tests/testthat/test-misc-issues.r

In gbm-developers/gbm3: Generalized Boosted Regression Models

context("miscellaneous tests - previous issues")
test_that("predicts correctly on unknown levels (issue #18)", {
  d <- data.frame(x=as.factor(1:20), y=1:20)
  
  train <- d[1:10,]
  test  <- d[11:20,]
  
  ## try this 10 times
  ## this isn't a deterministic bug
  params <- training_params(bag_fraction=1, num_trees = 1, min_num_obs_in_node = 1,
                            num_train = round(0.5 * nrow(train)), shrinkage=1, id=seq_along(train[,1]))
  for (ind in seq_len(10)) {
    g <- gbmt(y ~ x,
             data=train, 
             train_params=params
             )
    p <- predict(g, newdata=test, n.trees=1) - g$initF
    if (sum(abs(p)) != 0) break
  }
  
  expect_equal(sum(abs(p)), 0)
})

test_that("print.GBMFit works without cross-validation (issue #5)", {
  df <- data.frame(
    x = runif(100),
    y = runif(100),
    z = sample(0:1, 100, replace = TRUE)
  )
  
  trained_gbm <- gbmt_fit(df[, c("x", "y")], df$z)
  
  expect_error(print(trained_gbm), regexp=NA)
})


test_that("predicts correctly on unknown levels (issue #18) - old api", {
    d <- data.frame(x=as.factor(1:20), y=1:20)

    train <- d[1:10,]
    test  <- d[11:20,]

    ## try this 10 times
    ## this isn't a deterministic bug
    for (ind in seq_len(10)) {
        g <- gbm(y ~ x,
                 distribution="Gaussian",
                 bag.fraction=1,
                 data=train, 
                 n.trees=1,
                 shrinkage=1,
                 n.minobsinnode=1)
        p <- predict(g, newdata=test, n.trees=1) - g$initF
        if (sum(abs(p)) != 0) break
    }

    expect_equal(sum(abs(p)), 0)
})

test_that("print.GBMFit works without cross-validation (issue #5) - old api", {
    df <- data.frame(
        x = runif(100),
        y = runif(100),
        z = sample(0:1, 100, replace = TRUE)
    )

    trained_gbm <- gbm.fit(df[, c("x", "y")], df$z)
    
    expect_error(print(trained_gbm), regexp=NA)
})