Explicit seed setting in worker processes

[topepo]

For models that have seed arguments, parallel processing will result in inconsistent results because the worker seeds are not controlled. This PR generates a set of seeds immediately when tune_grid() or fit_resamples() are called, then sets them in the iter_*() functions. This is the same approach as caret. Unit tests will be in the extratests package.

For example:

library(tidymodels)
data(ames)
ames <- mutate(ames, Sale_Price = log10(Sale_Price))

set.seed(123)
ames_split <- initial_split(ames, prob = 0.80, strata = Sale_Price)
ames_train <- training(ames_split)
ames_test  <-  testing(ames_split)

## -----------------------------------------------------------------------------

rf_model <-
  rand_forest(trees = 1000) %>%
  set_engine("ranger") %>%
  set_mode("regression")

rf_wflow <-
  workflow() %>%
  add_formula(
    Sale_Price ~ Neighborhood + Gr_Liv_Area + Year_Built + Bldg_Type +
      Latitude + Longitude) %>%
  add_model(rf_model)

set.seed(55)
ames_folds <- vfold_cv(ames_train, v = 10)

keep_pred <- control_resamples(save_pred = TRUE)

## -----------------------------------------------------------------------------

set.seed(130)
rf_res_seq_1 <-
  rf_wflow %>%
  fit_resamples(resamples = ames_folds, control = keep_pred)

set.seed(130)
rf_res_seq_2 <-
  rf_wflow %>%
  fit_resamples(resamples = ames_folds, control = keep_pred)

## -----------------------------------------------------------------------------

library(doParallel)
cl <- makePSOCKcluster(10)
registerDoParallel(cl)

set.seed(130)
rf_res_par_1 <-
  rf_wflow %>%
  fit_resamples(resamples = ames_folds, control = keep_pred)

set.seed(130)
rf_res_par_2 <-
  rf_wflow %>%
  fit_resamples(resamples = ames_folds, control = keep_pred)

## -----------------------------------------------------------------------------

all.equal(bind_rows(rf_res_seq_1$.metrics),
          bind_rows(rf_res_seq_2$.metrics))
# CRAN tune results:
# [1] TRUE

# new results
# [1] TRUE

all.equal(bind_rows(rf_res_par_1$.metrics),
          bind_rows(rf_res_par_2$.metrics))
# CRAN tune results:
# [1] "Component “.estimate”: Mean relative difference: 0.001654902"

# new results
# [1] TRUE

all.equal(bind_rows(rf_res_seq_1$.metrics),
          bind_rows(rf_res_par_1$.metrics))

# CRAN tune results:
# [1] "Component “.estimate”: Mean relative difference: 0.001052047"

# new results
# [1] TRUE

[github-actions]

This pull request 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.