Testing

Tests are run by pytest.

Implement new tests

Choose or add a new file into the tests/ directory to make a new test. This filename should start with test_ to be targeted by pytest.

Into each function of this test file, you can run multiple types of tests, e.g. you can run some assert statements.

You can also use some fixtures, i.e. tests whose output will be reused multiple times, which can be very useful when you need to load data or train your model.

from mypackage import Model

@pytest.fixture
def trained_model(init_data):
    model = Model()
    model.fit()
    return model

# Here, the argument is the previous function name, but it actually contains the function output (i.e., the model)
def test_alpha_positive(trained_model):
    assert (trained_model.alpha > 0).all()

def test_is_symmetric(trained_model):
    x1, x2 = trained_model.x, -trained_model.x
    y1, y2 = trained_model(x1), trained_model(x2)

    assert torch.isclose(y1, y2, atol=1e-6).all() # To test if two tensor are equal, it's safer to use torch.isclose

Run your tests

At the project's root, run:

pytest