add refit.NNETAR

[Tim-TU]

No description provided.

[mitchelloharawild]

Looks great. Some small changes requested. I'll fix support for the scaling of inputs.

[mitchelloharawild]

This can be simplified to:
scale_in <- length(unlist(object$scales)) != 0

[mitchelloharawild]

A simple trick to refit without re-estimation is to set maxit=0:

# from ?nnet::nnet
library(nnet)
ir <- rbind(iris3[,,1],iris3[,,2],iris3[,,3])
targets <- class.ind( c(rep("s", 50), rep("c", 50), rep("v", 50)) )
samp <- c(sample(1:50,25), sample(51:100,25), sample(101:150,25))
ir1 <- nnet(ir[samp,], targets[samp,], size = 2, rang = 0.1,
            decay = 5e-4, maxit = 200)
#> # weights:  19
#> initial  value 55.564730 
#> iter  10 value 45.612146
#> iter  20 value 11.169247
#> iter  30 value 2.334405
#> iter  40 value 2.300697
#> iter  50 value 2.285848
#> iter  60 value 2.269265
#> iter  70 value 2.214906
#> iter  80 value 1.942610
#> iter  90 value 1.779559
#> iter 100 value 1.760020
#> iter 110 value 1.752948
#> iter 120 value 1.751550
#> iter 130 value 1.750537
#> iter 140 value 1.750014
#> iter 150 value 1.749923
#> iter 160 value 1.749906
#> iter 170 value 1.749886
#> iter 180 value 1.749876
#> final  value 1.749870 
#> converged
ir2 <- nnet(ir[samp,], targets[samp,], size = 2, rang = 0.1,
            decay = 5e-4, maxit = 0, Wts = ir1$wts)
#> # weights:  19
identical(ir1$fitted.values, ir2$fitted.values)
#> [1] TRUE

^{Created on 2020-06-30 by the reprex package (v0.3.0)}

[mitchelloharawild]

Oops, unintentionally merged this when trying to push to your pull request. I'll fix the above requests.

Thanks again for this pull request!