Forum: help

RE: MaxLik unused arguments [ Reply ]
By: Arne Henningsen on 2018-08-13 08:11

Hi Chris

The error occurs inside your loglik() function (in the call to dnorm()):

> startPar <- c(0.1260504, -6.5373661, 1.0222147 )
> loglik( params = startPar, data = x )
Error in log(mean = mu, sd = sigma) :
unused arguments (mean = -6.5373661, sd = 1.0222147)
> traceback()
2: dnorm(data, log(mean = mu, sd = sigma)) at #9
1: loglik(params = startPar, data = x)

Do you mean "dnorm(data, mean = mu, sd = sigma)" instead of "dnorm(data, log(mean = mu, sd = sigma))"?

Best regards,
Arne

MaxLik unused arguments [ Reply ]
By: Chris Jensen on 2018-05-13 17:50

[forum:45947]

I use maxLik to find parameters from a MLE to left-censored data, code is as follows;
LOD <- (LODppb/1000)*0.4/70
x<- c(0.000260414, 0.001164827, 0.001077806, 0.000206215, 0.003244783, 0.001254606, 0.001298323,
0.000401082, 0.000947205, 0.003621713, 0.003054214, 0.008090681, 0.004951853, 0.001570994,
0.002745296)
library(maxLik)
require(fitdistrplus)
loglik <- function(params,data){

p <- params[1]
mu <- params[2]
sigma <- params[3]
Q <- pnorm(log(LOD), mean = mu, sd = sigma, lower.tail = TRUE, log.p = FALSE)
# 212
t1 <- 15*log((1-p)+p*Q)
t4 <- (119-15)*log(p)+sum(dnorm(data,log(mean = mu, sd = sigma)))
ll <- t1+t4
return(ll)}
newdistparameters <- maxLik(loglik,data = x, start=c(0.1260504, -6.5373661, 1.0222147))
summary( newdistparameters )

But I get an error due to unused arguments. SD value is (inf) and not an actual value, could it be caused by SD going below 0 ? if so, how do I stop that from happening ?

Error in log(mean = mu, sd = sigma) :
unused arguments (mean = -6.5373661, sd = 1.0222147)
8.
dnorm(data, log(mean = mu, sd = sigma))
7.
fnOrig(theta, ...)
6.
fn(start1, fixed = fixed, sumObs = TRUE, returnHessian = returnHessian,
...)
5.
maxNRCompute(fn = function (theta, fnOrig, gradOrig = NULL, hessOrig = NULL,
fixed, sumObs = FALSE, returnHessian = TRUE, ...)
{
nParam <- length(theta) ...
4.
eval(as.call(cl))
3.
eval(as.call(cl))
2.
maxRoutine(fn = logLik, grad = grad, hess = hess, start = start,
constraints = constraints, ...)
1.
maxLik(loglik, data = x, start = c(0.1260504, -6.5373661, 1.0222147))

> summary( newdistparameters )
--------------------------------------------
Maximum Likelihood estimation
Newton-Raphson maximisation, 17 iterations
Return code 1: gradient close to zero
Log-Likelihood: -45.07829
3 free parameters
Estimates:
Estimate Std. error t value Pr(> t)
[1,] 0.8739496 Inf 0 1
[2,] -6.6636431 Inf 0 1
[3,] 0.0007371 Inf 0 1

I would appreciate any help.