Best regards George ----- Original Message ----- From: "Michel Juillard" <michel.juillard@ens.fr> To: "List for Dynare developers" <dev@dynare.org> Sent: Wednesday, December 02, 2009 8:02 PM Subject: Re: [DynareDev] k_order_perturbation DLL DR1and log-linear models
Hi George,
you are right. Thanks for pointing it out.
Best
Michel
G. Perendia wrote:
Hi Sebastien
There appears to be a slight problem with the current implementation of k_order_pert.m.
It is called from the top of dr1.m and dr1.m then returns after running it but it then does not appear to provide for the log-linear adjustment of ghx and ghu usually performed further down the line in dr1.m when if the model is log-linear - though I may be missing something.
There is also a C++ log-linear adjustment built in the C++ version of the dr1 for C++ DsgeLikelihood and that code may be integrated into k_order_perturbation.cpp too to run adjustment within C++ when if needed.
Best regards
George
_______________________________________________ Dev mailing list Dev@dynare.org http://www.dynare.org/cgi-bin/mailman/listinfo/dev
_______________________________________________ Dev mailing list Dev@dynare.org http://www.dynare.org/cgi-bin/mailman/listinfo/dev
Good idea. Thanks George Best, Michel G. Perendia wrote:
I have copied the log-linear correction from dr1 to k_ord_pert.m as a quick fix and can upload that once I test it if that is ok.
Best regards
George ----- Original Message ----- From: "Michel Juillard" <michel.juillard@ens.fr> To: "List for Dynare developers" <dev@dynare.org> Sent: Wednesday, December 02, 2009 8:02 PM Subject: Re: [DynareDev] k_order_perturbation DLL DR1and log-linear models
Hi George,
you are right. Thanks for pointing it out.
Best
Michel
G. Perendia wrote:
Hi Sebastien
There appears to be a slight problem with the current implementation of k_order_pert.m.
It is called from the top of dr1.m and dr1.m then returns after running
it
but it then does not appear to provide for the log-linear adjustment of
ghx
and ghu usually performed further down the line in dr1.m when if the
model
is log-linear - though I may be missing something.
There is also a C++ log-linear adjustment built in the C++ version of
the
dr1 for C++ DsgeLikelihood and that code may be integrated into k_order_perturbation.cpp too to run adjustment within C++ when if
needed.
Best regards
George
_______________________________________________ Dev mailing list Dev@dynare.org http://www.dynare.org/cgi-bin/mailman/listinfo/dev
_______________________________________________ Dev mailing list Dev@dynare.org http://www.dynare.org/cgi-bin/mailman/listinfo/dev
_______________________________________________ Dev mailing list Dev@dynare.org http://www.dynare.org/cgi-bin/mailman/listinfo/dev
Hi Sebastien I added the loglinear correction and the 2nd output argument for the 1st order as there must be 1+order output variables to k_order_pert.m However: 1) Why stoch_simul is not accepting loglinear option. I may be missing something but my expectation is that would yield incorrect IRFs (and few other results for stoch_simul) when if used with loglinear models (that is, when if not run after the estimation with loglinear option!). Temporary workaround (or consistency issue?!): Run stoch_simul after estimation with loglinear option and stoch_simul will "inherit" options_.loglinear=1 ! 2) I guess the new preprocessor does not support k-order_solver for estimation as it may be misleading. as the estimation currently works for 1st order only, (though there is still a minor performance benefit form using k_order_solver in estimation even for the 1st order). However, if one runs estimation after stoch_simul with k_order_solver or order=3 option, the estimation will "inherit" the k_order_solver option and use it for the 1st order only anyway! Further suggestions 3) use of k-order_solver should implicitly enforce use_dll model option, 4) keyword loglinear could be a common feature of the model (i.e. alongside of linear) regardless of the operation but we would need to retain current use in estimation for downwards compatibility. Best regards George ----- Original Message ----- From: "Michel Juillard" <michel.juillard@ens.fr> To: "List for Dynare developers" <dev@dynare.org> Sent: Wednesday, December 02, 2009 8:57 PM Subject: Re: [DynareDev] k_order_perturbation DLL DR1and log-linear models
Good idea. Thanks George Best,
Michel
G. Perendia wrote:
I have copied the log-linear correction from dr1 to k_ord_pert.m as a quick fix and can upload that once I test it if that is ok.
Best regards
George ----- Original Message ----- From: "Michel Juillard" <michel.juillard@ens.fr> To: "List for Dynare developers" <dev@dynare.org> Sent: Wednesday, December 02, 2009 8:02 PM Subject: Re: [DynareDev] k_order_perturbation DLL DR1and log-linear models
Hi George,
you are right. Thanks for pointing it out.
Best
Michel
G. Perendia wrote:
Hi Sebastien
There appears to be a slight problem with the current implementation of k_order_pert.m.
It is called from the top of dr1.m and dr1.m then returns after running
it
but it then does not appear to provide for the log-linear adjustment of
ghx
and ghu usually performed further down the line in dr1.m when if the
model
is log-linear - though I may be missing something.
There is also a C++ log-linear adjustment built in the C++ version of
the
dr1 for C++ DsgeLikelihood and that code may be integrated into k_order_perturbation.cpp too to run adjustment within C++ when if
needed.
Best regards
George
_______________________________________________ Dev mailing list Dev@dynare.org http://www.dynare.org/cgi-bin/mailman/listinfo/dev
_______________________________________________ Dev mailing list Dev@dynare.org http://www.dynare.org/cgi-bin/mailman/listinfo/dev
_______________________________________________ Dev mailing list Dev@dynare.org http://www.dynare.org/cgi-bin/mailman/listinfo/dev
_______________________________________________ Dev mailing list Dev@dynare.org http://www.dynare.org/cgi-bin/mailman/listinfo/dev
Hi George, thanks for the correction, but I removed yss from the outputs of k_order_perturbation DLL. This DLL doesn't handle the steady state. It is provided to the DLL both as oo_.steady_state and dr.ys. In the log-linear lines, we should just use dr.ys What is k_order_pert_opt ?
1) Why stoch_simul is not accepting loglinear option. I may be missing something but my expectation is that would yield incorrect IRFs (and few other results for stoch_simul) when if used with loglinear models (that is, when if not run after the estimation with loglinear option!).
loglinear was an ill thought option that brings more drawbacks than advantages. This needs to be rethought. We keep it because it works for fs2000 estimation example, but I don't put emphasis on it. You are right, making it available in stoch_simul would require making the output clear and consistent. The problems that you point to are related to the global character of options and we will deal with it when we make options local to a Dynare instruction (version 4.2)
Temporary workaround (or consistency issue?!):
Run stoch_simul after estimation with loglinear option and stoch_simul will "inherit" options_.loglinear=1 !
2) I guess the new preprocessor does not support k-order_solver for estimation as it may be misleading. as the estimation currently works for 1st order only, (though there is still a minor performance benefit form using k_order_solver in estimation even for the 1st order).
The goal was to make 3rd order solution available. Using k_order for order = 1, 2 is possible, but only for people who know what they are doing. However, I'm surprised that k_order_solver is not accepted by the preprocessor for estimation. I thought that all stoch_simul options were made available to estimation instruction. This needs to be checked.
However, if one runs estimation after stoch_simul with k_order_solver o order=3 option, the estimation will "inherit" the k_order_solver option and use it for the 1st order only anyway!
Further suggestions 3) use of k-order_solver should implicitly enforce use_dll model option,
We decided to avoid too many implicit behviors. The presence of use_dll should be checked and an error reported if k_order_solver is used without use_dll. This important to attract the user's attention on the need to have a proper compiler installed with Matlab on their machine.
4) keyword loglinear could be a common feature of the model (i.e. alongside of linear) regardless of the operation but we would need to retain current use in estimation for downwards compatibility.
Yes, with the possibility to transform only some variables. This should be put in version 4.2 All the best, Michel
Best regards
George
----- Original Message ----- From: "Michel Juillard" <michel.juillard@ens.fr> To: "List for Dynare developers" <dev@dynare.org> Sent: Wednesday, December 02, 2009 8:57 PM Subject: Re: [DynareDev] k_order_perturbation DLL DR1and log-linear models
Good idea. Thanks George Best,
Michel
G. Perendia wrote:
I have copied the log-linear correction from dr1 to k_ord_pert.m as a
quick
fix and can upload that once I test it if that is ok.
Best regards
George ----- Original Message ----- From: "Michel Juillard" <michel.juillard@ens.fr> To: "List for Dynare developers" <dev@dynare.org> Sent: Wednesday, December 02, 2009 8:02 PM Subject: Re: [DynareDev] k_order_perturbation DLL DR1and log-linear
models
Hi George,
you are right. Thanks for pointing it out.
Best
Michel
G. Perendia wrote:
Hi Sebastien
There appears to be a slight problem with the current implementation
of
k_order_pert.m.
It is called from the top of dr1.m and dr1.m then returns after
running
it
but it then does not appear to provide for the log-linear adjustment
of
ghx
and ghu usually performed further down the line in dr1.m when if the
model
is log-linear - though I may be missing something.
There is also a C++ log-linear adjustment built in the C++ version of
the
dr1 for C++ DsgeLikelihood and that code may be integrated into k_order_perturbation.cpp too to run adjustment within C++ when if
needed.
Best regards
George
_______________________________________________ Dev mailing list Dev@dynare.org http://www.dynare.org/cgi-bin/mailman/listinfo/dev
_______________________________________________ Dev mailing list Dev@dynare.org http://www.dynare.org/cgi-bin/mailman/listinfo/dev
_______________________________________________ Dev mailing list Dev@dynare.org http://www.dynare.org/cgi-bin/mailman/listinfo/dev
_______________________________________________ Dev mailing list Dev@dynare.org http://www.dynare.org/cgi-bin/mailman/listinfo/dev
_______________________________________________ Dev mailing list Dev@dynare.org http://www.dynare.org/cgi-bin/mailman/listinfo/dev
participants (3)
-
G. Perendia -
Michel Juillard -
Sébastien Villemot