The filter routines there have names and should follow calling rules for the par matlab/kalman/likelihood ones but they do not yet take mf and still expect Z matrix of 1s to map system to observables instead and Pinf must not be empty but at least be initialised to a zeros matrix of Pstar size.
 
The smoother is generic as it was, not suitable for running at all
 
Best regards
 
George
 
----- Original Message -----
From: Stéphane Adjemian
To: List for Dynare developers
Sent: Wednesday, May 27, 2009 2:46 PM
Subject: Re: [DynareDev] Kalman Filter

Hi George

2009/5/27 G. Perendia <george@perendia.orangehome.co.uk>
Hi
 
The C++ Kalman dll driver  routines in /mex/kalman/matlab are still in development

Sure...
 
and the code there is still subject to developer testing

... But I'd like to play with it.
 
and the Makefile there is  not yet complete but it should use blas/lapack from  Matlab libraries.

Ok. I will adapt your Makefile for my platform.

Best, Stéphane.

 
 
 
Best regards
 
George Perendia
Tel.: 02072815392
Mob: 07951415480
----- Original Message -----
Sent: Wednesday, May 27, 2009 2:18 PM
Subject: Re: [DynareDev] Kalman Filter

Oui, oui... I never use the presampling option... And with all the examples I have ever considered it takes much more than 10 iterations to get to the steady state kalman filter.

By the way, is there somewhere linux a Makefile for the cc kalman routines? Also, why do we need Atlas (if we also use matlab's lapack/blas libraries)?

Stéphane. 

2009/5/27 Michel Juillard <michel.juillard@ens.fr>
Si le filtre converge a l'etat stationnaire a une date avant start (la date a partir de laquelle on cumule la vraisemblance), alors additionner tous les determinants des iterations dans le filtre stationnaire a la derniere periode fausse le calcul de la constante.

amicalement

Michel

Stéphane Adjemian wrote:
Thanks Michel. Your commit is ok for me. I do not yet understand the problem raised by George. I need to go through his example...

Stéphane.

2009/5/27 Michel Juillard <michel.juillard@ens.fr <mailto:michel.juillard@ens.fr>>

   Thanks Stephane,

   I just uploaded a new version of kalman_filter.m Tell me what you
   think

   Best

   Michel

   Stéphane Adjemian wrote:

       Hi all,

       I agree, the matlab code is very unclear (even if I had fun
       writting it this way ;-) and prone to errors if one uses the
       vector lik (Marco is using it). I would rather prefer to add
       the constants outside of the loop with a (sub)vector
       operation, this should be more efficient. I will do it today
       or tomorrow.

       Best,
       Stéphane.

       2009/5/27 Michel Juillard <michel.juillard@ens.fr
       <mailto:michel.juillard@ens.fr> <mailto:michel.juillard@ens.fr

       <mailto:michel.juillard@ens.fr>>>


          On closer inspection, I don't think that the expression
       pointed by
          George in kalman_filter.m is wrong:

          1. reste = smpl-t or the number of periods during which the
       filter
          is stationary. This shouldn't be larger than T-start+1

          2. it is problematic (see below) but not wrong to add all the
          determinants at once in the last period of the stationary
       filter

          3. I don't think this explains the difference with the C++
       version
          of the filter and we still have to look for it.

          4. it remains that the current code is very unclear and that if
          LIK is correct the vector lik doesn't have the correct
       constants
          on each elements.

          5. I would like to simplify the code and add the correct
       constant
          to each element of the lik vector. It would be a little bit
       less
          efficient in Matlab than the current code, but I doubt it
       would be
          noticeable.
          Stephane, what do you think?


          Best

          Michel

          G. Perendia wrote:

              Dear Michel

              I think I found an error in Dynare Matlab
       kalman_filter. suite
              of utilities
              which affects the likelihood LIK  results with start>1
       (i.e.
              presampling>0):

              the calculation speed-up construct which relies on
       converged
              covariance
              matrix

              lik(t) = lik(t) + reste*log(dF);

              adds reste * log(dF) to the last-1 (i.e. the smpl)
       member of lik
              (the last, the lik(smpl+1) one contains smpl*pp*log(2*pi))
              but reste is usually larger than T-start+1 so that

              LIK = .5*(sum(lik(start:end))-(start-1)*lik(smpl+1)/smpl)

              has much more log(dF)s added than required since they
       are all
              concentrated
              in the last-1 (the T) member

              For example, if I change the above construct to
              lik(t) = lik(t) + min(reste,(smpl-start+1))*log(dF);

              the reported likelihood for presample=40 from Matlab KF is
              1640935.5855267849
              which is nearly the same as that from C++ KF below:
              1640935.5854489324

              Shall I make changes to kalman/likelihood/ KFs and
       upload the
              .m files?
              This problem affects also the older versions of
              DiffuseLikelihood**.m  too.

              Best regards

              George
              artilogica@btconnect.com
       <mailto:artilogica@btconnect.com>
       <mailto:artilogica@btconnect.com

       <mailto:artilogica@btconnect.com>>


              ----- Original Message ----- From: "Michel Juillard"
              <michel.juillard@ens.fr <mailto:michel.juillard@ens.fr>
       <mailto:michel.juillard@ens.fr <mailto:michel.juillard@ens.fr>>>


              To: "G. Perendia" <george@perendia.orangehome.co.uk
       <mailto:george@perendia.orangehome.co.uk>
              <mailto:george@perendia.orangehome.co.uk
       <mailto:george@perendia.orangehome.co.uk>>>

              Sent: Tuesday, May 26, 2009 10:32 AM
              Subject: Re: Kalman Filter+PS


                                Hi George,


                                          Re 1) below:
                      I modified C++ KF so that it reports log-likelihood
                      for given
                      start/preampling in same/similar manner as the
       Matlab
                      KFs do and I am
                      getting approximately close results, e.g.

                      ll=  -1640935.5854489324 for C++ and
                       (-) 1640482.4179242959 for Matlab KF (for
       start=41, i.e.
                                        presample=40).
                                    whilst they appear same for presample=0
                      (e.g.2.5906e+006), i.e.
                      -2590556.989730841 vs
                       2590556.989778722

                      Are those results acceptably close or should I
                      investigate further where
                                        the
                                    above difference may come form?


                                            This indicates a problem . The difference should be the
                  same with and
                  without presample. It may come from the computation
       of the
                  likelihood
                  constant. This is done in a very obscure manner in
       Dynare
                  Matlab.
                   


             

          _______________________________________________
          Dev mailing list
          Dev@dynare.org <mailto:Dev@dynare.org>
       <mailto:Dev@dynare.org <mailto:Dev@dynare.org>>


          http://www.dynare.org/cgi-bin/mailman/listinfo/dev




       --        Stéphane Adjemian
       CEPREMAP & Université du Maine

       Tel: (33)1-43-13-62-39
       ------------------------------------------------------------------------



       _______________________________________________
       Dev mailing list
       Dev@dynare.org <mailto:Dev@dynare.org>
       http://www.dynare.org/cgi-bin/mailman/listinfo/dev
       


   _______________________________________________
   Dev mailing list
   Dev@dynare.org <mailto:Dev@dynare.org>
   http://www.dynare.org/cgi-bin/mailman/listinfo/dev




--
Stéphane Adjemian
CEPREMAP & Université du Maine

Tel: (33)1-43-13-62-39
------------------------------------------------------------------------

_______________________________________________
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



--
Stéphane Adjemian
CEPREMAP & Université du Maine

Tel: (33)1-43-13-62-39


_______________________________________________
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




--
Stéphane Adjemian
CEPREMAP & Université du Maine

Tel: (33)1-43-13-62-39


_______________________________________________
Dev mailing list
Dev@dynare.org
http://www.dynare.org/cgi-bin/mailman/listinfo/dev