search for: davidkatzconsult

...e the problem) 2) Recompile R to get bigger memory capability? (I'll have to cross-post to some R forums too) This will be a challenge for a Linux newbie...like me. 3) Any other suggestions? My goal is to create a bigger neural network than fits in my Windows R version. -- David Katz www.davidkatzconsulting.com 541 482-1137 [[alternative HTML version deleted]]

...the user is willing to specify the design matrix in matrix.csr form. This is often advantageous in large problems to reduce memory requirements. I need some help or a reference that will show how to create the design matrix from data in matrix.csr form. Thanks for any help. -- David Katz www.davidkatzconsulting.com 541 482-1137 [[alternative HTML version deleted]]

Looking for help with a project for the US Navy, requires knowledge of Bayesian Statistics, Bayesian Networks and Survival Analysis. Please respond with CV. Thanks. -- David Katz www.davidkatzconsulting.com [[alternative HTML version deleted]]

What is the most efficient alternative to x[order(x)][1:n] where length(x)>>n? I also need the positions of the mins/maxs perhaps by preserving names. Thanks for any suggestions. -- View this message in context: http://www.nabble.com/Largest-N-Values-Efficiently--tf4788033.html#a13697535 Sent from the R help mailing list archive at Nabble.com.

I was surprised to see this unexpected behavior of subset in a for loop. I looked in subset.data.frame and it seemed to me that both versions should work, since the subset call should be evaluated in the global environment - but perhaps I don't understand environments well enough. Can someone enlighten me? In any case, this is a bit of a gotcha for naive users of subset. input.data <-

Hi, I have spatially autocorrelated data (with a binary response variable and continuous predictor variables). I believe I need to do an autologistic model, does anyone know a method for doing this in R? Many thanks C Bell

Dear R gurus, first let me apologize for a question that might hve been answered before. I was not able to find the solution yet. I want to concatenate two lists of lists at their lowest level. Suppose I have two lists of lists: list.1 <- list("I"=list("A"=c("a", "b", "c"), "B"=c("d", "e", "f")),

Hello I apologise for the length of this entry but please bear with me. In short: I need a way of subsampling communities from all possible communities of n taxa taken 1:n at a time without having to calculate all possible combinations (because this gives me a memory error - using combn() or expand.grid() at least). Does anyone know of a function? Or can you help me edit the combn or

Several searches turned up nothing. Perhaps I will try to implement it if nobody else has. Thanks. -- View this message in context: http://www.nabble.com/RFclustering---is-it-available-in-R--tf4274225.html#a12165636 Sent from the R help mailing list archive at Nabble.com.

The documentation for predict.gam in library mgcv gives an example of using an "lpmatrix" to do approximate prediction via interpolation. However, the code is specific to the example wrt the number of smooth terms, df's for each,etc. (which is entirely appropriate for an example) Has anyone generalized this to directly generate code from a gam object (eg SAS or C code)? I wanted to

This is in regards to the suggested use of type="lpmatrix" in the documentation for mgcv::predict.gam. Could one not get the same result more simply by using type="terms" and interpolating each term directly? What is the advantage of the lpmatrix approach for prediction outside R? Thanks. -- View this message in context:

Sometimes, for specific models, I get this error from predict.gam in library mgcv: Error in complete.cases(object) : negative length vectors are not allowed Here's an example: model.calibrate <- gam(meansalesw ~ s(tscore,bs="cs",k=4), data=toplot, weights=weight, gam.method="perf.magic") > test <- predict(model.calibrate,newdata) Error in

I've been using Rterm with ESS to run R for some time. Recently I've experienced lockups when displaying graphics; the first display seems to work, but then refuses to respond and must be killed with dev.off(). Rgui has no problems. I've tried eliminating all other processes that might cause conflicts, to no avail. I'm using win XP and R 2.9.0. Here's a transcript using rterm:

In Dr. Wood's book on GAM, he suggests in section 4.1.6 that it might be useful to shrink a single smooth by adding S=S+epsilon*I to the penalty matrix S. The context was the need to be able to shrink the term to zero if appropriate. I'd like to do this in order to shrink the coefficients towards zero (irrespective of the penalty for "wiggliness") - but not necessarily all the

Trying to learn Proto. This threw me: #startup r... > > library(proto) > a <- proto(x=10) > a$x [1] 10 > x <- proto(x=100) > x$x Error in get("x", env = x, inherits = TRUE) : invalid 'envir' argument > Do I simply need to be careful to name proto objects and proto components uniquely? Is this the desired behavior for proto objects? Thanks. -- View

The verbose option gives a display like: > rf.500 <- + randomForest(new.x,trn.y,do.trace=20,ntree=100,nodesize=500, + importance=T) | Out-of-bag | Tree | MSE %Var(y) | 20 | 0.9279 100.84 | What is the meaning of %var(y)>100%? I expected that to correspond to a model that was worse than random, but the predictions seem much better than that on