[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

# st: RE: Command for doing numerical optimization with a generic loss function?

 From "Nick Cox" To Subject st: RE: Command for doing numerical optimization with a generic loss function? Date Sun, 27 Nov 2005 17:29:20 -0000

```In addition to other answers,

STB-38  sg71  . . . . . . . . . . . . . . . .  Routines to maximize a function
(help amoeba if installed)  . . . . . . . . . . . . . . . . C. Ferrall
7/97    pp.22--26; STB Reprints Vol 7, pp.176--181
routines to maximize a multi-dimensional function

STB-32  sg56  . . . . Implementation of a simplex-based maximization algorithm
(help simplex if installed) . . . . . . . . . . . . . . . . .  T. Barr
7/96    pp.23--27; STB Reprints Vol 6, pp.134--140
likelihood-maximization routine based on the simplex method,
also known as the Amoeba method

Nick
n.j.cox@durham.ac.uk

Pablo Mitnik

> I need to do numerical optimization for a function that is not a
> likelihood function but a (rather complex) loss function. Given that
> stata has the command ml, and that more in general it uses
> algorithms of
> numerical optimization in many  other commands as well, I guess that I
> should be able to find the vector of parameters that minimize my loss
> function by simply writing an appropriate evaluator and using an
> existing command to call it. But I haven't been able to
> figure out what
> command/s to use. I thought of "tricking" ml (d0) by giving
> it minus my
> loss function as if it were log L, but the way in which (in my limited
> understanding) ml works, with the thetas defined as linear
> functions of
> the parameters, seems to preclude this alternative (my loss function,
> which would produce the scalar lnf, is not linear on the parameters).
>

*
*   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/
```

 © Copyright 1996–2015 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index