Obtaining the search region of the latin_hypercube grid function

Question

I use tune_grid() with no parameter grid to tune my hyperparameters (see below, please). According to the help page of the tune_grid() function, a parameter grid will be created using dials::grid_latin_hypercube(). I understand (please correct me if I am wrong) that the hypercube function divides the search region into equal subdivisions and randomly picks values from each of these subdivisions. I need to obtain the range of each of these subdivisions. How can I obtain that?

set.seed(345)
Data_RF_fit <- 
  Data_RF_wflow %>% 
  tune_grid(val_set,
            grid = 25, 
            control = control_grid(save_pred = TRUE),
            metrics = metric_set(rmse))

set.seed(345)
Data_KKNN_fit <- 
  Data_KKNN_wflow %>%  
  tune_grid(val_set,
            grid = 25,
            control = control_grid(save_pred = TRUE),
            metrics = metric_set(rmse))

Answer 1

The function grid_latin_hypercube() creates a grid based on the ranges of the input values that you set. If you don't set any value, it will use the default ranges.

In your case, since I believe you're using k-nearest neighbors, you can check these values using ?nearest_neighbors to find the tunable hyperparameters. Since these hyperparameters in tidymodels can be ran as commands in the console, you can either check each option by running it as a command, which will output its default state, or check its help page.

So, for example, if you wanted to check the defaults for the neighbors parameter you could either run

neighbors()
#> # Nearest Neighbors (quantitative)
#> Range: [1, 10]

or

?neighbors

Hope this helps

Answer 2

You can extract_parameter_set_dials() from a workflow you have created, and then use that to see the parameter that is created:

library(tidymodels)

knn_spec <-
    nearest_neighbor(neighbors = tune()) %>%
    set_mode("classification") %>%
    set_engine("kknn")

knn_wf <- workflow(Class ~ ., knn_spec)

pset <- extract_parameter_set_dials(knn_wf)
pset$object
#> [[1]]
#> # Nearest Neighbors (quantitative)
#> Range: [1, 15]
## if you need to compute on these:
pset$object[[1]]$range
#> $lower
#> [1] 1
#> 
#> $upper
#> [1] 15

^{Created on 2022-05-18 by the reprex package (v2.0.1)}