inst/test_mnist_digits_png_full.R

In rTorch: R Bindings to 'PyTorch'

library(testthat)

skip()
skip_on_cran()
skip_on_travis()

source("tensor_functions.R")

context("methods of train_dataset")

train_data_path = file.path('~/mnist_digits_png_full/training/')
test_data_path  = file.path('~/mnist_digits_png_full/testing/')

# read the datasets without normalization or corrections; as-is
train_dataset <<-  torchvision$datasets$ImageFolder(
  root = normalizePath(train_data_path),
  transform = torchvision$transforms$ToTensor()
)

test_dataset <<- torchvision$datasets$ImageFolder(
  root = normalizePath(test_data_path),
  transform = torchvision$transforms$ToTensor()
)

context("Sanity of local folder for MNIST images")

test_that("folders exist", {
  expect_true(dir.exists("~/"))
  expect_true(dir.exists(normalizePath(train_data_path)))
  expect_true(dir.exists(normalizePath(test_data_path)))
})



context("methods of train_dataset")

test_that("names of methods", {
  expect_equal(names(train_dataset),
               c("class_to_idx", "classes", "extensions", "extra_repr",
                 "imgs", "loader", "root", "samples", "target_transform",
                 "targets", "transform", "transforms"))
})


test_that("attributes of train_dataset", {

  expect_true(all(py_list_attributes(train_dataset) %in% c(
    "class_to_idx", "classes", "extensions", "extra_repr",
    "imgs", "loader", "root", "samples", "target_transform",
    "targets", "transform", "transforms",

    "__add__", "__class__", "__delattr__",
    "__dict__","__dir__", "__doc__",
    "__eq__","__format__", "__ge__",
    "__getattribute__", "__getitem__", "__gt__",
    "__hash__", "__init__", "__init_subclass__",
    "__le__", "__len__", "__lt__",
    "__module__", "__ne__", "__new__",
    "__reduce__", "__reduce_ex__", "__repr__",
    "__setattr__", "__sizeof__", "__str__",
    "__subclasshook__", "__weakref__", "_find_classes",
    "_format_transform_repr", "_repr_indent"))
  )
})

test_that("train_dataset has __getitem()__ method", {
  if (rTorch:::is_linux()) expect_error(!train_dataset[0])
  if (rTorch:::is_linux()) expect_error(!train_dataset[[0]])
  if (rTorch:::is_windows()) expect_true(rTorch:::is_tensor(train_dataset[[2]][[1]]))
  result <- train_dataset$`__getitem__`(0L)
  expect_equal(length(result), 2)
})

test_that("train_dataset returns a list of 2 elements with py_get_item()", {
  result <- py_get_item(train_dataset, 0L)
  expect_equal(class(result), "list")
  expect_equal(length(result), 2)
})

test_that("train_dataset is a tuple in Python", {
  expect_equal(as.character(py_eval("type(r.train_dataset[0])")),
               "<class 'tuple'>")
})

test_that("1st member of train_dataset list is a tensor: image", {
  result <- py_get_item(train_dataset, 0L)[[1]]
  # print(result)
  expect_true(rTorch:::is_tensor(result))
})

test_that("2nd member of train_dataset list is an integer: label", {
  result <- py_get_item(train_dataset, 0L)[[2]]
  expect_true(is.integer(result))
})

test_that("train_dataset is a Python tuple object but an R list", {
  result <- train_dataset$`__getitem__`(0L)
  expect_equal(class(result), "list")
  expect_equal(as.character(py_eval("type(r.train_dataset[0])")),
               "<class 'tuple'>")
})

test_that("dimension of the tensor is 3D", {
  result <- py_get_item(train_dataset, 0L)[[1]]
  expect_equal(tensor_dim_(result), 3)
})

test_that("dimensions of the tensor is 3x28x28", {
  result <- py_get_item(train_dataset, 0L)[[1]]
  expect_equal(tensor_dim(result), c(3, 28, 28))
})

test_that("length of the train_dataset is 60000", {
  result <- py_len(train_dataset)
  expect_equal(result, 60000)
})

test_that("length of the test_dataset is 10000", {
  result <- py_len(test_dataset)
  expect_equal(result, 10000)
})

test_that("1st label of the train_dataset is 0", {
  result <- py_get_item(train_dataset, 0L)[[2]]
  expect_equal(result, 0)
  result <- py_get_item(train_dataset, 59999L)[[2]]
})

test_that("59999th label of the train_dataset is 9", {
  result <- py_get_item(train_dataset, 59999L)[[2]]
  expect_equal(result, 9)
})

test_that("last label of the train_dataset is 9", {
  result <- py_get_item(train_dataset, py_len(train_dataset)-1L)[[2]]
  expect_equal(result, 9)
})

test_that("last label of the train_dataset using py_object_last()", {
  result <- py_get_item(train_dataset, py_object_last(train_dataset))[[2]]
  expect_equal(result, 9)
})

test_that("train and test datasets have __len__ method in Python", {
  expect_true(py_has_length(train_dataset))
  expect_true(py_has_length(test_dataset))
})

test_that("these objects do not have Python __len__", {
  expect_true(py_has_length(py_get_item(train_dataset, 0L)[[1]]))
  expect_false(py_has_length(py_get_item(train_dataset, 0L)[[2]]))
  expect_false(py_has_length(py_get_item(train_dataset, 0L)))
  result <-  r_to_py(py_get_item(train_dataset, 0L)[[2]])
  expect_equal(as.character(result$`__class__`), "<class 'int'>")
})

test_that("dataset label is of type integer", {
  expect_true(is.integer(py_get_item(train_dataset, 0L)[[2]]))
  result <-  r_to_py(py_get_item(train_dataset, 0L)[[2]])
  expect_equal(as.character(result$`__class__`), "<class 'int'>")
})