Here it is: the problem was in my likelihood evaluation: I was parsing
$ML_b directly, while I should have been parsing whatever my program
was supplied through args. Now it does work. See, [ML] should explain
more than just mentioning the names of the globally used macros! :))
On 3/9/06, Stas Kolenikov <[email protected]> wrote:
> Alan,
>
> Indeed, with all zeroes, the log likelihood is = . (as it should be in
> my model). I tried it with my starting values, with Stata's random
> search, whatever -- it is just going that way. It may have to do with
> a constraint I am imposing -- I am fixing one of the parameters, and
> for some reason Stata does not respect that in the random search
> (which I think is a bug!!!), so when it switches from -ml search- to
> -ml max-, it loses a few hundreds in the log-likelihood. That must
> puzzle -ml-; it then takes regressively smaller steps trying to crawl
> back to the nice value it had, and fails.
>
> My previous experience with -ml- was that is was able to incorporate
> the constraint successfully.
>
> On 3/9/06, Feiveson, Alan H. (JSC-SK311) <[email protected]> wrote:
> > Stas - Have you tried changing your initial values? I believe if you
> > don't specify them they default to being zero. If the log likelihood
> > evaluated at all zeros has partials that are also zero, the procedure
> > may not be able to find a place to go next.
> >
> > Al Feiveson
> >
> > -----Original Message-----
> > From: [email protected]
> > [mailto:[email protected]] On Behalf Of Stas
> > Kolenikov
> > Sent: Thursday, March 09, 2006 12:50 PM
> > To: [email protected]
> > Subject: st: lazy -ml-!
> >
> > A really strange thing is happenning with my -ml- programming. I wrote
> > an -lf- program that passes -ml check-, however when it comes to the
> > real -ml max-imization, it just does not make any steps (I am checking
> > with -noi mat li $ML_b- in the whatever_ll.ado) -- and hence my program
> > returns the same value of the likelihood, and hence -ml- exits with
> > r(430) could not calculate numerical derivatives, flat or discontinuous
> > region encountered. That is absolutely weird, it does not depend on
> > starting values, search (which does make good steps imporving the
> > likelihood), it just happens that way!
> >
> > Stata/SE 9.1 is up to date.
> >
> > . which ml
> > C:\Stata9\ado\updates\m\ml.ado
> > *! version 6.2.6 01jul2005
> >
> > . which ml_max
> > C:\Stata9\ado\updates\m\ml_max.ado
> > *! version 7.3.1 30jan2006
> >
> > Has there been anything drastic in the latest updates? I just won't
> > venture digging into ml_max.ado code.
> >
> > --
> > Stas Kolenikov
> > http://stas.kolenikov.name
> >
> > *
> > * For searches and help try:
> > * http://www.stata.com/support/faqs/res/findit.html
> > * http://www.stata.com/support/statalist/faq
> > * http://www.ats.ucla.edu/stat/stata/
> >
> > *
> > * For searches and help try:
> > * http://www.stata.com/support/faqs/res/findit.html
> > * http://www.stata.com/support/statalist/faq
> > * http://www.ats.ucla.edu/stat/stata/
> >
>
>
> --
> Stas Kolenikov
> http://stas.kolenikov.name
>
--
Stas Kolenikov
http://stas.kolenikov.name
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
