Attempting my first randomForest model in R and am working through tuning hyperparameters. I created a spec first: tune_spec<- decision_tree() %>% set_engine("rpart") %>% set_mode("regression")
And then I tried to create a tuning grid: tree_grid<- grid_regular(parameters(tune_spec), levels=3)
but I got this error:
Error in parameters():
! parameters objects cannot be created from objects of class decision_tree.
The parameters() function was deprecated in 0.2.0 of {tune} (2022-03-18). The function to use is extract_parameter_set_dials().
library(tidymodels)
tune_spec <- decision_tree() |>
set_engine("rpart") |>
set_mode("regression")
tune_spec |>
extract_parameter_set_dials()
#> Collection of 0 parameters for tuning
#>
#> [1] identifier type object
#> <0 rows> (or 0-length row.names)
We are getting back a empty parameters object, because you need to specify which variables you want to use with tune()
tune_spec <- decision_tree(tree_depth = tune(), min_n = tune()) |>
set_engine("rpart") |>
set_mode("regression")
tune_spec |>
extract_parameter_set_dials()
#> Collection of 2 parameters for tuning
#>
#> identifier type object
#> tree_depth tree_depth nparam[+]
#> min_n min_n nparam[+]
And once you have done that, then you can pass it to grid_regular() or the other functions
tune_spec |>
extract_parameter_set_dials() |>
grid_regular(levels = 3)
#> # A tibble: 9 × 2
#> tree_depth min_n
#> <int> <int>
#> 1 1 2
#> 2 8 2
#> 3 15 2
#> 4 1 21
#> 5 8 21
#> 6 15 21
#> 7 1 40
#> 8 8 40
#> 9 15 40