def test_PC_nd(self, mascaret_data):
space_ = copy.deepcopy(mascaret_data.space)
space_.max_points_nb = 1000
sample = space_.sampling(1000, 'halton')
surrogate = PC(distributions=mascaret_data.dists, sample=sample, degree=10,
strategy='LS')
input_ = surrogate.sample
output = mascaret_data.func(input_)
surrogate.fit(input_, output)
pred = np.array(surrogate.evaluate(mascaret_data.point)).reshape(-1)
npt.assert_almost_equal(mascaret_data.target_point, pred, decimal=1)
surrogate = PC(distributions=mascaret_data.dists, degree=5,
strategy='Quad')
input_ = surrogate.sample
output = mascaret_data.func(input_)
surrogate.fit(input_, output)
# Compute predictivity coefficient Q2
q2 = sklearn_q2(mascaret_data.dists, mascaret_data.func, surrogate.evaluate)
assert q2 == pytest.approx(1, 0.1)
surrogate = PC(distributions=mascaret_data.dists, degree=10,
sample=sample,
strategy='SparseLS')
input_ = surrogate.sample
output = mascaret_data.func(input_)
surrogate.fit(input_, output)
# Compute predictivity coefficient Q2
q2 = sklearn_q2(mascaret_data.dists, mascaret_data.func, surrogate.evaluate)
assert q2 == pytest.approx(1, 0.1)
def test_GP_1d(self, ishigami_data):
surrogate = Kriging(ishigami_data.space, ishigami_data.target_space)
# Test one point evaluation
pred, _ = np.array(surrogate.evaluate(ishigami_data.point))
assert pred == pytest.approx(ishigami_data.target_point, 0.2)
# Compute predictivity coefficient Q2
def wrap_surrogate(x):
evaluation, _ = surrogate.evaluate(x)
return evaluation
q2 = sklearn_q2(ishigami_data.dists, ishigami_data.func, wrap_surrogate)
assert q2 == pytest.approx(1, 0.1)
# Kernel and noise
surrogate = Kriging(ishigami_data.space, ishigami_data.target_space,
kernel=Matern(), noise=0.8)
surrogate = Kriging(ishigami_data.space, ishigami_data.target_space,
kernel=Matern(), noise=True)
# Optimizer
surrogate = Kriging(ishigami_data.space, ishigami_data.target_space,
global_optimizer=False)
pred, _ = np.array(surrogate.evaluate(ishigami_data.point))
assert pred == pytest.approx(ishigami_data.target_point, 0.2)
def test_sk_regressors_nd(self, mascaret_data):
regressor = 'RandomForestRegressor()'
surrogate = SklearnRegressor(mascaret_data.space, mascaret_data.target_space,
regressor)
q2 = sklearn_q2(mascaret_data.dists, mascaret_data.func, surrogate.evaluate)
assert q2 == pytest.approx(0.77, 0.2)
def test_PC_1d(self, ishigami_data):
space_ = copy.deepcopy(ishigami_data.space)
space_.max_points_nb = 2000
sample = space_.sampling(2000, 'halton')
surrogate = PC(distributions=ishigami_data.dists, sample=sample, degree=10,
strategy='LS', stieltjes=False)
input_ = surrogate.sample
assert len(input_) == 2000
output = ishigami_data.func(input_)
surrogate.fit(input_, output)
pred = np.array(surrogate.evaluate(ishigami_data.point))
assert pred == pytest.approx(ishigami_data.target_point, 0.1)
# Compute predictivity coefficient Q2
q2 = sklearn_q2(ishigami_data.dists, ishigami_data.func, surrogate.evaluate)
assert q2 == pytest.approx(1, 0.1)
surrogate = PC(distributions=ishigami_data.dists, degree=10,
strategy='Quad')
input_ = surrogate.sample
assert len(input_) == 1331
output = ishigami_data.func(input_)
surrogate.fit(input_, output)
# Compute predictivity coefficient Q2
q2 = sklearn_q2(ishigami_data.dists, ishigami_data.func, surrogate.evaluate)
assert q2 == pytest.approx(1, 0.2)
surrogate = PC(distributions=ishigami_data.dists, degree=10, sample=sample,
sparse_param={"max_considered_terms": 130,
"most_significant": 30,
"significance_factor": 5e-5},
strategy='SparseLS')
input_ = surrogate.sample
assert len(input_) == 2000
output = ishigami_data.func(input_)
surrogate.fit(input_, output)
# Compute predictivity coefficient Q2
q2 = sklearn_q2(ishigami_data.dists, ishigami_data.func, surrogate.evaluate)
assert q2 == pytest.approx(1, 0.2)
def test_evofusion(mufi_data):
f_e, f_c = mufi_data.func
surrogate = Evofusion(mufi_data.space, mufi_data.target_space)
# Test one point evaluation
pred, _ = np.array(surrogate.evaluate(mufi_data.point))
assert pred == pytest.approx(mufi_data.target_point, 0.1)
# Compute predictivity coefficient Q2
def wrap_surrogate(x):
evaluation, _ = surrogate.evaluate(x)
return evaluation
q2 = sklearn_q2(mufi_data.dists, f_e, wrap_surrogate)
assert q2 == pytest.approx(1, 0.1)
def test_no_pod(ishigami_data, tmp, settings_ishigami):
test_settings = copy.deepcopy(settings_ishigami)
test_settings.pop('pod')
driver = Driver(test_settings, tmp)
driver.sampling()
pred, _ = driver.prediction(write=True, points=ishigami_data.point)
assert pred == pytest.approx(ishigami_data.target_point, 0.2)
assert os.path.isdir(os.path.join(tmp, 'predictions'))
def wrap_surrogate(x):
evaluation, _ = driver.prediction(points=x)
return evaluation
q2 = sklearn_q2(ishigami_data.dists, ishigami_data.func, wrap_surrogate)
assert q2 == pytest.approx(1, 0.1)
def test_GP_nd(self, mascaret_data):
# Scaling as class SurrogateModel does it
scaler = preprocessing.MinMaxScaler()
scaler.fit(np.array(mascaret_data.space.corners))
space_scaled = scaler.transform(mascaret_data.space)
surrogate = Kriging(space_scaled, mascaret_data.target_space,
kernel=Matern())
surrogate = Kriging(space_scaled, mascaret_data.target_space)
# Compute predictivity coefficient Q2
def wrap_surrogate(x):
x_scaled = scaler.transform(x)
evaluation, _ = surrogate.evaluate(x_scaled)
return evaluation
q2 = sklearn_q2(mascaret_data.dists, mascaret_data.func, wrap_surrogate)
assert q2 == pytest.approx(1, 0.1)
def test_RBFnet_1d(self, ishigami_data):
surrogate = RBFnet(ishigami_data.space, ishigami_data.target_space)
# Test one point evaluation
pred = np.array(surrogate.evaluate(ishigami_data.point))
assert pred == pytest.approx(ishigami_data.target_point, 0.3)
# Test space evaluation
pred = np.array(surrogate.evaluate(ishigami_data.space))
npt.assert_almost_equal(ishigami_data.target_space, pred, decimal=1)
# Compute predictivity coefficient Q2
def wrap_surrogate(x):
evaluation = surrogate.evaluate(x)
return evaluation
q2 = sklearn_q2(ishigami_data.dists, ishigami_data.func, wrap_surrogate)
assert q2 == pytest.approx(0.86, 0.1)
surrogate = RBFnet(ishigami_data.space, ishigami_data.target_space, regtree=1)
def test_sk_regressors_1d(self, ishigami_data):
# From an object
l_scale = (1.0,) * 3
scale_bounds = [(0.01, 100)] * 3
kernel = ConstantKernel() * Matern(length_scale=l_scale,
length_scale_bounds=scale_bounds)
regressor = GaussianProcessRegressor(kernel=kernel, normalize_y=True)
surrogate = SklearnRegressor(ishigami_data.space, ishigami_data.target_space,
regressor)
# Test space evaluation
pred = np.array(surrogate.evaluate(ishigami_data.space))
npt.assert_almost_equal(ishigami_data.target_space, pred, decimal=1)
# From a str
regressor = 'RandomForestRegressor()'
surrogate = SklearnRegressor(ishigami_data.space, ishigami_data.target_space,
regressor)
q2 = sklearn_q2(ishigami_data.dists, ishigami_data.func, surrogate.evaluate)
assert q2 == pytest.approx(0.77, 0.2)