寓言:: NNETAR的可变重要性？

Question

是否有一种方法可以从fable::NNETAR模型中获得可变重要性的视图？像下面的小编nnet示例一样？

library(tidyverse)
library(fable)
#> Loading required package: fabletools
library(NeuralNetTools)
library(nnet)

ts <- tibble(index = 1:1000) %>%
  mutate(
    y = seq(0.5, 1000 / 2, by = 0.5) + arima.sim(model = list(
      ar = c(0.2, 0.3),
      ma = 0.2,
      order = c(2, 1, 1)
    ), 999) %>% as.double(),
    x1 = seq(0.5, 1000 / 2, by = 0.5),
    x2 = 1:1000,
    x3 = 1
  ) %>% 
  as_tsibble(index = index)

# Possible to extract variable importance from fable::NNETAR ...
mod <- ts %>%
  model(NNet = NNETAR(y ~ x1 + x2 + x3))
#> Warning: Constant xreg column, setting `scale_inputs=FALSE`

# ... as per the nnet / garson example?
nnet(y ~ x1 + x2 + x3, data = ts, size = 1) %>% 
  garson() +
  ggtitle("NN Variable Importance")
#> # weights:  6
#> initial  value 53799287.386267 
#> final  value 53627119.744357 
#> converged

^{由reprex package（v0.3.0）于2019-09-18创建}