similar to: abstol in nnet

Hi, I am using nnet to learn transfer functions. For each transfer function I can estimate the best possible Mean Squared Error (MSE). So, rather than trying to grind the MSE to 0, I would like to use abstol to stop training once the best MSE is reached. Can anyone confirm that the abstol parameter in the nnet function is the MSE, or is it the Sum-of-Squares (SSE)? Best regards, Sam.

Hi, I am trying to make a neural network learning a "noisy sine wave". Suppose I generate my data like so.. x <- seq(-2*pi, 2*pi, length=500) y <- sin(x) + rnorm(500, sd=sqrt(0.075)) I then train the neural net on the first 400 points using c <- nnet(as.matrix(x[1:400]),as.matrix(y[1:400]), size=3, maxit=10000, abstol=0.075, decay=0.007) Inspecting the fit of the training

Hello there, I'm back again with another question about the neural network package. I'm having trouble getting the network to run for the maximum number of iterations. It always stops early, usually after 100 iterations claiming to have converged at an answer. Now, for my purposes I want it to run for the entire number of epochs, and I'm been looking at modifying the abstol

Hello All, Although I have downloaded some libraries such as multivariate data analysis library (multiv) and ade4, their functions such as pca or reconst are not recognised. Should I install any thing else or use any instruction so that R could find the location of libraries? Thanks for your quick response, Haleh Yasrebi

Hello, Like QR decomposition, I am looking for decomposition to get U and V matrix of SVD decomposition (according to LINPACK, X = UDV'). Do you know if there is a function which could calculate this decomposition? Look forward to your reply, Haleh

Hi, I'm using the R nnet package and have a few simple (?) questions. What is the "value " that is output after every 10 iterations during the training of the network and how is it calculated? # weights: 177 initial value 506.134586 iter 10 value 128.222774 iter 20 value 95.399782 iter 30 value 87.184564 ... Is the "value" the error, if not, is there any way

I am new to R and use nnetTs - calls. If a time series of let's say 80000 Datapoints and did call nnetTs I want make a new net for the old ponts plus the next 1000 points (81000 datapoints total) what would again cost much calculation time. So I want to pre-init the new net with the former wonnen net to reduce the necessary iteration numbers. Is thee a possibility to do that and how? i.e.:

Hi, I just want to point out a discrepancy between the documentation of predict.nnet & the function definition. >?predict.nnet => predict.nnet package:nnet R Documentation Predict New Examples by a Trained Neural Net Description: Predict new examples by a trained neural net. Usage: predict.nnet(object, x,

Hello, Why do I get doubt (NA) in the factor of test classification even if I fix l (minimum vote)? By setting l, no doubt should be occurred. Look forward to your reply Haleh

Dear R people, I am using tune.nnet from e1071 package to tune the parameters for nnet. I am using the following syntax: tuneclass <-c(rep(1,46),rep(2,15)) tunennet <-tune.nnet(x=traindata,y=tuneclass,size=c(50,75,100),decay=c(0,0.005,0.010),MaxNWts = 20000) Here traindata is the training data that I want to tune for nnet which is a matrix with 61 rows(samples) and 200

Hello, I've recently been using the nnet package to do some basic forecast predictions. I've found the package to be quite useful and I am getting some good results. However, I am in the midst of writing a small paper on the results I am getting and wish to clarify some things about the nnet package that are not made clear in the documentation. In particular I would like to know

One possibility is to look at predict.nnet(), and - Write an R function that write out parts of an nnet object that are needed by predict.nnet() to an external file. - Re-write predict.nnet() in C, reading the model information from the external file. Obviously you'll also need the C source for the code that predict.nnet() calls, and modify those as needed to strip out dependency on R, if

Hello All, I am working with samr and nnet packages. I am following the steps given below: 1> I take a input file with signal values for 9506 genes and 36 chips , belonging to two classes. 2> I perform samr analysis on 80% of chip data from both the classes.(selected by random sampling) 3> I then use the data of only the significant genes from this samr analysis to train nnet. 4>

Hi Version 0.7 of package tsDyn presented at useR! 2009 is now on CRAN, extended with several new features. The package tsDyn is aimed at estimating nonlinear time series models which exhibit regime specific properties. The regime switching dynamics can either be described by smooth transition (STAR and LSTAR) or threshold effects (SETAR). The package furthermore offers nonlinear models

Hi Version 0.7 of package tsDyn presented at useR! 2009 is now on CRAN, extended with several new features. The package tsDyn is aimed at estimating nonlinear time series models which exhibit regime specific properties. The regime switching dynamics can either be described by smooth transition (STAR and LSTAR) or threshold effects (SETAR). The package furthermore offers nonlinear models

Hello, I am pretty new to R. I am working on neural network classifiers and I am feeding the nnet input from different regions of interest (fMRI data). The script that I am using is this: library (MASS) heap_lda <- data.frame(as.matrix(t(read.table(file="R_10_5runs_matrix9.txt")))*100000,syll = c(rep("heap",3),rep("hoop",3),rep("hop",3))) library(nnet)

Dear R people, I want to use the tune.nnet function of e1071 package to tune nnet . I am unable to understand the parameters of tune.nnet from the e1071 pdf document. I have performed nnet on a traindata and want to test it for class prediction with a testdata. I want to know the values of size,decay,range etc. parameters for which the prediction of testdata is best. Can anyone please tell me

HI, Dear R community, My data set has 2409 variables, the last one is response variable. I have used the nnet after feature selection and works. But this time, I am using nnet to fit a model without feature selection. I got the following error information: > dim(train) [1] 1827 2409 nnet.fit<-nnet(as.factor(out) ~ ., data=train, size=3, rang=0.3, decay=5e-4, maxit=500) # model

Hello ! I'd like to know to which of the FANN package network corresponds the R nnet network ? In more details, what is the R nnet activation function, what is the training algorithm (rprop, quickprop, ...) ? Also, it seems that the R nnet "decay" parameter in nnet corresponds to the "learning_rate" parameter in FANN. Correct ? Many thanks in advance ! Luc Moulinier

The TITLE for the nnet package is garbled: it comes out as nnet Feed-forward neural networks and multinomial log-linear nnet Feed-forward neural networks and multinomial log-linear models The problem is in maketitle: auk% cat DESCRIPTION Bundle: VR Version: 6.1-9 Date: 2000/07/11 Depends: R (>= 1.1) Author: S original by Venables & Ripley. R port by Brian Ripley