similar to: weights in nls

Sorry to report further difficulties with nls and profiling and constraints ... the problem this time (which I didn't check for in my last round of testing) is that the nls profiler doesn't seem to respect constraints that have been set when using the port algorithm. See test code below ... If I can I will try to hack the code, but I will probably start by redefining my function with

Some students of mine are trying to use gnls, the generalized non-linear least squares function within the nlme library, to study evolutionary questions where correlations between traits at the species level are non-independent because of the evolutionary relatedness of the species. Specifically, they're using a non-linear function (log(sexual dimorphism) ~ log(a + b*variation in mating

I found what seems to be a glaring bug in nls when using constraints, but it is so glaring that I'm a bit nervous about having been stupid. I have (1) tried to make sure I'm up to date: platform i486-pc-linux-gnu arch i486 os linux-gnu system i486, linux-gnu status major 2 minor 2.0 year 2005 month 10 day 06 svn rev 35749 (2) skimmed through the SVN

I'm probably missing something very basic here, but: I've written some C code that I load into R dynamically. In the course of this C code, I generate some multinomial random deviates. I initially used the publically available "randlib" library, which also implements its own random number generator and binomial deviates (which are used to generate the multinomial deviates).

In the optim() documentation, the control parameter "maxit" says "There is no other stopping criteria." That should be "are", or "criterion" ... While I'm at it -- poking at the code a little bit, it looks as if the initial simplex is set from the initial point by displacing each parameter value by max(0.1,0.1*pmax(fabs(Bvec))), where Bvec is the

There was some interest in the commands for creating an HTML file of 3D graphics that can be shown with a Java applet. Looking at things I discovered (of course) that I should really clean up quite a few things before releasing it for real. I hope to do some of that this weekend. In the meanwhile, here are a couple of pointers to the Java applet & documentation (apparently free for

I am trying to find a way to analyze a "simple" mixed model with two levels of a treatment, a random blocking factor, and (wait for it) negative binomial count distributions as the response variable. As far as I can tell, the currently available R offerings (glmmGibbs, glmmPQL in MASS, and Jim Lindsey's glmm code) aren't quite up to this. From what I have read (e.g.

I'm trying to answer a question from a student about memory use in R (I won't go into the details right here). I have a really vague memory of having read a document, possibly by Venables or Ripley, discussing the awfulness of memory allocation in S-PLUS, and giving (in the context of a bootstrapping analysis of shoe size data??) some general strategies for conserving memory in S-PLUS.

I'm very excited by the new mle package now incorporated in stats4. If possible, I'd like to help develop it. In the past I wrote a similar package (mleprof, available from http://www.zoo.ufl.edu/bolker/R/src), and would like to see if there's anything that my package does that I could contribute (in particular, I'd like to make sure that the code is as robust as possible in

This is off-topic (apologies), but I thought I might get a lead or two here. I'm interested in generating random deviates from a multivariate distribution which is a generalization of the beta distribution -- the Bayesian canonical distribution for the parameter estimates of a multinomial distribution. Given a vector (length n-1) of probabilities p and a vector (length n) of shape

I'm having problems (or to be precise a student is having problems, which I'm having problems helping her with) trying to use glm.nb() from the MASS package to do some negative binomial fits on a data set that is, admittedly, wildly overdispersed (some zeros and some numbers in the hundreds). glm.nb is failing to converge, and furthermore is (to my surprise) producing values of theta

I want to add my two cents of congratulation to the R core team. I also want to encourage everyone who uses R to be an active, not a passive user -- the fastest way R will get better is if the folks who use it submit bug reports, suggestions, R code for their particular fields, documentation, even patches and code fixes. R is big and complicated enough now that we can't leave testing to

Can we add a note to the documentation that setting par(fig) resets par(mfrow) and par(mfcol) to c(1,1)? Or are mfrow and mfcol now deprecated in favor of all the split screen stuff? (I was spending the morning trying to write some code that plotted multiple subplots within whatever plot region was active at the moment; I was able to set and reset fig successfully, but got very confused as to

There are a few inconsistencies, at least, in some of the functions that generate random deviates from particular distributions (I think they're bugs because they're inconvenient, but maybe someone can make an argument for the current behavior). If people think these are really bugs I can submit them, together or separately. 1. rlnorm(n,mean,sd) gives NaN for sd=0, rather than always

For various reasons, I've opted to make my packages available from my own web page rather than submitting them to CRAN (mostly laziness -- for a long time I didn't have the packages quite cleaned up enough to pass all the tests). It occurred to me to wonder about the scripts used by CRAN maintainers to generate the PACKAGES file, and to generate PACKAGES.html from PACKAGES. Are

Just a quick question, in case I'm doing something really boneheaded that could be easily sorted out. I'm attempting to save() datasets on Linux (R 1.1.1) and load() them on Win2K (rw1011, fetched from CRAN today). I get the "restore file corrupted" message every time. I've tried saving with ascii=TRUE and FALSE, and the ASCII versions look OK (it's my impression

Just thought people might be interested to know that the latest version of R-win now runs under Wine, the windows emulator for Linux (ver. 20001202). (I originally tried this a couple of releases back because a student seemed to be having system-specific problems and I was attempting to be lazy.) The recent switch from crtdll to msvcrt seems to have helped with Wine compatibility. I

Some of my students were interested in a plot where the sizes of the symbols were proportional to the number of repeats in the data (I have them plotting discrete data, so there are often overlaps). Here's the solution I came up with, which seems clever (to me) but not necessarily the best way. Can anyone suggest improvements? sizeplot <- function(x,y,maxsize=0.2,

When using test="F" in stat.anova() / anova.glm(), R uses the assumed dispersion parameter for the specified family (e.g. scale=1 for binomial), while S-PLUS automatically uses the estimated dispersion parameter (residual deviance/residual df). I think there are good reasons for the behavior in R -- it fits with the "you get what you actually asked for" philosophy -- and

I have found what seems to be a bug in warning(), but perhaps I'm being really boneheaded (it's happened before). Essentially, warning() seems to segfault if its argument is greater than 8191 characters (8192 is defined as BUFSIZE in errors.c, so a quick workaround would be to boost this ...) The bug was initially provoked by trying to concatenate two long tables -- the warning message