lightgbm model doesn't see sample_size parameter from boost_tree

[bwilkowski]

While I was testing bonsai package I noticed some odd behaviour regarding tuning parameters. When I indicated which parameters should be tuned via boost_tree I saw that lighgbm did not use sample_size. Is that correct behaviour?
I used following code to configure model.

pre_proc <-
recipe(status ~ ., data = data_train) %>%
update_role(cif, data_alertu, okres_alertu, new_role = "id variable") %>%
step_zv(all_predictors()) %>%
step_corr(all_predictors(), threshold = .9)

gbm_mod <-
boost_tree(
mode = "classification",
trees = tune(),
tree_depth = tune(),
min_n = tune(),
loss_reduction = tune(),
sample_size = tune(),
mtry = tune(),
learn_rate = tune()
) %>%
set_engine("lightgbm")

gbm_wflow <-
workflow() %>%
add_model(gbm_mod) %>%
add_recipe(pre_proc)

gbm_set <- extract_parameter_set_dials(gbm_wflow) %>% update(mtry = mtry(c(10,50)))

gbm_set returned only 6 parameters, although I set 7 to be tuned. I saw that sample_size is missing. Shouldn't it be translated to lightgbm's bagging_fraction parameter?

[simonpcouch]

Ah, my apologies! This is indeed an issue.

A quick reprex to jump off from when I have a moment for this:

library(tidymodels)
library(bonsai)
data(penguins, package = "modeldata")

bt <-
  boost_tree(sample_size = tune()) %>%
  set_engine(engine = "lightgbm") %>%
  set_mode(mode = "classification")

grid <-
  tune_grid(
    bt,
    species ~ flipper_length_mm + island,
    bootstraps(penguins)
  )
#> Warning: No tuning parameters have been detected, performance will be evaluated
#> using the resamples with no tuning. Did you want to [tune()] parameters?

^{Created on 2022-07-06 by the reprex package (v2.0.1)}

The usage of sample_prop is only internal and need not be user-facing—just helps us come up with reasonable values to tune over. Looks like parsnip has a way of handling this that I'll either need to PR over there for or emulate in bonsai.

I'm focused elsewhere at the moment, but a fix for this will definitely be included in the next release of the package. :)

Thanks again for the report.

[github-actions]

This issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with a reprex: https://reprex.tidyverse.org) and link to this issue.