similar to: Interpreting negative diagonal values in a hessian

I've been working on a new package and I have a few questions regarding the behaviour of the nlm function. I've been (for better or worse) using the nlm function to fit a linear model without suppling the hessian or gradient attributes in the objective function. I'm curious as to why the nlm requires 31 iterations (for the linear model), and then it doesn't work when I try to add

Hi R-help! I'm trying to understand how R's nlm function updates its estimate of the Hessian matrix. The Dennis/Schnabel book cited in the references presents a number of different ways to do this, and seems to conclude that the positive-definite secant method (BFGS) works best in practice (p201). However, when I run my code through the optim function with the method as "BFGS",

Dear R experts, In R-intro, under the 'Nonlinear least squares and maximum likelihood models' there are ttwo examples considered how to use 'nlm' function. In 'Least squares' the Standard Errors obtained as follows: After the fitting, out$minimum is the SSE, and out$estimates are the least squares estimates of the parameters. To obtain the approximate standard

Dear all, I am trying to minimize a function with 3 parameters using nlm. I have worked out the 2nd derivatives (incl the cross-product terms) and would like to supply them to nlm for evaluation. What I am not sure about is how to set up the Hessian matrix for nlm. That is, attr(lhat, "hessian") <- c(???) Do I have to enter all 9 of the entries or just the lower triangle of the

Out of curiosity, I have tried, without success, to use the new facility in nlm to specify the gradient and hessian. (It is many years since I had a problem simple enough to make analytic derivation of these worthwhile.) The help now says that the function must have attributes with these names but gives no indication as to what should be in the attributes. The online example and demo do not use

GENERAL REFERENCE ON NONLINEAR OPTIMIZATION? What are your favorite references on nonlinear optimization? I like Bates and Watts (1988) Nonlinear Regression Analysis and Its Applications (Wiley), especially for its key insights regarding parameter effects vs. intrinsic curvature. Before I spent time and money on several of the refences cited on the help pages for "optim",

In the course of revising a paper I have had occasion to attempt to maximize a rather complicated log likelihood using the function nlm(). This is at the demand of a referee who claims that this will work better than my proposed use of a home- grown implementation of the Levenberg-Marquardt algorithm. I have run into serious hiccups in attempting to apply nlm(). If I provide gradient and

Hello RUser! I try to use ace for an ancestral state reconstruction but got back an error message. ace(FacVar,Tree, type="discrete") Warning messages: 1: In nlm(function(p) dev(p), p = rep(ip, length.out = np), hessian = TRUE) : NA/Inf durch gr??te positive Zahl ersetzt (NA/Inf replaced by positive number) 2: In nlm(function(p) dev(p), p = rep(ip, length.out = np), hessian

R users; A question about optimization within R. I've been using both optim() and nlminb() to estimate parameters and all seems to be working fine. For context (but without getting into specifics - sorry), I'm working with a problem that is known to have correlated parameters, and parameter estimation can be difficult. I have a question on optim() - I'm using

Dear Helper, I used "constrOptim.nl" and got the value of par. The estimations looks good even if the number of iterations is only 16. But the values of hessian and convergence are both "NULL". I tested the objective function and gradient function by "optim" and didn't see any problem there. With these functions, "optim" gives the convergence value

Hello all, I am trying to get the variance-covariance (VCOV) matrix of the parameter estimates produced from the nlminb minimizing function, using vcov.nlminb, but it seems to have been expunged from the MASS library. The hessian from nlminb is also producing NaNs, although the estimates seems to be right, so I can't VCOV that way either. I also tried using the vcov function after minimizing

Dear expeRts, first of all I'd like to thank you for the quick help on my last which() problem. Here is another one I could not tackle: I have data on an absorption measurement which I want to fit with an voigt profile: fn.1 <- function(p){ for (i1 in ilong){ ff <- f[i1] ex[i1] <- exp(S*n*L*voigt(u,v,ff,p[1],p[2],p[3])[[1]]) } sum((t-ex)^2) } out <-

Dear R experts! I try to minimize a function with external C fitting function. I get the hessian matrix. Here it is: [,1] [,2] [,3] [,4] [1,] 1.8816631 0 0.8859803 0 [2,] 0.0000000 0 0.0000000 0 [3,] 0.8859803 0 0.4859983 0 [4,] 0.0000000 0 0.0000000 0 Second and fourth rows/columns have zero values only. That's OK, because that ones related

Hi, Use the `auglag' function in "alabama" if you want to get the Hessian at convergence. This typically tends to perform better than `constrOptim.nl'. Also, `constrOptim.nl' does not compute the Hessian. You should not specify method="L-BFGS-B". The default method "BFGS" is better in this setting. Hope this helps, Ravi

Hi, R developers. when running mle inside a loop I found a nasty behavior. From time to time, my model had a degenerate minimum and the loop just crashed. I tracked it down to "vcov <- if (length(coef)) solve(oout$hessian)" line, being the hessian singular. Note that the minimum reached was good, it just did not make sense to calculate the covariance matrix as the inverse of a

Hi everyone. I have a problem that I have been unable to determine either the best way to proceed and why the methods I'm trying to use sometimes fail. I'm using the pf() function in an optimization function to find a noncentrality parameter that leads to a specific value at a specified quantile. My goal is to have a general function that returns the noncentrality parameter that

It would be really nice if nlm took a set of "..." optional arguments that were passed through to the objective function. This level of hacking is probably slightly beyond me: is there a reason it would be technically difficult/inefficient? (I have a vague memory that it used to work this way either in S-PLUS or in some previous version of R, but I could easily be wrong.) Here's

Dear R users, I have looked in the reference Schnabel, R. B., Koontz, J. E. and Weiss, B. E. (1985) A modular system of algorithms for unconstrained minimization. _ACM Trans. Math. Software_, *11*, 419-440. cited in the nlm help. This article says that the algorithm permits the use of step selection (line search, dogleg and optimal step), analytic or finite diference gradient

Hello: Below is a toy logistic regression problem. When I wrote my own code, Newton-Raphson converged in three iterations using both the gradient and the Hessian and the starting values given below. But I can't get nlm() to work! I would much appreciate any help. > x [1] 10.2 7.7 5.1 3.8 2.6 > y [1] 9 8 3 2 1 > n [1] 10 9 6 8 10 derfs4=function(b,x,y,n) {

Admittedly I am using an old version 1.7.1, but can anyone tell if this is or was a problem. I can only get nlm (nonlinear minimization) to adjust the first three components of function variable. No gradient or hessian is supplied. E.G.; fnoise function(y) { y[5]/(y[4]*sp2) * exp(-((x[,3]-y[1]-y[2]*x[,1]-y[3] *x[,2])/y[4])^2/2) + (1-y[5])/(y[9]*sp2) * exp(-((x[,3]-y[6]-y[7]*x[,1]-y[8]