tests/testthat/test_gbm_unify.R

In ModelOriented/treeshap: Compute SHAP Values for Your Tree-Based Models Using the 'TreeSHAP' Algorithm

library(treeshap)
suppressWarnings(library(gbm, quietly = TRUE))

x <- fifa20$data
x['value_eur'] <- fifa20$target

gbm_with_cat_model <- gbm::gbm(
  formula = value_eur ~ .,
  data = x,
  distribution = "laplace",
  n.trees = 10,
  interaction.depth = 2,
  n.cores = 1
)

x <- x[colnames(fifa20$data) != 'work_rate']

gbm_num_model <- gbm::gbm(
  formula = value_eur ~ .,
  data = x,
  n.trees = 50,
  distribution = 'gaussian',
  n.cores = 1
)

test_that('gbm.unify returns an object of appropriate class', {
  expect_true(is.model_unified(gbm.unify(gbm_num_model, x)))
  expect_true(is.model_unified(unify(gbm_num_model, x)))
})


test_that('gbm.unify returns an object with correct attributes', {
  unified_model <- gbm.unify(gbm_num_model, x)

  expect_equal(attr(unified_model, "missing_support"), TRUE)
  expect_equal(attr(unified_model, "model"), "gbm")
})

test_that('the gbm.unify function does not support models with categorical features', {
  expect_error(gbm.unify(gbm_with_cat_model), "Models built on data with categorical features are not supported - please encode them before training.")
})

test_that('the gbm.unify function returns data frame with columns of appropriate column', {
  unifier <- gbm.unify(gbm_num_model, x)$model
  expect_true(is.integer(unifier$Tree))
  expect_true(is.integer(unifier$Node))
  expect_true(is.character(unifier$Feature))
  expect_true(is.factor(unifier$Decision.type))
  expect_true(is.numeric(unifier$Split))
  expect_true(is.integer(unifier$Yes))
  expect_true(is.integer(unifier$No))
  expect_true(is.integer(unifier$Missing))
  expect_true(is.numeric(unifier$Prediction))
  expect_true(is.numeric(unifier$Cover))
})

test_that("shap calculates without an error", {
  unifier <- gbm.unify(gbm_num_model, x)
  expect_error(treeshap(unifier, x[1:3,], verbose = FALSE), NA)
})

test_that("gbm: mean prediction calculated using predict == using covers", {
  unifier <- gbm.unify(gbm_num_model, x)

  intercept_predict <- mean(predict(unifier, x))

  ntrees <- sum(unifier$model$Node == 0)
  leaves <- unifier$model[is.na(unifier$model$Feature), ]
  intercept_covers <- sum(leaves$Prediction * leaves$Cover) / sum(leaves$Cover) * ntrees

  #expect_true(all(abs((intercept_predict - intercept_covers) / intercept_predict) < 10**(-14)))
  expect_equal(intercept_predict, intercept_covers)
})

test_that("gbm: covers correctness", {
  unifier <- gbm.unify(gbm_num_model, x)

  roots <- unifier$model[unifier$model$Node == 0, ]
  expect_true(all(roots$Cover == nrow(x)))

  internals <- unifier$model[!is.na(unifier$model$Feature), ]
  yes_child_cover <- unifier$model[internals$Yes, ]$Cover
  no_child_cover <- unifier$model[internals$No, ]$Cover
  if (all(is.na(internals$Missing))) {
    children_cover <- yes_child_cover + no_child_cover
  } else {
    missing_child_cover <- unifier$model[internals$Missing, ]$Cover
    missing_child_cover[is.na(missing_child_cover)] <- 0
    missing_child_cover[internals$Missing == internals$Yes | internals$Missing == internals$No] <- 0
    children_cover <- yes_child_cover + no_child_cover + missing_child_cover
  }
  expect_true(all(internals$Cover == children_cover))
})