def test_fit_multiple_trajectores(data_multiple_trajctories):
x, t = data_multiple_trajctories
u = [np.ones((xi.shape[0], 2)) for xi in x]
model = SINDy()
# Should fail if multiple_trajectories flag is not set
with pytest.raises(ValueError):
model.fit(x, u=u, t=t)
# Should fail if either x or u is not a list
with pytest.raises(ValueError):
model.fit(x, u=u[0], multiple_trajectories=True)
with pytest.raises(ValueError):
model.fit(x[0], u=u, multiple_trajectories=True)
# x and u should be lists of the same length
with pytest.raises(ValueError):
model.fit([x[:-1]], u=u, multiple_trajectories=True)
model.fit(x, u=u, multiple_trajectories=True)
check_is_fitted(model)
model.fit(x, u=u, t=t, multiple_trajectories=True)
assert model.score(x, u=u, t=t, multiple_trajectories=True) > 0.8
model = SINDy()
model.fit(x, u=u, x_dot=x, multiple_trajectories=True)
check_is_fitted(model)
model = SINDy()
model.fit(x, u=u, t=t, x_dot=x, multiple_trajectories=True)
check_is_fitted(model)
def test_fit_multiple_trajectores(data_multiple_trajctories):
x, t = data_multiple_trajctories
model = SINDy()
# Should fail if multiple_trajectories flag is not set
with pytest.raises(ValueError):
model.fit(x, t=t)
model.fit(x, multiple_trajectories=True)
check_is_fitted(model)
model.fit(x, t=t, multiple_trajectories=True)
assert model.score(x, t=t, multiple_trajectories=True) > 0.8
model = SINDy()
model.fit(x, x_dot=x, multiple_trajectories=True)
check_is_fitted(model)
model = SINDy()
model.fit(x, t=t, x_dot=x, multiple_trajectories=True)
check_is_fitted(model)
# Test validate_input
t[0] = None
with pytest.raises(ValueError):
model.fit(x, t=t, multiple_trajectories=True)
def test_fit_discrete_time(data_discrete_time_c):
x, u = data_discrete_time_c
model = SINDy(discrete_time=True)
model.fit(x, u=u)
check_is_fitted(model)
model = SINDy(discrete_time=True)
model.fit(x[:-1], u=u[:-1], x_dot=x[1:])
check_is_fitted(model)
def test_fit_discrete_time(data_discrete_time):
x = data_discrete_time
model = SINDy(discrete_time=True)
model.fit(x)
check_is_fitted(model)
model = SINDy(discrete_time=True)
model.fit(x[:-1], x_dot=x[1:])
check_is_fitted(model)
def test_simulate_errors(data_lorenz):
x, t = data_lorenz
model = SINDy()
model.fit(x, t)
with pytest.raises(ValueError):
model.simulate(x[0], t=1)
model = SINDy(discrete_time=True)
with pytest.raises(ValueError):
model.simulate(x[0], t=[1, 2])
def test_simulate_errors(data_lorenz_c_1d):
x, t, u, u_fun = data_lorenz_c_1d
model = SINDy()
model.fit(x, u=u, t=t)
with pytest.raises(ValueError):
model.simulate(x[0], t=1, u=u)
model = SINDy(discrete_time=True)
with pytest.raises(ValueError):
model.simulate(x[0], t=[1, 2], u=u)
def test_differentiate(data_lorenz, data_multiple_trajctories):
x, t = data_lorenz
model = SINDy()
model.differentiate(x, t)
x, t = data_multiple_trajctories
model.differentiate(x, t, multiple_trajectories=True)
model = SINDy(discrete_time=True)
with pytest.raises(RuntimeError):
model.differentiate(x)
def test_t_default(data):
x, t, u, _ = data
dt = t[1] - t[0]
model = SINDy()
model.fit(x, u=u, t=dt)
model_t_default = SINDy(t_default=dt)
model_t_default.fit(x, u=u)
np.testing.assert_allclose(model.coefficients(),
model_t_default.coefficients())
np.testing.assert_almost_equal(model.score(x, u=u, t=dt),
model_t_default.score(x, u=u))
def test_improper_shape_input(data_1d):
x, t = data_1d
# Ensure model successfully handles different data shapes
model = SINDy()
model.fit(x.flatten(), t)
check_is_fitted(model)
model = SINDy()
model.fit(x.flatten(), t, x_dot=x.flatten())
check_is_fitted(model)
model = SINDy()
model.fit(x, t, x_dot=x.flatten())
check_is_fitted(model)
def test_fit_discrete_time_multiple_trajectories(
data_discrete_time_multiple_trajectories, ):
x = data_discrete_time_multiple_trajectories
# Should fail if multiple_trajectories flag is not set
model = SINDy(discrete_time=True)
with pytest.raises(ValueError):
model.fit(x)
model.fit(x, multiple_trajectories=True)
check_is_fitted(model)
model = SINDy(discrete_time=True)
model.fit(x, x_dot=x, multiple_trajectories=True)
check_is_fitted(model)
def test_cross_validation(data_lorenz):
x, t = data_lorenz
dt = t[1] - t[0]
model = SINDy(t_default=dt,
differentiation_method=SINDyDerivative(kind="spline",
s=1e-2))
param_grid = {
"optimizer__threshold": [0.01, 0.1],
"differentiation_method__kwargs": [
{
"kind": "spline",
"s": 1e-2
},
{
"kind": "finite_difference",
"k": 1
},
],
"feature_library__degree": [1, 2],
}
search = RandomizedSearchCV(model,
param_grid,
cv=TimeSeriesSplit(n_splits=3),
n_iter=5)
search.fit(x)
check_is_fitted(search)
def test_integration_smoothed_finite_difference(data_lorenz):
x, t = data_lorenz
model = SINDy(differentiation_method=SmoothedFiniteDifference())
model.fit(x, t=t)
check_is_fitted(model)
def test_simulate(data):
x, t, u, u_fun = data
model = SINDy()
model.fit(x, u=u, t=t)
x1 = model.simulate(x[0], t=t, u=u_fun)
assert len(x1) == len(t)
def test_multiple_trajectories_errors(data_multiple_trajctories, data_discrete_time):
x, t = data_multiple_trajctories
model = SINDy()
with pytest.raises(TypeError):
model._process_multiple_trajectories(np.array(x, dtype=object), t, x)
with pytest.raises(TypeError):
model._process_multiple_trajectories(x, t, np.array(x, dtype=object))
# Test an option that doesn't get tested elsewhere
model._process_multiple_trajectories(x, t, x, return_array=False)
x = data_discrete_time
model = SINDy(discrete_time=True)
with pytest.raises(TypeError):
model._process_multiple_trajectories(x, t, np.array(x, dtype=object))
def test_bad_t(data):
x, t = data
model = SINDy()
# No t
with pytest.raises(ValueError):
model.fit(x, t=None)
# Invalid value of t
with pytest.raises(ValueError):
model.fit(x, t=-1)
# t is a list
with pytest.raises(ValueError):
model.fit(x, list(t))
# Wrong number of time points
with pytest.raises(ValueError):
model.fit(x, t[:-1])
# Two points in t out of order
t[2], t[4] = t[4], t[2]
with pytest.raises(ValueError):
model.fit(x, t)
t[2], t[4] = t[4], t[2]
# Two matching times in t
t[3] = t[5]
with pytest.raises(ValueError):
model.fit(x, t)
def test_trapping_inequality_constraints(thresholds, relax_optim,
noise_levels):
t = np.arange(0, 40, 0.05)
x = odeint(lorenz, [-8, 8, 27], t)
x = x + np.random.normal(0.0, noise_levels, x.shape)
# if order is "feature"
constraint_rhs = np.array([-10.0, -2.0])
constraint_matrix = np.zeros((2, 30))
constraint_matrix[0, 6] = 1.0
constraint_matrix[1, 17] = 1.0
feature_names = ["x", "y", "z"]
opt = TrappingSR3(
threshold=thresholds,
constraint_lhs=constraint_matrix,
constraint_rhs=constraint_rhs,
constraint_order="feature",
inequality_constraints=True,
relax_optim=relax_optim,
)
poly_lib = PolynomialLibrary(degree=2)
model = SINDy(
optimizer=opt,
feature_library=poly_lib,
differentiation_method=FiniteDifference(drop_endpoints=True),
feature_names=feature_names,
)
model.fit(x, t=t[1] - t[0])
assert np.all(
np.dot(constraint_matrix, (model.coefficients()).flatten("F")) <=
constraint_rhs) or np.allclose(
np.dot(constraint_matrix,
(model.coefficients()).flatten("F")), constraint_rhs)
def test_libraries(data_lorenz, library):
x, t = data_lorenz
model = SINDy(feature_library=library)
model.fit(x, t)
s = model.score(x, t)
assert s <= 1
def test_mixed_inputs(data):
x, t = data
# Scalar t
model = SINDy()
model.fit(x, t=2)
check_is_fitted(model)
# x_dot is passed in
model = SINDy()
model.fit(x, x_dot=x)
check_is_fitted(model)
model = SINDy()
model.fit(x, t, x_dot=x)
check_is_fitted(model)
def test_simulate(data):
x, t = data
model = SINDy()
model.fit(x, t)
x1 = model.simulate(x[0], t)
assert len(x1) == len(t)
def test_predict(data, optimizer):
x, t = data
model = SINDy(optimizer=optimizer)
model.fit(x, t)
x_dot = model.predict(x)
assert x.shape == x_dot.shape
def test_simulate_discrete_time(data_discrete_time_c):
x, u = data_discrete_time_c
model = SINDy(discrete_time=True)
model.fit(x, u=u)
n_steps = x.shape[0]
x1 = model.simulate(x[0], t=n_steps, u=u)
assert len(x1) == n_steps
def test_print_discrete_time(data_discrete_time, capsys):
x = data_discrete_time
model = SINDy(discrete_time=True)
model.fit(x)
model.print()
out, _ = capsys.readouterr()
assert len(out) > 0
def test_get_feature_names_len(data_lorenz_c_1d):
x, t, u, _ = data_lorenz_c_1d
model = SINDy()
model.fit(x, u=u, t=t)
# Assumes default library is polynomial features of degree 2
assert len(model.get_feature_names()) == 15
def test_simulate_discrete_time(data_discrete_time):
x = data_discrete_time
model = SINDy(discrete_time=True)
model.fit(x)
n_steps = x.shape[0]
x1 = model.simulate(x[0], n_steps)
assert len(x1) == n_steps
def test_not_fitted(data_lorenz_c_1d):
x, t, u, u_fun = data_lorenz_c_1d
model = SINDy()
with pytest.raises(NotFittedError):
model.predict(x, u=u)
with pytest.raises(NotFittedError):
model.simulate(x[0], t=t, u=u_fun)
def test_parallel(data_lorenz):
x, t = data_lorenz
model = SINDy(n_jobs=4)
model.fit(x, t)
x_dot = model.predict(x)
s = model.score(x, x_dot=x_dot)
assert s >= 0.95
def test_parallel(data_lorenz_c_1d):
x, t, u, _ = data_lorenz_c_1d
model = SINDy(n_jobs=4)
model.fit(x, u=u, t=t)
x_dot = model.predict(x, u=u)
s = model.score(x, u=u, x_dot=x_dot)
assert s >= 0.95
def test_simulate_with_vector_control_input(data):
x, t, u, _ = data
model = SINDy()
model.fit(x, u=u, t=t)
x1 = model.simulate(x[0], t=t, u=u)
assert len(x1) == len(t) - 1
def test_simulate_with_interp(data):
x, t, u, _ = data
model = SINDy()
model.fit(x, u=u, t=t)
u_fun = interp1d(t, u, axis=0)
x1 = model.simulate(x[0], t=t[:-1], u=u_fun)
assert len(x1) == len(t) - 1
def test_model_sindy_equivalence(data_lorenz_c_1d):
x, t, u, _ = data_lorenz_c_1d
model = SINDyEstimator().fit(x, t=t, u=u).fetch_model()
sindy_model = SINDy().fit(x, t=t, u=u)
assert_allclose(model.coefficients(), sindy_model.coefficients())
assert model.n_input_features_ == sindy_model.n_input_features_
assert model.n_output_features_ == sindy_model.n_output_features_
assert model.n_control_features_ == sindy_model.n_control_features_