Welcome, Guest. Please Login
YaBB - Yet another Bulletin Board
 
  HomeHelpSearchLogin  
 
Latin Hypercube Sampling question (Read 2778 times)
rss
YaBB Newbies
*
Offline


Curious HyPhy user

Posts: 6
Latin Hypercube Sampling question
Aug 29th, 2013 at 3:44am
 
I'm trying to perform parameter resampling using Latin Hypercube sampling in HYPHY, and had a couple of questions:

1) Is it possible to perform LHS from the command line in HYPHY?

2) After resampling, it looks like some of my LHS samples have a likelihood value of -infinity.  I thought that the re-samples are chosen with probability somehow proportional to their likelihood, so that impossible parameter combinations would be excluded?  Any idea why some of the re-samples might have a likelihood value of -infinity?   (I chose M << N, so there should have been plenty of initial samples to choose from.

3)Is it correct to turn LHS samples into confidence bounds by taking the interval containing 95% of the resampled parameter estimates?

Thanks!!
Back to top
 
 
IP Logged
 
Sergei
YaBB Administrator
*****
Offline


Datamonkeys are forever...

Posts: 1658
UCSD
Gender: male
Re: Latin Hypercube Sampling question
Reply #1 - Aug 29th, 2013 at 8:11am
 
Hi rss,

1). Yes, you can perform such an analysis. Assuming you have a NEXUS file containing the HyPhy block.
[To export such a file assuming you have a likelihood function lfID (or any other name defined), just use
Code:
Export (lfNEXUS, lfID);
fprintf (file_to_save_to, CLEAR_FILE, lfNEXUS);
 


]

To do so call

$HYPHYMP -p nexus_with_lf_data

Using the '-p' flag will prompt HyPhy to offer you a list of post-processing options, including LHC

2). This is definitely NOT normal (could be a bug, or something wonky with the likelihood function). -Infinity basically means that the likelihood of your data is exactly 0, which should not happen.
Can you send me an example?

3). That is the idea, yes.

Sergei
Back to top
 

Associate Professor
Division of Infectious Diseases
Division of Biomedical Informatics
School of Medicine
University of California San Diego
WWW WWW  
IP Logged