无法使用 Tidy 模型调整我的随机森林（带护林员）中的 mtry 参数 R

Question

I am trying to tune the parameters for a random forest model using tune() and the Tidy model environment in R. I am using ranger as the engine and this is a classification model, but I cannot tune the mtry parameter.

I tried:

random_forest <- rand_forest(min_n = tune(),
                            trees = tune(),
                            mtry = tune()) %>%
                    set_mode("classification") %>%
                    set_engine("ranger")

set.seed(3)
grid <- grid_random(parameters(random_forest), size = 20)

It works perfectly fine for trees and and min_n parameters, but if I try to use mtry = tune(), then create a grid, then I get the following error message:

Error in grid_random() : : These arguments contains unknowns: mtry . See the . See the finalize()` function. Traceback:

1. grid_random(parameters(random_forest), size = 20)
2. grid_random.parameters(parameters(random_forest), size = 20)
3. make_random_grid(,,,params. size = size. original = original. . filter = {. { . filter . } . })
4. validate_params(..., call = call)
5. rlang::abort(paste0("These arguments contains unknowns: ", paste0(" ", . bad_param, " ", collapse = ","), ". See the finalize() function."), . call = call)
6. signal_abort(cnd, .file)`

I have tried grid_random and grid_regular and different values for size or levels, but it will only work if I remove mtry as it doesn't to recognise it as a valid argument.

Answer 1

The values that the mtry hyperparameter of the model can take on depends on the training data . Before you give some training data to the parameters, it is not known what would be good values for mtry . You can finalize() the parameters by passing in some of your training data:

library(tidymodels)

rf_spec <- 
  rand_forest(min_n = tune(),
              trees = tune(),
              mtry = tune()) %>%
  set_mode("classification") %>%
  set_engine("ranger")

set.seed(3)
extract_parameter_set_dials(rf_spec)
#> Collection of 3 parameters for tuning
#> 
#>  identifier  type    object
#>        mtry  mtry nparam[?]
#>       trees trees nparam[+]
#>       min_n min_n nparam[+]
#> 
#> Model parameters needing finalization:
#>    # Randomly Selected Predictors ('mtry')
#> 
#> See `?dials::finalize` or `?dials::update.parameters` for more information.

extract_parameter_set_dials(rf_spec) %>%
  finalize(mtcars)  ## imagine mtcars was your training data
#> Collection of 3 parameters for tuning
#> 
#>  identifier  type    object
#>        mtry  mtry nparam[+]
#>       trees trees nparam[+]
#>       min_n min_n nparam[+]

^{Created on 2023-02-01 with reprex v2.0.2}