def test_LFM_gradient(artificial_data, models):
reg_truth = Regressor(ss=models[0])
reg_truth._use_penalty = False
reg_truth._use_jacobian = True
dt, u, u1, y, *_ = reg_truth._prepare_data(artificial_data, ['To', 'Qh'],
'Ti')
reg_lfm = Regressor(ss=models[1])
reg_lfm._use_penalty = False
reg_lfm._use_jacobian = True
eta_truth = deepcopy(reg_truth.ss.parameters.eta_free)
eta_lfm = deepcopy(reg_lfm.ss.parameters.eta_free)
grad_truth = reg_truth._eval_dlog_posterior(eta_truth, dt, u, u1, y)[1]
grad_lfm = reg_lfm._eval_dlog_posterior(eta_lfm, dt, u, u1, y)[1]
fct = nd.Gradient(reg_truth._eval_log_posterior)
grad_truth_approx = fct(eta_truth, dt, u, u1, y)
assert np.all(eta_truth == eta_lfm)
assert ned(grad_truth, grad_lfm) < 1e-7
assert ned(grad_truth, grad_truth_approx) < 1e-7
assert np.all(np.sign(grad_truth) == np.sign(grad_truth_approx))
assert np.all(np.sign(grad_truth) == np.sign(grad_lfm))
assert grad_truth == pytest.approx(grad_truth_approx, rel=1e-6)
assert grad_truth == pytest.approx(grad_lfm, rel=1e-6)
def check_grad_fd(data, reg):
reg._use_penalty = False
reg._use_jacobian = True
reg.ss.method = 'mfd'
dt, u, u1, y, _ = reg._prepare_data(data, None, 'y')
grad = reg._eval_dlog_posterior(reg.ss.parameters.eta_free, dt, u, u1,
y)[1]
grad_fct = nd.Gradient(reg._eval_log_posterior)
grad_fd = grad_fct(reg.ss.parameters.eta_free, dt, u, u1, y)
assert ned(grad, grad_fd) < 1e-7
assert np.all(np.sign(grad) == np.sign(grad_fd))
assert grad == pytest.approx(grad_fd, rel=1e-6)
def test_gradient_RCModel(artificial_data_rc, reg, inputs, outputs):
"""Compare regressor gradient with numerical differentiation"""
reg._use_penalty = False
reg._use_jacobian = True
dt, u, u1, y, *_ = reg._prepare_data(artificial_data_rc, inputs, outputs)
eta = reg.ss.parameters.eta_free.copy()
grad = reg._eval_dlog_posterior(eta, dt, u, u1, y)[1]
fct = nd.Gradient(reg._eval_log_posterior)
grad_fd = fct(eta, dt, u, u1, y)
assert ned(grad, grad_fd) < 1e-7
assert np.all(np.sign(grad) == np.sign(grad_fd))
assert grad == pytest.approx(grad_fd, rel=1e-6)
def check_grad(data, reg1, reg2):
reg1._use_penalty = False
reg1._use_jacobian = True
reg2._use_penalty = False
reg2._use_jacobian = True
reg1.ss.method = 'mfd'
reg2.ss.method = 'mfd'
dt, u, u1, y, _ = reg1._prepare_data(data, None, 'y')
grad1 = reg1._eval_dlog_posterior(reg1.ss.parameters.eta_free, dt, u, u1,
y)[1]
grad2 = reg2._eval_dlog_posterior(reg2.ss.parameters.eta_free, dt, u, u1,
y)[1]
assert ned(grad1, grad2) < 1e-7
assert np.all(np.sign(grad1) == np.sign(grad2))
assert grad1 == pytest.approx(grad2, rel=1e-6)