Re: [R] Kmeans cluster analysis

An idea for an interactive approach is to create a distance matrix with 1 or 0 according as pairs are on the same or different clusters, input that to hclust, plot, and run identify.hclust. regards, Farrar Monica Pisica [EMAIL PROTECTED] wrote: Hi Nataniel, As far as i know

Re: [R] How do I back transforme ordinary log-krigged prdiction values?

Zia, I'm not an expert on Kriging, or on the particular package. It sounds like you just want to know how to anti-log the predictions,. which shouldn't be hard. However, depending on what you actually need to predict, it seems your predictions might be biased, right? regards,

Re: [R] Minimum valid number of observations for rpart

Regarding interpretable output, I assume you have looked at the mds plot? regards, Farrar [EMAIL PROTECTED] wrote: Hi, I wonder if anyone knows a study dealing with the minimum valid number of observations when using CART?. On top of that, when using RandomForest, is it

Re: [R] Stepwise Logistic Regression

Stella, An obvious answer is ?step However, I'm having a bit of a problem with it, lately. I got it wording with backwards selection, then it didn't work when I changed the direction from backward to both (backwards and forwards). I would like to know whatever you find that works.

Re: [R] question regression trees

Have a look at the function arguments. I think the function may not split a set of fewer than 20 objects, which is a default setting that can be changed. In addition to rpart, you might want to look at tree. regards, Farrar Wensui Liu [EMAIL PROTECTED] wrote: with seeing more

[R] rpart with overdispersed count data?

I would like to do recursive partitioning when the response is a count variable subject to overdispersion, using say negative binomial likelihood or something like quasipoisson in glm. Would appreciate any thoughts on how to go about this (theory/computation). If I understand the rpart

[R] monitor a simulation with a special console box?

I like to monitor simulation by reporting some current values to the console, every 25th iteration say. I think it might be nice to have that appear in a separate window. Anyone know how? regards, David Farrar New River Analytic [[alternative HTML version

Re: [R] Bayesian data mining

that package. You might have a look at the bayesm package. For that, you can find lots of examples in a text on Bayes in marketing (Rossi et al.) David Farrar New River Analytic 540-818-7373 Frank Grex [EMAIL PROTECTED] wrote: Hi, I need a help to know whether I can perform

Re: [R] any way to make the code more efficient ?

Something like this may be reasonably efficient. I create a length-500 list of 1400*4 matrices of uniform random numbers. Farrar numArrays - 500 # number of arrays arrDim1 - 1400 # array num. rows arrDim2 - 4 # array num. cols arrList - list(numArrays,

Re: [R] test of spatial dependence?? - ask an ecologist?

at Legendre Legendre text book. HTH, Miltinho Brazil David Farrar escreveu: In addition to the 25 numbers, I assume you have coordinates of each field. Otherwise, I don't understand what you are trying to do. I think ecologists like to use a test due to Mantel in this situation

Re: [R] stat question - not R question so ignore if not interested

I seem to remember a term lurking variable. Bert Gunter [EMAIL PROTECTED] wrote: ... But of course this is always the question underlying all empirical -- or maybe even scientific -- analysis: is there some other perhaps more fundamental variable out there that I'm missing that would

Re: [R] test of spatial dependence?? - ask an ecologist?

book. HTH, Miltinho Brazil David Farrar [EMAIL PROTECTED] escreveu: In addition to the 25 numbers, I assume you have coordinates of each field. Otherwise, I don't understand what you are trying to do. I think ecologists like to use a test due to Mantel in this situation

Re: [R] test of spatial dependence?? - ask an ecologist?

In addition to the 25 numbers, I assume you have coordinates of each field. Otherwise, I don't understand what you are trying to do. I think ecologists like to use a test due to Mantel in this situation. The prefix auto means self, of course, the idea being that measurements

Re: [R] Making a case for using R in Academia

I am just joining this thread. Regarding a tendency of journals to lock out the use of particular packages, there are rumours that SAS proc mixed has to be used for particular things. I wonder if whether R might displace SAS or proc mixed in such a role could depend on wether there the QA

[R] maps. display hierararchical cluster analysis results on a map?

All, I have done a cluster analysis analysis of some spatial locations, based on variables other than lat-long, and want to show the results on a map to see if my clusters have some geographic meaning. For kmeans (etc.) I would just use different symbols (etc.) to distinguish

Re: [R] [Fwd: Re: Organisation of medium/large projects with multiple analyses]

It's good to see this sort of thing discussed. For my current approach, I keep a fairly static directory for function libraries, another one for large data sets, and others for projects. I try to define tasks (probably like your analyses) within projects. There is a project

[R] identify.hclust() not working for me

V2.4.0. hc Call: hclust(d = dist(mtx2, method = manh), method = average) Cluster method : average Distance : manhattan Number of objects: 9 plot(hc) identify(hc) Error in cutree(x, k = 2:MAXCLUSTER) : elements of 'k' must be between 1 and 9 David Farrar, Ph.D