tools/misc.R

In brunomontezano/masters: Master's Thesis Analysis

load("data/cohort_2007.rda")

initial_split <- rsample::initial_split(cohort_2007, strata = ocup_atual_dic_t2)

train <- rsample::training(initial_split)
test <- rsample::testing(initial_split)

resamples <- rsample::vfold_cv(
    train,
    v = 3,
    repeats = 1,
    strata = ocup_atual_dic_t2)

rec <- train |>
    recipes::recipe(ocup_atual_dic_t2 ~ .) |>
    recipes::update_role(rec, new_role = "id") |>
    recipes::step_filter_missing(recipes::all_predictors(), threshold = 0.1) |>
    recipes::step_nzv(recipes::all_predictors()) |>
    recipes::step_other(recipes::all_nominal_predictors(), threshold = 0.2) |>
    recipes::step_dummy(recipes::all_nominal_predictors())

mod <- parsnip::logistic_reg(penalty = tune::tune(), mixture = 1) |>
    parsnip::set_mode("classification") |>
    parsnip::set_engine("glmnet")

wf <- workflows::workflow() |>
    workflows::add_recipe(rec) |>
    workflows::add_model(mod)

rs <- tune::tune_grid(
    object = wf,
    resamples = resamples,
    grid = dials::grid_regular(dials::penalty(range = c(-4, -1)),
                               levels = 20),
    metrics = yardstick::metric_set(yardstick::roc_auc,
                                    yardstick::accuracy),
    control = tune::control_grid(verbose = TRUE, allow_par = TRUE)
)

rs |> tune::show_best()

work_last_fit <- tune::last_fit(
    object = wf,
    split = initial_split
)