Optimization using PORT routines

Unconstrained and box-constrained optimization using PORT routines.

opt_nlminb_julia(
  par0,
  fitMeth = "doubleLog_Beck",
  y,
  t,
  w = NULL,
  ylu = NULL,
  lower = NULL,
  upper = NULL,
  ...
)

Arguments

par0	Initial values for the parameters to be optimized over.
fitMeth	Curve fitting methods, one of c("doubleLog_Beck", "doubleLog_Elmore", "doubleLog_AG", "doubleLog_Zhang")
y	Numeric vector, vegetation index time-series
t	Numeric vector, Date variable
w	(optional) Numeric vector, weights of y. If not specified, weights of all NA values will be wmin, the others will be 1.0.
ylu	ymin, ymax, which is used to force ypred in the range of ylu.
lower	vectors of lower and upper bounds, replicated to be as long as start. If unspecified, all parameters are assumed to be unconstrained.
upper	vectors of lower and upper bounds, replicated to be as long as start. If unspecified, all parameters are assumed to be unconstrained.
...	ignored parameters

Value

A list object of

• par: The optimal parameters

• convergence:

  • 0: convergent;

  • 1: Non-convergent

• iterations

• evaluations: list(function, gradient)

• objective

See also

stats::nlminb()

Examples

if (FALSE) {

t    = seq(1.0, 366, 8)
fun  = doubleLog_Beck
par  = c(0.1 , 0.7, 50, 0.1, 250, 0.1)
par0 = c(0.05, 0.6 , 45, 0.1, 200, 0.2)

ypred = t*0
y     = fun(par, t)

julia_init()
r_julia  <- opt_nlminb_julia(par0, "doubleLog_Beck", y, t)
r_R <- opt_nlminb(par0, f_goal, fun = fun, y = y, t = t, pred = ypred)

list(julia = r_julia, R = r_R) %>%
    map(~c(.$par, .$objective, .$value)) %>%
    do.call(rbind, .)# %>%

n <- length(t)
w <- rep(0.2, n)
# julia is 5 times faster
{
    # microbenchmark::microbenchmark : 18.939826 ms in R
    info <- rbenchmark::benchmark(
        r1 <- opt_nlminb_julia(par0, "doubleLog_Beck", y, t, w),
        r2 <- opt_nlminb(par0, f_goal, fun = fun, y = y, t = t, pred = ypred),
        replications = 500
    )
    print(info)
}

}