/torch-totem

Torch test module

Primary LanguageLuaBSD 3-Clause "New" or "Revised" LicenseBSD-3-Clause

Build Status

Totem - Torch test module

Totem is a small test package in the style of torch.Tester. The interface is essentially the same as torch.Tester but differs/extends it in the following ways:

  • the test runner has different output (with colours!)
  • the test runner can output test results in a simple machine-readable file
  • the package includes some standard tests for nn modules

Basic test description

A test script can be written as follows:

require 'totem'

local tests = totem.TestSuite()

local tester = totem.Tester()

function tests.TestA()
  local a = 10
  local b = 10
  tester:asserteq(a, b, 'a == b')
  tester:assertne(a,b,'a ~= b')
end

function tests.TestB()
  local a = 10
  local b = 9
  tester:assertlt(a, b, 'a < b')
  tester:assertgt(a, b, 'a > b')
end

return tester:add(tests):run()

The command totem-init can be used to generate an empty test.

Set-up and tear-down functions

It is sometimes useful to introduce initialization or destruction code that is shared among all tests in a test suite. Totem allows you to define special _setUp and _tearDown test functions. The _setUp function is interpreted as a set-up function that is called before every test. Similarly, the _tearDown is interpreted as a tear-down function that is called after every test. The following example illustrates this.

require 'totem'

local tests = totem.TestSuite()

local tester = totem.Tester()

function tests.TestA()
  ....
end

function tests.TestB()
  ....
end

function tests._setUp(testName)
  .... set-up / initialization code
end

function tests._tearDown(testName)
  .... tear-down / clean-up up code
end

return tester:add(tests):run()

Command-line usage

When running the script from the command-line you get a number of options:

Run tests

Usage:

  ./simple.lua [options] [test1 [test2...] ]

Options:

  --list print the names of the available tests instead of running them.
  --log-output (optional file-out) redirect compact test results to file.
        This contains one line per test in the following format:
        name #passed-assertions #failed-assertions #exceptions
  --no-colour suppress colour output
  --summary print only pass/fail status rather than full error messages.
  --full-tensors when printing tensors, always print in full even if large.
        Otherwise just print a summary for large tensors.
  --early-abort (optional boolean) abort execution on first error.
  --rethrow (optional boolean) errors make the program crash and propagate up
        the stack

If any test names are specified only the named tests are run. Otherwise
all the tests are run.

Additionally the script totem-run can be used to run all test files (i.e. files with names of the form test*.lua in a directory (by default the current directory).

Nesting tests

It's possible to nest test cases. Individual test files are still assumed to be runnable as stand-alone scripts, but a test case can include the outputs of such files. For example

require 'totem'

local tester = totem.Tester()
tester:add('test_nn.lua')
tester:add('test_simple.lua')
tester:add('test_tensor.lua')
return tester:run()

will first run all the tests in each of the listed test files and then report the overall test results. Each test is considered to pass only if all of its subtests pass.

Running several tests

The script scripts/totem-run will run all the files with a filename test*.lua that are inside the folder specified by the argument --folder (current folder by default). The rest of the arguments are passed to all the individual tests.

Example:

totem-run --folder tests --summary