def test_get_model_sparsity():
"""Tests if the method get_model_sparsity in utils.py works correctly."""
qmodel = create_quantized_network()
# Generate sparsity levels to test
sparsity_levels = np.concatenate((np.random.rand(10), [1.0, 0.0])).round(2)
# Test various sparsity levels
for true_sparsity in sparsity_levels:
qmodel = set_network_sparsity(qmodel, true_sparsity)
calc_sparsity = get_model_sparsity(qmodel)
assert np.abs(calc_sparsity - true_sparsity) < 0.01
def test_get_po2_model_sparsity():
"""Tests get_model_sparsity on a po2-quantized model.
Models quantized with po2 quantizers should have a sparsity near 0 because
if the exponent is set to 0, the value of the weight will equal 2^0 == 1 != 0
"""
qmodel = create_quantized_po2_network()
# Generate sparsity levels to test
sparsity_levels = np.concatenate((np.random.rand(10), [1.0, 0.0])).round(2)
# Test various sparsity levels
for set_sparsity in sparsity_levels:
qmodel = set_network_sparsity(qmodel, set_sparsity)
calc_sparsity = get_model_sparsity(qmodel)
assert np.abs(calc_sparsity - 0) < 0.01