tests/testthat/test_loghaz.R

In survivalmodels: Models for Survival Analysis

skip_if_no_pycox()

set.seed(1)
np <- reticulate::import("numpy")
np$random$seed(1L)
torch <- reticulate::import("torch")
torch$manual_seed(1L)

test_that("silent", {
  # suppress cuts warning
  expect_silent({fit <- loghaz(Surv(time, status) ~ ., data = rats[1:50, ],
                                                verbose = FALSE,
                               frac = 0.3)})
  expect_silent({predict(fit, newdata = rats[51:100, ])})
})

test_that("custom", {
  expect_silent(loghaz(Surv(time, status) ~ ., data = rats[1:10, ],
                         custom_net = build_pytorch_net(3L, 10L),
                         verbose = FALSE))
})

# test_that("cutpoints", {
#   expect_equal(loghaz(Surv(time, status) ~ ., data = rats[1:50, ], verbose = FALSE,
#                        cutpoints = c(20, 50, 70))$model$duration_index,
#                array(c(20, 50, 70)))
#   expect_silent(loghaz(Surv(time, status) ~ ., data = rats[1:50, ], verbose = FALSE,
#                        cuts = 3))
# })

test_that("interpolate", {
  fit <- loghaz(Surv(time, status) ~ ., data = rats[1:50, ], verbose = FALSE)
  expect_silent(predict(fit, interpolate = TRUE))
})

test_that("auto sanity", {
  sanity_check(model = "loghaz",
               pars = list(frac = 0.3, activation = "relu", num_nodes = c(2L, 2L),
                           dropout = 0.1, early_stopping = TRUE, epochs = 100L, batch_size = 32L))
})