Re: st: numerical derivatives and -ml- command

[Nick Cox <njcoxstata@gmail.com>]

There's another simple comment that doesn't contradict Maarten's line.

Sun Yutao is asking for more documentation on -ml-, even asking for
further details about Stata's hidden proprietary code. My guess is
that this is a case where those who know cannot speak. There is
already a large fraction of [P] and an entire book written by
StataCorp staff about -ml- and although they evidently don't answer
Sun Yutao's questions I doubt that there will be more comment. Even
user-programmers with much experience of using -ml- as an engine tend
to treat it as a black box, partly they because they have no choice
and partly because it's a good strategy any way.

Nick

On Tue, Nov 6, 2012 at 8:52 AM, Maarten Buis <maartenlbuis@gmail.com> wrote:
> On Mon, Nov 5, 2012 at 10:26 PM, Sun Yutao wrote:
>> I'm trying to write an maximum likelihood algorithm that can handle a
>> particularly large number of fixed effects (which is not possible with Stata's
>> #var and matsize limitation). I have been comparing the performance of my ml
>> with the Stata -ml- these days and I find something really strange.
>
> Writing your own maximizer is something I would not do. These are
> extremely subtle problems, and it is just way too easy to get it
> wrong. Moreover, as you noticed Stata's maximizers are very good. It
> is best to leave these kinds of problems to the professionals, you
> already paid for their work after all. Typically that means you just
> have to be a bit more creative in thinking about such problems such
> that you can use Stata's maximizers even with a large number of fixed
> effects.
>
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/