[R] Estimates at each iteration of optim()?

[Ravi Varadhan]

Deepankar,

Here is an example using BFGS:

> fr <- function(x) {   ## Rosenbrock Banana function
+     x1 <- x[1]
+     x2 <- x[2]
+     100 * (x2 - x1 * x1)^2 + (1 - x1)^2
+ }
> grr <- function(x) { ## Gradient of 'fr'
+     x1 <- x[1]
+     x2 <- x[2]
+     c(-400 * x1 * (x2 - x1 * x1) - 2 * (1 - x1),
+        200 *      (x2 - x1 * x1))
+ }
> optim(c(-1.2,1), fr, grr, method = "BFGS", control=list(trace=TRUE))
initial  value 24.200000 
iter  10 value 1.367383
iter  20 value 0.134560
iter  30 value 0.001978
iter  40 value 0.000000
final  value 0.000000 
converged
$par
[1] 1 1

$value
[1] 9.594955e-18

$counts
function gradient 
     110       43 

$convergence
[1] 0

$message
NULL

> 

This example shows that the parameter estimates are printed out every 10
iterations.  However, trying different integer values for trace from 2 to 10
(trace = 1 behaves the same as trace=TRUE) did not change anything.  If you
want to get estimates at every iteration, look at the source code for BFGS
(which I assume is in FORTRAN). You may have to modify the source code and
recompile it yourself to get more detailed trace for BFGS. 

However, you can get parameter iterates at every step for "L-BFGS-B" using
trace=6, although this gives a lot more information than just the parameter
estimates.  Alternatively, you can use the "CG" methods with trace=TRUE or
trace=1, which is a generally a lot slower than BFGS or L-BFGS-B.

Why do you want to look at parameter estimates for each step, anyway?


Ravi.

----------------------------------------------------------------------------
-------

Ravi Varadhan, Ph.D.

Assistant Professor, The Center on Aging and Health

Division of Geriatric Medicine and Gerontology 

Johns Hopkins University

Ph: (410) 502-2619

Fax: (410) 614-9625

Email: rvaradhan at jhmi.edu

Webpage:  http://www.jhsph.edu/agingandhealth/People/Faculty/Varadhan.html

 

----------------------------------------------------------------------------
--------

-----Original Message-----
From: r-help-bounces at stat.math.ethz.ch
[mailto:r-help-bounces at stat.math.ethz.ch] On Behalf Of DEEPANKAR BASU
Sent: Monday, April 23, 2007 11:34 AM
To: Peter Dalgaard
Cc: r-help at stat.math.ethz.ch
Subject: Re: [R] Estimates at each iteration of optim()?

I read the description of the trace control parameter in ?optim and then
also looked at the examples given at the end. In one of the examples I found
that they had used "trace=TRUE"  with the method "SANN". I am using the
method "BFGS" and I tried using "trace=TRUE" too but I did not get the
parameter estimates at each iteration. As you say, it might be method
dependent. I tried reading the source code for "optim" but could not find
out what I was looking for. Hence, I was wondering if anyone could tell me
what option to use with the method "BFGS" to get the parameter estimates at
each iteration of the optimization.

Deepankar


----- Original Message -----
From: Peter Dalgaard <p.dalgaard at biostat.ku.dk>
Date: Monday, April 23, 2007 2:46 am
Subject: Re: [R] Estimates at each iteration of optim()?

> DEEPANKAR BASU wrote:
> > I am trying to maximise a complicated loglikelihood function with 
> the "optim" command. Is there some way to get to know the estiamtes 
> at each iteration? When I put "control=list(trace=TRUE)" as an 
> option in "optim", I just got the initial and final values of the 
> loglikelihood, number of iterations and whether the routine has 
> converged or not. I need to know the estimate values at each 
> iteration.>
> >   
> It might help if you actually _read_ the description of the trace 
> control parameter (hint: it is not an on/off switch) in ?optim... 
> And, 
> as it says, this is method dependent, so you may have to study the 
> source code.
> 
> > Deepankar
> >
> > ______________________________________________
> > R-help at stat.math.ethz.ch mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-help
> > PLEASE do read the posting guide http://www.R-project.org/posting-
> guide.html> and provide commented, minimal, self-contained, 
> reproducible code.
> >   
> 
>

______________________________________________
R-help at stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.