def test_no_side_effects(self):
"""Test that the input is not mutated."""
# Generate random data
np.random.seed(42)
costs = np.random.randn(10, 3)
# Copy the data before passing into function
costs_before = costs.copy()
crowding_distance(costs)
# The input argument is not changed/mutated in any way
assert costs_before.shape == costs.shape
assert np.all(costs_before == costs)
def test_invariance_under_permutations(self, seed):
"""Test that the algorithm in invariant under random permutations of data."""
# Create some random data
np.random.seed(seed)
n_costs = np.random.randint(1, 9)
n_objectives = np.random.randint(1, 4)
# Permutation invariance does not hold if rows are duplicated
# Therefore we use floats here
costs = np.random.randn(n_costs, n_objectives)
# Get masks
distances = crowding_distance(costs)
# Verify that one distance is returned for each row
assert len(distances) == n_costs
# Permute the data
permutation = np.random.permutation(np.arange(n_costs))
# Permutation invariance should hold
assert np.all(distances[permutation] == crowding_distance(costs[permutation]))
def test_on_known_instance(self):
"""Test on a small instance computed by hand."""
costs = np.array([[1, 2], [3, 9], [5, 3], [9, 1]], dtype=np.float)
dist = crowding_distance(costs)
assert np.all(dist == np.array([np.inf, np.inf, 0.8125, np.inf]))