tests/testthat/test_py_object_slicing.R

In rTorch: R Bindings to 'PyTorch'

library(testthat)

source("helper_utils.R")

skip_if_no_python()

# function narrow() extracts part of a tensor ----------------------------------
#
context("extract parts of a Python object 1")

test_that("extract a slice from a list", {
    builtins    <- import_builtins()
    py_run_string("li = [0, 1, 2, 3, 4, 5]")
    sl <- builtins$slice(1L, 3L)
    py_li <- r_to_py(py$li)
    expect_equal(py_li$`__getitem__`(sl), r_to_py(py_eval("[1, 2]")))
    expect_equal(py$li[c(1, 3)], c(0, 2))
    expect_equal(py$li[c(2, 3)], c(1, 2))
    py_run_string("del li")  # remove variable from Python environment
})


context("extract parts of a Python object 2")
test_that("tensor dimension is 4D: 60000x3x28x28", {
    skip_if_no_torch()

    img <<- torch$ones(60000L, 3L, 28L, 28L)
    expect_equal(tensor_dim(img), c(60000, 3, 28, 28))
    expect_equal(tensor_ndim(img), 4)
    # print(img[10:11][0:1])
    expect_equal(tensor_dim(py_eval("r.img[0:10]")), c(10, 3, 28, 28))


    # error with new slicing
    # expect_equal(tensor_dim(img[0:9]), c(10, 3, 28, 28))
    expect_equal(tensor_dim(img[1:10,,,]), c(10, 3, 28, 28))
    # print(tensor_dim(img[1:10,,,]))

    expect_equal(tensor_dim(py_eval("r.img[0:10, 0:1]")), c(10, 1, 28, 28))

    # ERROR with new slicing
    # expect_error(tensor_dim(img[0:10, 0:1]))
    expect_equal(tensor_dim(img[1:10, 1:2,,]), c(10, 2, 28, 28))

    # ERROR with new slicing
    # expect_error(tensor_dim(img[c(list(0:10), list(0:1))]))
    expect_equal(tensor_dim(img[1:10, 1:2, 1:5, 1:5]), c(10, 2, 5, 5))


    slice1D <- py_eval("[x for x in range(100)]")
    slice2D <- py_eval("[x for x in range(0)]")
    # print(tensor_dim(img[10:10][1]))
    # print(tensor_dim(img[10:10][0:0]))
    # print(tensor_dim(img[10:10][0:1]))
    # print(tensor_dim(img[10:10][1:1]))
})