[R] Non-finite finite difference error

[Ravi Varadhan]

Hi Jason,

The error message indicates that there was problem in estimating the gradient of objective function.  It has nothing to do with your second data point.  This could happen for a variety of reasons, but the most proximate cause of the problem seems due to a parameter being negative during the iteration.  The best solution, most often, is to provide a better (or atleast a different) starting value.  Look at the example in the help page for "fitdistr":

fitdistr(x, dgamma, list(shape = 1, rate = 0.1), lower = 0.01)

Note that the above command specifies lower bounds on both the shape and the rate parameter (hence a different optimziation algorithm will be used in "optim").

Best,
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: JS.Augustyn at gmail.com
Date: Tuesday, December 23, 2008 11:27 pm
Subject: [R] Non-finite finite difference error
To: r-help at r-project.org


> Hello, I'm trying to use fitdistr() from the MASS package to fit a 
> gamma  
>  distribution to a set of data. The data set is too large (1167 
> values) to  
>  reproduce in an email, but the summary statistics are:
>  
>  Min. 1st Qu. Median Mean 3rd Qu. Max.
>  116.7 266.7 666.7 1348.0 1642.0 16720.0
>  
>  The call I'm trying to make is:
>  fitdistr(x,"gamma")
>  
>  and the error is:
>  Error in optim(x = c(3466.676842, 1666.749002, 2500.067852, 
> 1200.053892, :
>  non-finite finite-difference value [2]
>  In addition: Warning message:
>  In dgamma(x, shape, scale, log) : NaNs produced
>  
>  I found a couple of other posts from folks who were getting the same 
> error  
>  from optim(), but did not find any useful tips for my situation. The 
> error  
>  seems to indicate a problem with value 2 in my data set 
> (1666.749002), but  
>  nothing seems odd about that value.
>  
>  I'm willing to pass along the full data set as an attachment if it 
> would  
>  help.
>  
>  Thank you in advance!
>  
>  Jason S. Augustyn, Ph.D.
>  
>  	[[alternative HTML version deleted]]
>  
>  ______________________________________________
>  R-help at r-project.org mailing list
>  
>  PLEASE do read the posting guide 
>  and provide commented, minimal, self-contained, reproducible code.