[R] Monotonic regression

[(Ted Harding)]

On 09-May-05 Raubertas, Richard wrote:
> The 'pava' function below looks like code that I wrote
> (with all the comments removed).  I have posted it two or
> three times over the years to the S/R lists.

If you recognise the code, then it is likely that it is what
I came across abut 3 years ago when I was browsing arond
for methods foer isotonic regression. At the time I was still
mainly using 'octave' for much statistics, and I "ported"
it from R to octave (a very simple transcription in this
case).

So thanks for contributing it: I have found it useful many
times, and happily acknowledge your authorship!

> As Martin Maechler has noted, the function 'isoreg' will also
> do monotonic regression (much faster for large data sets).
> However, it does not allow weights (at least as of R 2.0.1).

Nor (looking at the code for 'isoreg' today, now that the
issue has come up) could I see any easy way to incorporate
weights into that code.

> I don't understand the original poster's comment about
> "local minimum".  Isotonic regression is a convex optimization
> problem and 'pava' or 'isoreg' will always produce the 
> unique solution.

Agreed. Nor did I!

Best wishes,
Ted.


--------------------------------------------------------------------
E-Mail: (Ted Harding) <Ted.Harding at nessie.mcc.ac.uk>
Fax-to-email: +44 (0)870 094 0861
Date: 09-May-05                                       Time: 19:32:46
------------------------------ XFMail ------------------------------