tests/testthat/test-reproducible.r

In gbm-developers/gbm3: Generalized Boosted Regression Models

context("reproducibility")
test_that("Setting the seed causes result to be reproducible (1 core)", {
  set.seed(18900217)
  mod <- gbmt(Species == 'setosa' ~ ., data=iris, distribution=gbm_dist("Bernoulli"),
             cv_folds=3)
  nt1 <- gbmt_performance(mod, method="cv")
  ri1 <- relative_influence(mod, num_trees=nt1)
  
  set.seed(18900217)
  mod <- gbmt(Species == 'setosa' ~ ., data=iris, distribution=gbm_dist("Bernoulli"),
             cv_folds=3)
  nt2 <- gbmt_performance(mod, method="cv")
  ri2 <- relative_influence(mod, num_trees=nt2)
  
  expect_equal(nt1, nt2,
               label="Number of trees match when same seed is used")
  expect_equal(ri1, ri2,
               label="Relative influences match when same seed is used")
})

test_that("Setting different seeds causes result to be different (1 core)", {
  set.seed(18900217)
  mod <- gbmt(Species == 'setosa' ~ ., data=iris, distribution=gbm_dist("Bernoulli"),
             cv_folds=3)
  nt1 <- gbmt_performance(mod, method="cv")
  ri1 <- relative_influence(mod, num_trees=nt1)
  
  set.seed(19620729)
  mod <- gbmt(Species == 'setosa' ~ ., data=iris, distribution=gbm_dist("Bernoulli"),
             cv_folds=3)
  nt2 <- gbmt_performance(mod, method="cv")
  ri2 <- relative_influence(mod, num_trees=nt2)
  
  expect_false(isTRUE(all.equal(ri1, ri2)),
               label="Relative influences don't match when different seeds are used")
})

context("reproducibility - old api")

test_that("Setting the seed causes result to be reproducible (1 core)", {
  set.seed(18900217)
  mod <- gbm(Species == 'setosa' ~ ., data=iris, distribution="Bernoulli",
             cv.folds=3, n.trees=100, shrinkage=.1)
  nt1 <- gbmt_performance(mod, method="cv")
  ri1 <- relative_influence(mod, num_trees=nt1)

  set.seed(18900217)
  mod <- gbm(Species == 'setosa' ~ ., data=iris, distribution="Bernoulli",
             cv.folds=3, n.trees=100, shrinkage=.1)
  nt2 <- gbmt_performance(mod, method="cv")
  ri2 <- relative_influence(mod, num_trees=nt2)
  
  expect_equal(nt1, nt2,
               label="Number of trees match when same seed is used")
  expect_equal(ri1, ri2,
               label="Relative influences match when same seed is used")
})

test_that("Setting different seeds causes result to be different (1 core)", {
  set.seed(18900217)
  mod <- gbm(Species == 'setosa' ~ ., data=iris, distribution="Bernoulli",
             cv.folds=3, n.trees=100, shrinkage=.1)
  nt1 <- gbmt_performance(mod, method="cv")
  ri1 <- relative_influence(mod, num_trees=nt1)

  set.seed(19620729)
  mod <- gbm(Species == 'setosa' ~ ., data=iris, distribution="Bernoulli",
             cv.folds=3, n.trees=100, shrinkage=.1)
  nt2 <- gbmt_performance(mod, method="cv")
  ri2 <- relative_influence(mod, num_trees=nt2)

  expect_false(isTRUE(all.equal(ri1, ri2)),
              label="Relative influences don't match when different seeds are used")
})