ArchiveOrangemail archive

R special interest group on Mixed Effect Models, notably lmer() related


r-sig-mixed-models.r-project.org
(List home) (Recent threads) (20 other R Project lists)

Subscription Options

  • RSS or Atom: Read-only subscription using a browser or aggregator. This is the recommended way if you don't need to send messages to the list. You can learn more about feed syndication and clients here.
  • Conventional: All messages are delivered to your mail address, and you can reply. To subscribe, send an email to the list's subscribe address with "subscribe" in the subject line, or visit the list's homepage here.
  • Moderate traffic list: up to 30 messages per day
  • This list contains about 12,508 messages, beginning Jan 2007
  • 4 messages added yesterday
Report the Spam
This button sends a spam report to the moderator. Please use it sparingly. For other removal requests, read this.
Are you sure? yes no

Hamiltonian Monte Carlo and Reimann manifold Langevin Monte Carlo and Mixed Effects

Ad
Dominick Samperi 1305292095Fri, 13 May 2011 13:08:15 +0000 (UTC)
Has anyone considered the possible advantages of Hamiltonian Monte
Carlo (http://www.cs.toronto.edu/~radford/) or Reimann manifold Langevin
Monte Carlo (http://www.dcs.gla.ac.uk/inference/rmhmc/) in the context
of mixed modeling?

The latter paper refers to early work by Bates and Watts (1980), and it
appears that the study of geometric aspects of regression goes back
(at least) to the Ph.D. thesis of D.M. Bates, so perhaps readers of this
list can provide some historical perspective.

A key advantage of HMC is that it can prevent getting stuck executing
a Metropolis random walk by taking more directed steps that have a
high probability of acceptance. The downside is that a lot of tuning is
required, and the new Reimann manifold method is advertised to
address this issue by using the intrinsic geometry to automate
some of this manual fine tuning (at the considerable expense of
computing Hessians).

But it is never really established why an intrinsic geometric
approach is practically relevant (and worth the
expense). A similar concern was expressed by some of the
comments following Bates and Watts (1980): to a geometer
a squashed beer can is the same as a full one, and a donut
is the same as a coffee cup.

It may be very helpful conceptually to understand the
underlying intrinsic geometric structure, but if this is all
that mattered there would be no advantage to eigenvalue
decompositions, QR transformations, and other change
of variables that obviously have very practical consequences.

Thus another question is what is the current applied view
on the geometric ideas that were discussed in
Bates and Watts (1980)? The same journal issue included
a number of papers on the geometry of statistical
inference including one by Efron.

My (possibly incorrect) impression is that well-designed
adaptive coordinate transformations have overshadowed
the use of intrinsic methods.

Finally, it seems like there is a link between HMC and
mixed modeling because the introduction of the momenum
variables looks a lot like the introduction of random effects.

Thanks,
Dominick
dave fournier 1305406648Sat, 14 May 2011 20:57:28 +0000 (UTC)
I implemented hybrid mcmc (Hamiltonian) in AD Model Builder's random 
effects module.
So far as I know it has not been used much.  It should be possible to 
access it for testing from
Ben Bolker's glmmadmb package for R without too much trouble.
Dominick Samperi 1305513862Mon, 16 May 2011 02:44:22 +0000 (UTC)
On Sat, May 14, 2011 at 1:12 PM, dave fournier  wrote:
> I implemented hybrid mcmc (Hamiltonian) in AD Model Builder's random effects
> module.
> So far as I know it has not been used much.  It should be possible to access
> it for testing from
> Ben Bolker's glmmadmb package for R without too much trouble.Thanks Dave,

I just installed glmmADMB from R-Forge (Version 0.5-2) but found
no information about Hybrid or Hamiltonian MCMC there.

Dominick
dave fournier 1305548198Mon, 16 May 2011 12:16:38 +0000 (UTC)
On 11-05-15 07:43 PM, Dominick Samperi wrote:

Yes well that is more difficult. Now that we are open source you are 
dealing with a committee.
I do the ADMB side. Ben and others do the R side (I guess). I have been 
trying to get the R
side people to make the R scripts more flexible so that command line 
options can be passed to
the ADMB part. You can request this either here or perhaps better so as 
not to clutter up the
R list with unwelcome ADMB stuff on the ADMB list.> On Sat, May 14, 2011 at 1:12 PM, dave fournier<davef@otter-rsch.com>  wrote:
>> I implemented hybrid mcmc (Hamiltonian) in AD Model Builder's random effects
>> module.
>> So far as I know it has not been used much.  It should be possible to access
>> it for testing from
>> Ben Bolker's glmmadmb package for R without too much trouble.
> Thanks Dave,
>
> I just installed glmmADMB from R-Forge (Version 0.5-2) but found
> no information about Hybrid or Hamiltonian MCMC there.
>
> Dominick
>
Home | About | Privacy