tests/testthat/test-cv.sail.R

In sahirbhatnagar/funshim: Sparse Additive Interaction Learning

library(doParallel)
doParallel::registerDoParallel(cores = 2)
expect_matrix <- function(x) inherits(x,"matrix")
set.seed(1234) # we set the seed so that the cv error curves remain identical when testing (randomness is introduced due to CV folds)

context("cv.sail model fit, parallel, predict, plot with both packaged datasets")

testthat::skip_on_cran()
testthat::skip_on_appveyor()

data("sailsim")
data("oasis")
f.basis <- function(i) splines::bs(i, degree = 3)

cvfit_sim <- try(cv.sail(x = sailsim$x, y = sailsim$y, e = sailsim$e, basis = f.basis,
                         dfmax = 5, nfolds = 3),
                 silent = TRUE)

cvfit_oasis <- try(cv.sail(x = oasis$x, y = oasis$y, e = oasis$e, basis = f.basis,
                           dfmax = 5, nfolds = 6, alpha = 0.8),
                   silent = TRUE)

cvfit_parallel <- try(cv.sail(x = sailsim$x, y = sailsim$y, e = sailsim$e,
                              dfmax = 5,
                              basis = f.basis, nfolds = 3, parallel = TRUE),
                      silent = FALSE)

cvfit_parallel_weak <- try(cv.sail(x = sailsim$x, y = sailsim$y, e = sailsim$e, strong = TRUE,
                              dfmax = 5,
                              basis = f.basis, nfolds = 3, parallel = TRUE),
                      silent = FALSE)

new_x <- replicate(20, rnorm(50))
new_e <- rnorm(50, sd = 0.5)

test_that("no error in fitting cv.sail and parallel version for both simulated and real data", {

  expect_false(inherits(cvfit_sim, "try-error"))
  expect_false(inherits(cvfit_oasis, "try-error"))
  expect_false(inherits(cvfit_parallel, "try-error"))
  expect_false(inherits(cvfit_parallel_weak, "try-error"))
  expect_is(cvfit_sim, "cv.sail")
  expect_is(cvfit_oasis, "cv.sail")
  expect_is(cvfit_parallel, "cv.sail")

})


test_that("no error in predict for cv.sail", {

  expect_true(expect_matrix(predict(cvfit_sim, type = "nonzero", s = "lambda.min")))
  expect_true(expect_matrix(predict(cvfit_oasis, type = "nonzero", s = "lambda.min")))
  expect_true(expect_matrix(predict(cvfit_parallel, type = "nonzero", s = "lambda.min")))

  expect_true(expect_matrix(predict(cvfit_sim)))
  expect_true(expect_matrix(predict(cvfit_oasis)))
  expect_true(expect_matrix(predict(cvfit_parallel)))

  expect_equivalent(predict(cvfit_sim, s = Inf),
                    matrix(rep(cvfit_sim$sail.fit$a0[1], cvfit_sim$sail.fit$nobs)))

  expect_equivalent(predict(cvfit_oasis, s = Inf),
                    matrix(rep(cvfit_oasis$sail.fit$a0[1], cvfit_oasis$sail.fit$nobs)))

  expect_equivalent(predict(cvfit_parallel, s = Inf),
                    matrix(rep(cvfit_parallel$sail.fit$a0[1], cvfit_parallel$sail.fit$nobs)))

  expect_matrix(predict(cvfit_sim, newx = new_x, newe = new_e))
  expect_equal(dim(predict(cvfit_sim, newx = new_x, newe = new_e, s = c(5,3,1,0.5))), c(nrow(new_x),4))

  expect_equivalent(class(coef(cvfit_sim)), "dgCMatrix")
  expect_equal(dim(coef(cvfit_sim))[[1]], dim(cvfit_sim$sail.fit$design)[[2]] + 1)

})



context("cv plot")

test_that("check plots for cv curve", {

  disp_cv_curve <- function() plot(cvfit_sim)
  vdiffr::expect_doppelganger("cv.sail cv curve", disp_cv_curve)

})