nlminb
nlminb.
nlminb.control(eval.max, iter.max, abs.tol, rel.tol, x.tol,
step.min, step.max, scale.init, scale.upd,
scale.fac, scale.tol, scale.mod, sing.tol,
sing.step, diff.g)
scale. If
scale.init >= 0, all components of the scale vector will be initialized to
scale.init before updating. If
scale.init < 0, the scale vector will not
be initialized internally. Default : 0.0
scale at the beginning of
an iteration. Each component of
scale is set either to the norm of the
corresponding diagonal element of the Hessian matrix or else to
scale.fac
times its current value, whichever is larger. Default : 0.6
scale during
automatic updating. If a component of
scale becomes smaller than
scale.tol
in magnitude, it is set to
scale.mod.
Default :
scale.tol = 10^(-6).
scale during
automatic updating. If a component of
scale becomes smaller than
scale.tol
in magnitude, it is set to
scale.mod.
Default :
scale.mod = 1.0.
sing.tol = max( 10^(-10), .Machine$double.eps^(2/3))
sing.step = 1
nlminb.
Automatic updating for
scale is available only if
hessian is supplied.
nlminb.control would normally be called only to set the
control parameter
in a call to
nlminb. Names may not be abbreviated.
For details, see the PORT Mathematical Subroutine Library Manual.
A. T. & T. Bell Laboratories (1984). PORT Mathematical Subroutine Library Manual.
norm.neg.2.11 <- function(theta, y)
{
length(y) * log(theta[2]) +
(1/theta[2]) * sum(y-theta[1])^2
}
set.seed(12)
my.obs <- rnorm(100, mean=10, sd=2)
nlminb(start = c(0,1), obj = norm.neg.2.11,
control = nlminb.control(iter.max = 100, x.tol = .0001),
lower = c(-Inf, 0), y = my.obs)