R help - Oct 2009 - Fast optimizer

Hi,

I'm using optim with box constraints to MLE on about 100 data points.
It goes quite slow even on 4GB machine. I'm wondering if R has any
faster implementation? Also, if I'd like to impose
equality/nonequality constraints on parameters, which package I should
use? Any help would be appreciated. Thank you.

rc

You have hardly given us any information for us to be able to help you.  Give us
more information on your problem, and, if possible, a minimal, self-contained
example of what you are trying to do.

Ravi.
____________________________________________________________________

Ravi Varadhan, Ph.D.
Assistant Professor,
Division of Geriatric Medicine and Gerontology
School of Medicine
Johns Hopkins University

Ph. (410) 502-2619
email: rvaradhan at jhmi.edu


----- Original Message -----
From: R_help Help <rhelpacc at gmail.com>
Date: Thursday, October 29, 2009 8:24 pm
Subject: [R] Fast optimizer
To: r-help at r-project.org, r-sig-finance at stat.math.ethz.ch

> Hi,
> 
> I'm using optim with box constraints to MLE on about 100 data points.
> It goes quite slow even on 4GB machine. I'm wondering if R has any
> faster implementation? Also, if I'd like to impose
> equality/nonequality constraints on parameters, which package I should
> use? Any help would be appreciated. Thank you.
> 
> rc
> 
> ______________________________________________
> R-help at r-project.org mailing list
> 
> PLEASE do read the posting guide 
> and provide commented, minimal, self-contained, reproducible code.

Dear R_Help Help,

The critical questions are 
a) how many parameters do you have 
b) how pathological is the log-likelihood function
c) how good are your initial values and 
d) how efficiently have you coded your objective function?

Of these, the last is most likely the critical one, and the one you should be
looking at first.

(and consider suing your parents for giving you such a name.)

Bill Venables.
________________________________________
From: r-help-bounces at r-project.org [r-help-bounces at r-project.org] On
Behalf Of R_help Help [rhelpacc at gmail.com]
Sent: 30 October 2009 10:22
To: r-help at r-project.org; r-sig-finance at stat.math.ethz.ch
Subject: [R] Fast optimizer

Hi,

I'm using optim with box constraints to MLE on about 100 data points.
It goes quite slow even on 4GB machine. I'm wondering if R has any
faster implementation? Also, if I'd like to impose
equality/nonequality constraints on parameters, which package I should
use? Any help would be appreciated. Thank you.

rc

______________________________________________
R-help at r-project.org 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.

> Date: Fri, 30 Oct 2009 09:29:06 +0100
> From: Christophe Dutang <dutangc at gmail.com>
> Subject: Re: [R] [R-SIG-Finance]  Fast optimizer
> To: R_help Help <rhelpacc at gmail.com>
> Cc: r-help at r-project.org
>> > Ok. I have the following likelihood function.
>> >
>> > L <- p*dpois(x,a)*dpois(y,b+c)+(1-p)*dpois(x,a+c)*dpois(y,b)
>> >
>> > where I have 100 points of (x,y) and parameters c(a,b,c,p) to
>> > estimate. Constraints are:
>> >
>> > 0 < p < 1
>> > a,b,c > 0
>> > c < a
>> > c < b
>> >
>> > I construct a loglikelihood function out of this. First ignoring
the
>> > last two constraints, it takes optim with box constraint about 1-2
min
>> > to estimate this. I have to estimate the MLE on about 200 rolling
>> > windows. This will take very long. Is there any faster
implementation?
> Take a look at the CRAN task view on optimisation, you may find faster  
> algorithms.
> 
There are several new or revised methods in development as well as a new
wrapper optimx() in the r-forge OptimizeR project
http://r-forge.r-project.org/R/?group_id=395

In particular Rvmmin is an all-R implementation of the algorithm at the
heart of optim's BFGS but with bounds and mask (fixed parameter)
constraints. I'm looking into how it can be made more convenient to
hot-start with suspected "good" parameters, which would be likely be
important here.

JN