def test_find_failure_indicators_with_threshold():
outputs = np.array([2, 3, -4, 6, 8, 1])
inputs = np.ones((7, 2))
mIS = MultiFidelityIS(limit_state=4)
failure_indicators = mIS._find_failure_indicators(inputs, outputs)
np.testing.assert_array_almost_equal(failure_indicators,
np.array([0, 0, 0, 1, 1, 0]))
def test_find_failure_indicators_with_limit_state_function():
outputs = np.array([2, 3, -4, 6, 8, 1])
inputs = np.ones((7, 2))
def limit_state(input):
return input - 4
mIS = MultiFidelityIS(limit_state=limit_state)
failure_indicators = mIS._find_failure_indicators(inputs, outputs)
np.testing.assert_array_almost_equal(failure_indicators,
np.array([0, 0, 0, 1, 1, 0]))
def test_calc_importance_weights_with_densities():
input_densities = np.array([.01, .1, 1])
biasing_dist_densities = np.array([4, 2, .5])
importance_sampling = MultiFidelityIS(limit_state=0)
expected_weights = input_densities / biasing_dist_densities
weights = importance_sampling._calc_importance_weights_with_densities(
input_densities=input_densities,
biasing_densities=biasing_dist_densities)
np.testing.assert_array_almost_equal(expected_weights, weights)
def test_mfis_estimate_raises_error_no_input_dist(mocker):
num_inputs = 10
mock_bias_distribution = mocker.create_autospec(InputDistribution)
mock_bias_distribution.evaluate_pdf.return_value = np.ones((num_inputs, ))
mIS = MultiFidelityIS(limit_state=0.5,
biasing_distribution=mock_bias_distribution)
dummy_inputs = np.ones((num_inputs, 3))
dummy_outputs = np.zeros((num_inputs, ))
with pytest.raises(ValueError):
mIS.get_failure_prob_estimate(inputs=dummy_inputs,
outputs=dummy_outputs)
def test_mfis_estimate_without_distributions(mocker):
n_samples = 10
input_densities = 1 / np.array(range(100, 100 + n_samples))
bias_densities = np.ones((n_samples, ))
dummy_importance_weights = input_densities / bias_densities
mIS = MultiFidelityIS(limit_state=-0.5)
expected_mfis_estimate = np.mean(input_densities)
mfis_estimate = mIS.get_failure_prob_estimate(
inputs=np.ones((n_samples, 3)),
outputs=np.zeros((n_samples, )),
importance_weights=dummy_importance_weights)
np.testing.assert_almost_equal(expected_mfis_estimate, mfis_estimate[0])
def test_calc_importance_weights(mocker):
input_densities = np.array([.01, .1, 1])
mock_input_distribution = mocker.create_autospec(InputDistribution)
mock_input_distribution.evaluate_pdf.return_value = input_densities
biasing_dist_densities = np.array([4, 2, .5])
mock_bias_distribution = mocker.create_autospec(InputDistribution)
mock_bias_distribution.evaluate_pdf.return_value = biasing_dist_densities
importance_sampling = MultiFidelityIS(limit_state=0,
biasing_distribution= \
mock_bias_distribution,
input_distribution= \
mock_input_distribution)
expected_weights = input_densities / biasing_dist_densities
weights = importance_sampling.calc_importance_weights(np.ones((3, 1)))
np.testing.assert_array_almost_equal(expected_weights, weights)
def test_mfis_estimate_is_correct(mocker):
n_samples = 10
weights = 1 / np.array(range(100, 100 + n_samples))
input_densities = weights
mock_input_distribution = mocker.create_autospec(InputDistribution)
mock_input_distribution.evaluate_pdf.return_value = input_densities
biasing_dist_densities = np.ones((n_samples, ))
mock_bias_distribution = mocker.create_autospec(InputDistribution)
mock_bias_distribution.evaluate_pdf.return_value = biasing_dist_densities
mIS = MultiFidelityIS(limit_state=-0.5,
biasing_distribution=mock_bias_distribution,
input_distribution=mock_input_distribution)
expected_mfis_estimate = np.sum(np.delete(weights, 0)) / n_samples
dummy_inputs = np.ones((n_samples, 3))
dummy_outputs = np.concatenate(
[-np.ones((1, )), np.zeros((n_samples - 1, ))])
mfis_estimate = mIS.get_failure_prob_estimate(dummy_inputs, dummy_outputs)
np.testing.assert_almost_equal(expected_mfis_estimate, mfis_estimate[0])
def test_calc_importance_weights_no_dists_raises_error(mocker):
importance_sampling = MultiFidelityIS(limit_state=0)
dummy_inputs = np.ones((10, 3))
with pytest.raises(ValueError):
weights = importance_sampling.calc_importance_weights(dummy_inputs)