def test_method(self):
# Create mcmc
xs = [
self.real_parameters * 1.1,
self.real_parameters * 1.05,
self.real_parameters * 0.9,
self.real_parameters * 0.95,
]
mcmc = pints.DreamMCMC(4, xs)
self.assertFalse(mcmc.constant_crossover())
# Starts in initial phase
self.assertTrue(mcmc.needs_initial_phase())
self.assertTrue(mcmc.in_initial_phase())
# Perform short run
chains = []
for i in range(100):
xs = mcmc.ask()
fxs = np.array([self.log_posterior(x) for x in xs])
ys, fys, ac = mcmc.tell(fxs)
if i == 20:
mcmc.set_initial_phase(False)
if i >= 50:
chains.append(ys)
if np.any(ac):
self.assertTrue(np.all(xs[ac] == ys[ac]))
self.assertTrue(np.all(fys[ac] == fxs[ac]))
chains = np.array(chains)
self.assertEqual(chains.shape[0], 50)
self.assertEqual(chains.shape[1], len(xs))
self.assertEqual(chains.shape[2], len(xs[0]))
# Repeat with constant crossover
mcmc = pints.DreamMCMC(4, xs)
mcmc.set_constant_crossover(True)
self.assertTrue(mcmc.constant_crossover())
# Perform short run
chains = []
for i in range(100):
xs = mcmc.ask()
fxs = np.array([self.log_posterior(x) for x in xs])
ys, fys, ac = mcmc.tell(fxs)
if i == 20:
mcmc.set_initial_phase(False)
if i >= 50:
chains.append(ys)
chains = np.array(chains)
self.assertEqual(chains.shape[0], 50)
self.assertEqual(chains.shape[1], len(xs))
self.assertEqual(chains.shape[2], len(xs[0]))
def test_flow(self):
# Test we have at least 3 chains
n = 2
x0 = [self.real_parameters] * n
self.assertRaises(ValueError, pints.DreamMCMC, n, x0)
# Test initial proposal is first point
n = 3
x0 = [self.real_parameters] * n
mcmc = pints.DreamMCMC(n, x0)
self.assertTrue(np.all(mcmc.ask() == mcmc._x0))
# Double initialisation
mcmc = pints.DreamMCMC(n, x0)
mcmc.ask()
self.assertRaises(RuntimeError, mcmc._initialise)
# Tell without ask
mcmc = pints.DreamMCMC(n, x0)
self.assertRaises(RuntimeError, mcmc.tell, 0)
# Repeated asks should return same point
mcmc = pints.DreamMCMC(n, x0)
# Get into accepting state
for i in range(100):
mcmc.tell([self.log_posterior(x) for x in mcmc.ask()])
x = mcmc.ask()
for i in range(10):
self.assertTrue(x is mcmc.ask())
# Repeated tells should fail
mcmc.tell([1, 1, 1])
self.assertRaises(RuntimeError, mcmc.tell, [1, 1, 1])
# Bad starting point
mcmc = pints.DreamMCMC(n, x0)
mcmc.ask()
self.assertRaises(ValueError, mcmc.tell, float('-inf'))
def test_set_hyper_parameters(self):
# Tests the hyper-parameter interface for this optimiser.
n = 5
x0 = [self.real_parameters] * n
mcmc = pints.DreamMCMC(n, x0)
self.assertEqual(mcmc.n_hyper_parameters(), 8)
# Test getting/setting b
b = mcmc.b()
self.assertEqual(mcmc.b(), b)
b += 0.01
self.assertNotEqual(mcmc.b(), b)
mcmc.set_b(b)
self.assertEqual(mcmc.b(), b)
mcmc.set_b(0)
self.assertRaises(ValueError, mcmc.set_b, -1)
# B star
x = mcmc.b_star() + 1
mcmc.set_b_star(x)
self.assertEqual(mcmc.b_star(), x)
self.assertRaises(ValueError, mcmc.set_b_star, -5)
# p_g
x = mcmc.p_g() + 0.1
mcmc.set_p_g(x)
self.assertEqual(mcmc._p_g, x)
self.assertRaises(ValueError, mcmc.set_p_g, -0.5)
self.assertRaises(ValueError, mcmc.set_p_g, 1.5)
# delta max
x = mcmc.delta_max() - 1
mcmc.set_delta_max(x)
self.assertEqual(mcmc.delta_max(), x)
self.assertRaises(ValueError, mcmc.set_delta_max, -1)
self.assertRaises(ValueError, mcmc.set_delta_max, 1000)
# CR
x = mcmc.CR() * 0.9
mcmc.set_CR(x)
self.assertEqual(mcmc.CR(), x)
self.assertRaises(ValueError, mcmc.set_CR, -0.5)
self.assertRaises(ValueError, mcmc.set_CR, 1.5)
# nCR
x = mcmc.nCR() + 1
mcmc.set_nCR(x)
self.assertEqual(mcmc.nCR(), 4)
self.assertRaises(ValueError, mcmc.set_nCR, 1)
# should implicitly convert floats to int
mcmc.set_nCR(2.1)
self.assertEqual(mcmc.nCR(), 2)
mcmc.set_hyper_parameters([0.50, 1.25, 0.45, 1, 0, 1, 0.32, 5])
self.assertEqual(mcmc._b, 0.50)
self.assertEqual(mcmc._b_star, 1.25)
self.assertEqual(mcmc._p_g, 0.45)
self.assertEqual(mcmc._delta_max, 1)
self.assertEqual(mcmc._initial_phase, False)
self.assertEqual(mcmc._constant_crossover, True)
self.assertEqual(mcmc._CR, 0.32)
self.assertEqual(mcmc._nCR, 5)