Re: [R] Likelihood returning inf values to optim(L-BFGS-B) other

From: <tfjbl_at_mail.uas.alaska.edu>
Date: Fri 06 Apr 2007 - 14:47:48 GMT


Hello,
A couple of ideas...

Im not clear on your whole problem however...

Consider making use of the lgamma function, which returns the natural log of the gamma function. This may help. The gamma function gets awfully, big very fast.

Also multivariable likelihoods can be bumpy like a mountain range, with minor peaks and valleys. It is possible that your likelihood has such a shape. Maybe each iteration Xn is trying to get closer to the main peak, but instead goes up the ridge of a valley and gets lost, ultimately reaching a boundary of the region.

You could try starting at a variety of locations. Possibly many hundreds of starting points, randomly selected from within your region. Then examine the ending point for each starting point.

If you do have a bumpy likelihood surface you might have to start very close to the actual maximum to get there. Surface plots might help, setting some variables to a constant. I know in 4D this will be tough.

Here is a useful reference that helped me recently with a similar maximization problem:

"Computational Statistics"
by Geof H. Givens and Jennifer A. Hoeting

They have R-code examples here:
http://www.stat.colostate.edu/computationalstatistics/

Good luck!

Joe Liddle
University of Alaska Southeast



R-help@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. Received on Sat Apr 07 01:52:04 2007

Archive maintained by Robert King, hosted by the discipline of statistics at the University of Newcastle, Australia.
Archive generated by hypermail 2.1.8, at Fri 06 Apr 2007 - 16:30:58 GMT.

Mailing list information is available at https://stat.ethz.ch/mailman/listinfo/r-help. Please read the posting guide before posting to the list.