Page 1 of 1
restar of MH algorithm from the point of interruption
Posted:
Wed Sep 22, 2010 3:32 pm
by tanya
Suppose computer was accidentally shut down during MH algorithm. The hard drive contains several *_mh*_blck*.mat files. Is it possible to start from the point close to the point of interruption? Or do I need to start from the beginning?
I appreciate there is load_mh_file command, but I cannot find the proper usage of this command. If I write
estimation(datafile=nameq,lik_init=2,nobs=n1,mode_compute=0,mode_file=name_mode,load_mh_fil, mh_replic= 150000,mh_scale=0.35,bayesian_irf,irf=100)
it complains that *_mh14_blck2 does not exist. It does not exist, indeed, as I have fewer files than needed to complete. It crashed on *_mh5_blck2
Is it possible to start from what what exists?
PS. I run Dynare 3.064, but I suspect this would not matter.
Many thanks
Re: restar of MH algorithm from the point of interruption
Posted:
Thu Sep 23, 2010 1:10 pm
by StephaneAdjemian
Dear Tanya, You have to use option mh_recover instead of load_mh_file. For obvious safety reasons, you should make a copy of your mh_files before trying this option, I am not certain it works correctly with dynare 3.xxx).
Best, Stéphane.
Re: restar of MH algorithm from the point of interruption
Posted:
Thu Sep 23, 2010 1:37 pm
by tanya
I did try this, but it did not recognise the command. As no help for version 3 is available any more, I cannot check.
On a different note, can I ask if there is any hope that you will make Dynare 4,5,... to call external functions with MULTIPLE variables output? I am kinda waiting (this is the reason I use Dynare 3), if there is no hope, I will try to invest into something different...
Re: restar of MH algorithm from the point of interruption
Posted:
Thu Sep 23, 2010 3:33 pm
by StephaneAdjemian
I can't remember when we added the keyword
mh_recover. But I am sure there was a recover mode with dynare 3. Maybe you can try to add
- Code: Select all
options_.mh_recover = 1;
before the estimation command.
It's possible (not convenient) to use n external functions instead of one external function returning n arguments. The first external function calls a function that put the n variables to be computed in globals (or in a global structure) and each external function returns one of the global variables (or one field of the global structure). It's not very nice...
I don't use external functions and prefer to compute what I need in the steady state file (the steady state file is called before the *_dynamic.m file evaluation). External variables are only needed if you consider functions of the endogenous variables.
Best, Stéphane.
Re: restar of MH algorithm from the point of interruption
Posted:
Mon Oct 11, 2010 9:30 am
by SébastienVillemot
Note that, using the current unstable version of Dynare, you can already use functions with multiple return arguments by using a small trick: just create wrapper functions that call your original function and return the desired argument. You need to create as many such functions as you have return arguments.
I know this is only a workaround, but it should work fine and do what you want.
I’m not sure that we will implement the syntax that you want, because it may imply too many changes: could you give an example of the syntax that you would like to have in Dynare 4 ?