search for: abstol

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. [[alternative HTML version deleted]]

Hello All, I would like to know what fit criterion (abstol arg) is in nnet. Is it the threshold for the difference btw the max output and target values? Is the value at each iteration also the difference btw max of output and target values over all output units (case of multiple classes)? How could value displayed at each iteration be related to SSE and...

...RR)) theta.0 = numeric(3*K) for(i in 1:K) {theta.0[(1+3*(i-1)):(3*i)] = rep(1/10,3)+i/100} # data1 gives an error but data2 does not! constrOptim(theta=theta.0, f=likelihood, grad=score, ui=RR,ci=zeros, data = data1, method="BFGS", control=list(fnscale=-1,abstol=1e-16,reltol=1e-16))$par constrOptim(theta=theta.0, f=likelihood, grad=score, ui=RR,ci=zeros, data = data2, method="BFGS", control=list(fnscale=-1,abstol=1e-16,reltol=1e-16))$par # If you want to further check run the simulation below. K = 3 N = rep(20,...

...le session: nmminDemo.c: #include <R_ext/Applic.h> #include <stdio.h> double parabola(int n, double *par, void *ex) { double xm = par[0] - 1; return xm * xm + 13; } int main() { double initial[1] = {1.5}; double result[1]; double value; int convergenceCode; const double abstol = 1e-16; const double reltol = 1e-8; const double alpha = 1.0; /* reflection factor */ const double beta = 0.5; /* contraction factor */ const double gamm = 2.0; /* expansion factor */ const int trace = 4; /* tracing on */ int fncount; const int maxit = 10000; nmmin(1, initial, resu...

...apack routine 'dsyevr'? Especially I am interested about the meaning and handling of error codes 1 and 2. In Lapack.c I only see the reference to the variable info in certain Fortran code: F77_CALL(dsyevr)(jobv, range, uplo, &n, rx, &n, &vl, &vu, &il, &iu, &abstol, &m, rvalues, rz, &n, isuppz, &tmp, &lwork, &itmp, &liwork, &info); if (info != 0) error(_("error code %d from Lapack routine '%s'"), info, "dsyevr"); lwork = (int) tmp; liwork = itmp; work = (double *) R_alloc(lwork,...

...int n = mydata->n; double mu = par[0]; double sum = 0.0; int i; for (i=0; i<n; i++) { double t = X[i] - mu; sum += t*t; } return sum; } int main(void) { int i,n=1, trace=1, maxit=100, fail, fncount; double *par, abstol=0.0001, intol=.0001, alpha=1, beta=.5, gamma=2, *xin, *x, Fmin; MyData ex; par=malloc(n*sizeof(double)); xin=malloc(n*sizeof(double)); x=malloc(n*sizeof(double)); par[0]=5; xin[0]=5; ex.n=5; ex.X=malloc(ex.n*sizeof(double)); for(i=0;i<ex.n;i++) ex.X[i]=i; nmmin(n,xin...

...) { char *argv[]= {"nmminDemo", "--gui=none", "--silent"}; const int argc = 3; double initial[1] = {1.5}; double result[1]; double value; int convergenceCode; /* * The following values are based on the help * page for optim. */ const double abstol = 1e-16; const double reltol = 1e-8; const double alpha = 1.0; /* reflection factor */ const double beta = 0.5; /* contraction factor */ const double gamm = 2.0; /* expansion factor */ const int trace = 0; /* tracing on */ int fncount; const int maxit = 10000; Rf_initEmbeddedR(argc,...

Dear all, I'd like tweaking the ... arguments that one user can pass in my function for fitting a model. More precisely, my objective function is (really) problematic to optimize using the "optim" function. Consequently, I'd like to add in the "control" argument of the latter function a "ndeps = rep(something, #par)" and/or "parscale =

...ion in SPSS software. The input data, model and parameters are below: choles <- c(94, 158, 133, 164, 162, 182, 140, 157, 146, 182); sbp <- c(105, 121, 128, 149, 132, 103, 97, 128, 114, 129); case <- c(1, 3, 3, 2, 1, 2, 3, 1, 2, 2); result <- multinom(case ~ choles + sbp + choles:sbp, abstol=1.0e-20, reltol=1.0e-20, MaxNWts=10000); However, the estimated coeffcients and standard errors of the coefficeints are different from the SPSS. For instance, the estimated coefficients of the variable "choles" are 0.1946555 and 0.6244513 from the above result, but the SPSS result are...

...ork 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 set against the actual values using: pmat<- plot(y[1:400]) lines(c$fitted.values, col="blue", lwd=2) It seems as though neural net is not learning the negative values. I have tried running nnet several times, but each time I ge...

...'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 parameter. However, if I set abstol to 0, it still doesn't force the network to run for the maximum specified epochs. From the description in the documentation, setting it to 0 should achieve the effect I want. :( I can't seem to find a way to prevent early stopping. Any help would...

...,0)<<endl; //cout << qnorm(0.7, 0.0, 1.0, 0, 0) << endl; int n = 5; double *dpar, *opar; int fncount, fail; dpar = new double [n]; opar = new double [n]; dpar[0] = 13.67318; dpar[1] = 0; dpar[2] = 17.02707; dpar[3] = 0; dpar[4] = 24.08231; double value; double abstol = 1e-16; double intol = 1e-8; int y = 12; MM t = (MM)malloc(y*(n+2)*sizeof(TT)); t-> tempX = new double * [y]; for(size_t ii = 0; ii < y; ii++) { t->tempX[ii] = new double [n]; } t->tempX[0][1] = -0.03; t->tempX[1][1] = -0.02; t->tempX[2][1] = -0.015; t->tempX[3]...

#optim package estimate<-optim(init.par,Linn,hessian=TRUE, method=c("L-BFGS-B"),control = list(trace=1,abstol=0.001),lower=c(0,0,0,0,-Inf,-Inf,-Inf,-Inf,-Inf,-Inf,-Inf,-Inf,-Inf),upper=c(1,1,1,1,Inf,Inf,Inf,Inf,Inf,Inf,Inf,Inf,Inf)) #nlminb package estimate<-nlminb(init.par,Linn,gr=NULL,hessian=TRUE,control = list(trace=1,factr=1),lower=c(0,0,0,0,-Inf,-Inf,-Inf,-Inf,-Inf,-Inf,-Inf,-Inf,-Inf),upper=c(1,...

...e in my C functions the error might be. I give the output below. Can this be an R bug? I suspect it has to do with repeated calls to 'vmmin' like this: for (...){ vmax = vmaxget(); vmmin(*p, b, &Fmin, bfun, bfun_gr, *maxit, *trace, mask, abstol, reltol, nREPORT, ext, &fncount, &grcount, &fail); vmaxset(vmax); ... } but I am only guessing. Any suggestions? Ubuntu 8.04, R-2.7.2. G?ran ------------------------------------------------------------------- > glmmboot(y ~ x, cluster = clu...

...Core 2 Linux. I tested a proportional odds logistic regression with MASS's polr and Design's lrm. Parameter estimates between the 2 are consistent, but the standard errors are quite different, and the conclusions from the t and Wald tests are dramatically different. I cranked the "abstol" argument up quite a bit in the polr method and it did not make the differences go away. So 1. Can you help me see why the std. errors in the polr are so much smaller, and 2. Can I hear more opinions on the question of t vs. Wald in making these signif tests. So far, I understand the t...

Like kile for LaTeX, Linux/KDE's kate editor is an excellent editor for R, with easy code submission to a running R process. Syntax highlighting is good. I have not been able to figure out two things: - how to automatically reformat a line or region of text using good indentation rules (Emacs/ESS make this so easy by just hitting Tab while the cursor is in a line, or highlighting a

Hi all~ I am wondering if it is possible to alter the stopping precision for parameters estimated using the 'optim'? If it helps, I am minimizing the log-likelihood of a function using constraints (i.e. L-BFG-S). -Jeremy

Dear list, I have a question related to the correct interpretation of the relative convergence criterion used by 'optim'. In the help of the function is it written that: "reltol:Relative convergence tolerance. The algorithm stops if it is unable to reduce the value by a factor of reltol * (abs(val) + reltol) at a step." and I was wondering if the previous criterion is

Hello all, Ive been playing with nnet (package 'nnet') and Ive come across this problem. nnet doesnt seems to like to have more than 1000 weights. If I do: > data(iris) > names(iris)[5] <- "species" > net <- nnet(species ~ ., data=iris, size=124, maxit=10) # weights: 995 initial value 309.342009 iter 10 value 21.668435 final value 21.668435 stopped after 10

HI sorry to be a nuisance to all!!! how can i see the code of a particular function? e.g. nnet just as an example