similar to: fastest R platform: follow-up and summary

Hello, everyone! I picked up R several months ago and have adopted it as my choice for statistical programming. Coming from a Java background, I can honestly say that R is not only free, it is better tha S-plus: the lexical scope in R makes it very simple to simulate Java's object model. For this, I encourage everyone to read the artcle: Robert Gentleman and Ross Ihaka (2000) "Lexical

Let A be a m by n matrix and b a length n vector. What is the fastest vectorized code for doing for(j in 1:n) A[, j] <- A[, j]/b[j] ? solution 1: t(t(A)/b) solution 2: B <- matrix( rep(b, m), byrow=T, nrow=m ) A/B anything else? I have a program that uses this kind of operation million of times and I appreciate your input. Thanks. Jason Liao ===== Jason G. Liao Department of

I have a program which needs to compute squared Euclidean distance between two vectors million of times, which the Rprof shows is the bottleneck. I wondered if there is any faster way than my own simple function distance2 = function(x1, x2) { temp = x1-x2 sum(temp*temp) } I have searched the R-help archives and can not find anything except when the arguments are matrices. Thanks for any

Hi Jason, I suppose you installed the Matrix library, and it is working on your computer? If yes, may be det.Matrix() was removed, or renamed in the Matrix library you have (I cannot check this for the latest version, because I am away of the office until August 1st), but I will do that next week. In the meantime, you can replace 'det.Matrix' by 'det.default', and it should run.

For those who are interested, I have made available a R function for noncentral hypergeometric distribution at http://www.geocities.com/jg_liao/software/Hypergeometric/hypergeometric_in_R.txt The paper that describes the algorithm will appear in The American Statistician. The function does not run on S-plus as the R's scoping rule is used. Here is how the function can be used: > n1

without having to check the vector element by element? Thanks a lot! Jason ===== Jason G. Liao, Ph.D. Division of Biometrics University of Medicine and Dentistry of New Jersey 335 George Street, Suite 2200 New Brunswick, NJ 08903-2688 phone (732) 235-8611, fax (732) 235-9777 http://www.geocities.com/jg_liao

1. allows underscore as part of a variable name 2. Uses C or Java style comments mark. Jason ===== Jason G. Liao, Ph.D. Division of Biometrics University of Medicine and Dentistry of New Jersey 335 George Street, Suite 2200 New Brunswick, NJ 08903-2688 phone (732) 235-8611, fax (732) 235-9777 http://www.geocities.com/jg_liao __________________________________________________ HotJobs - Search

Hello, everyone! I guess no one is still using R on Sun Sparc these days. But our department has a (pretty new) two-CPU Sun server. We recently compiled R as a 64 bit application and expected a performance boost. But it runs 25-30% slower than the 32 bit version of R. Anyone knows why this is so? Thanks! Jason ===== Jason Liao, http://www.geocities.com/jg_liao Dept. of Biostatistics,

I have had the same problem and I wrote this function rmulti <- function(n, size, p) { NrDim <- length(p) if(NrDim<2) stop("The simulated variabel has to be at least 2-dimensional") res <- matrix(data=NA, nrow=n, ncol=NrDim) p <- p/sum(p) TempSize <- size for(i in 1:NrDim) { TempP <- p[i]/sum(p[i:NrDim]) TempBin <- rbinom(n=n, size=TempSize,

I am excited to learn that Prof. Tierney is bringing to us compiled R. I would like to learn when it will be available. This information will be useful in scheduling some of my projects. Thanks. Jason Jason Liao, http://www.geocities.com/jg_liao Dept. of Biostatistics, http://www2.umdnj.edu/bmtrxweb University of Medicine and Dentistry of New Jersey 683 Hoes Lane West, Piscataway‚ NJ 08854

First, happy holidays, everyone! Thanks to the R team for bringing out 1.4 before the new year. I have 10000 integer triplets stored in A[1:10000, 1:3]. I would like to find the unique triplets among the 10000 ones with possible duplications. What is the easiest way for this. I know the function unique(), which apply to a vector, not the 10000*3 array in my problem. Thanks in advance. Jason

My coauthor made a new R package GeneLogit (100% R code) which installs ok on R 2.3 on Windows. But when I type library(GeneLogit) it gave the error message Error in library(GeneLogit) : 'GeneLogit' is not a valid package -- installed < 2.0.0? It runs on R 1.9.0 just fine. It seems that others have encountered same problem but no solution is found by googling How can I fix this

I have been comparing R with other languages and systems. One peculiar feature of R is there is no scalar. Instead, it is just a vector of length one. I wondered how much performance penalty this deign cause, particular in situations with many scalars in a program. Thanks. Jason Liao, http://www.geocities.com/jg_liao Associate Professor of Biostatistics Drexel University School of Public

This is some kind of follow-up to my previous posts. I have further improved the speed of my program 6 times by eliminating all the apply(). It turns out that apply is slow, is slower than direct loop, it is an order slower than a matrix operation alternative. Here is one example. The first apply version runs 19 seconds, the second loop version runs 13 seconds, the third matrix version runs 1

I did some serach around. It seems that ROC curve computation is not supported on R. Anyone has some leads? Thanks. Jason ===== Jason G. Liao, Ph.D. Division of Biometrics UMDNJ School of Public Health 335 George Street, Suite 2200 New Brunswick, NJ 08903-2688 phone (732) 235-9748, fax (732) 235-9777 http://www.geocities.com/jg_liao __________________________________________________ Great

I have received a few private emails asking for some simple demonstration of calling Java code from R using the rJava package (which can be installed directly inside R). Here is one example for convolving two vectors (an example in the R manual about linking C with R). First write a Java program public class my_convolve { public static double[] convolve(double[] a, double[] b)

Good evening! I have a multi-dimensional minimization problem whose gradient is pretty hard to code. I tried Nelder and Mead method implemented in function optim and it does not work well. I also tried the quasi Newton method in optim using difference as approximate derivative. It does not work well either. I just went through Numerical Recipes book. The book discusses another method without

I did several simulation sessions and the result turned out to be a surprise. After some investigation, I found that different R sessions of the program used the same seed. Simply, in R210, if I start R and type rnorm(1), I always get the same random number. This is contradictary to what is in the R document Initially, there is no seed; a new one is created from the current time when one

One of the real advances (in my humble oppinion of course) of 2003 is the Very Fast Multivariate Kernel Density Estimation algorithm by Alex Gray which achieves several order of speed improvement by using Computational Geometry to organize the data. The algorithm is now implemented in C++ with Mathlab interface by Alexander Ihler of MIT: http://ssg.mit.edu/~ihler/code/kde.shtml I wondered if a

I tried to modify the R function bw.SJ (from the stats package) for my own use. But if I just get the source code and run directly (without any modification), it can no longer find some key functions called within it. I understand this has something to do with searching path which I do not understand well. Can anyone tell me how to modify the source code of an R function and still make it part of