def test_theta_computed_boolean(self):
theta_init = self.random_theta(2, 1)
mcmc = KurslMCMC(theta_init, niter=2)
self.assertFalse(mcmc._theta_computed(), "Without run should fail")
mcmc.set_sampler(np.linspace(0, 1, 100), np.random.random(100))
mcmc.run()
self.assertTrue(mcmc._theta_computed(), "After run should return true")
def test_run_default(self):
_cos = lambda l: l[2] * np.cos(l[0] * t + l[1])
theta = [
[15, 0, 1, 0],
[35, 2, 3, 0],
]
theta_std = [
[1.0, 0.1, 0.8, 0.01],
[1.5, 0.1, 1.5, 0.01],
]
theta, theta_std = np.array(theta), np.array(theta_std)
t = np.linspace(0, 1, 100)
c1 = _cos(theta[0])
c2 = _cos(theta[1])
S = c1 + c2
theta_init = np.array(theta, dtype=np.float64)
theta_init[:, 0] += 2 - np.random.random(2) * 0.5
theta_init[:, 2] += 1 - np.random.random(2) * 0.5
mcmc = KurslMCMC(theta_init, theta_std, nwalkers=40, niter=200)
mcmc.set_threshold(0.001)
mcmc.set_sampler(t, S)
mcmc.run()
# After simulation is finished check for correctness
theta_computed = mcmc.get_theta()
self.assertTrue(
np.allclose(theta, theta_computed, atol=0.5),
"Expected:\n{}\nReceived:\n{}".format(theta, theta_computed))
def test_run_start_from_solution(self):
_cos = lambda l: l[2] * np.cos(l[0] * t + l[1])
theta = np.array([
[15, 0, 1, 0],
[35, 2, 3, 0],
], dtype=np.float64)
t = np.linspace(0, 1, 100)
c1 = _cos(theta[0])
c2 = _cos(theta[1])
S = c1 + c2
# Initial guess is far off. Just for an example.
theta_init = np.array(theta, dtype=np.float64)
theta_init += np.random.random(theta_init.shape) * 3
mcmc = KurslMCMC(theta_init)
mcmc.set_sampler(t, S)
# However, execution starts close to solution.
pos = np.tile(theta.flatten(), (mcmc.nwalkers, 1))
pos += np.random.random(pos.shape) * 0.3
mcmc.run(pos=pos)
niter_executed = mcmc.get_lnprob().shape[0]
self.assertTrue(
niter_executed < mcmc.niter,
"Starting close solution should allow for quick convergence "
"and not all iterations would be executed.")
def test_run_with_incorrect_pos(self):
t = np.linspace(0, 1, 100)
S = np.random.random(t.size)
mcmc = KurslMCMC(self.random_theta(3, 2))
mcmc.set_sampler(t, S)
with self.assertRaises(ValueError) as error:
mcmc.run(pos=self.random_theta(3, 2))
self.assertTrue("pos.shape" in str(error.exception))
def test_set_sampler(self):
t = np.linspace(0, 1, 100)
S = np.random.random(t.size)
s = S[:-1]
s_var = np.sum((s - s.mean()) * (s - s.mean()))
theta_init = self.random_theta(3, 2)
mcmc = KurslMCMC(theta_init)
self.assertEqual(mcmc.model.s_var, 1, "Default var is 1")
self.assertTrue(
mcmc.sampler is None,
"Without explicit assignment there should be no sampler.")
mcmc.set_sampler(t, S)
self.assertEqual(
type(mcmc.sampler).__name__, "EnsembleSampler",
"Executing `set_sampler` should create EnsembleSampler `sampler`")
self.assertEqual(mcmc.model.s_var, s_var, "Updated var for the model")