Re: st: [Mata newbie] optimize() vs -ml-

[Antoine Terracol <Antoine.Terracol@univ-paris1.fr>]

Jeff,

again, thanks for your answer (I did post a reply to your answer a few 
hours ago, but it hasn't appearedon the list yet).

Playing with your code, I noticed that -optimize- is indeed faster than 
-ml- for small sample sizes, but that the reverse seems to be true for 
larger N

I ran a few simulations (-simulate-, 100 replications) with varying N.

Below is the results using -v2- and -d2- (the same pattern can be found 
with v1/d1 and v0/d0):

N=100
    Variable |       Obs        Mean    Std. Dev.       Min        Max
-------------+--------------------------------------------------------
    optimize |       100      .00423    .0069947          0       .016
          ml |       100      .05278    .0082298       .046       .078

N=1000
    Variable |       Obs        Mean    Std. Dev.       Min        Max
-------------+--------------------------------------------------------
    optimize |       100      .03333    .0061727       .015       .062
          ml |       100      .07604    .0079085       .062       .094

N=5000
    Variable |       Obs        Mean    Std. Dev.       Min        Max
-------------+--------------------------------------------------------
    optimize |       100      .16598     .015475        .14       .218
          ml |       100      .18982     .014276       .171        .25

N=10000
    Variable |       Obs        Mean    Std. Dev.       Min        Max
-------------+--------------------------------------------------------
    optimize |       100      .33812    .0282104       .296       .422
          ml |       100      .33324    .0244937       .296       .406

N=50000
    Variable |       Obs        Mean    Std. Dev.       Min        Max
-------------+--------------------------------------------------------
    optimize |       100     2.00641    .2400445      1.625      2.875
          ml |       100     1.73035    .2316291      1.406      2.719

Antoine


Jeff Pitblado, StataCorp LP wrote:


> If Antoine can derive formulas for producing the gradient for an -ml- type
> -d1- evaluator and the parameter level scores for an -optimize()- type -v1-
> evaluator, a speed comparison will show that -ml- with -d1- and -optimize()-
> with -v1- will be faster than -lf- and -v0-, respectively, and -optimize()-
> will tend to be faster than -ml-.
--
Ce message a ete verifie par MailScanner
pour des virus ou des polluriels et rien de
suspect n'a ete trouve.

*
*   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/