def test_reversible_pi(dtype):
C = np.array([[7048, 6, 0], [6, 2, 3], [0, 3, 2933]]).astype(dtype)
pi = np.array([0.70532947, 0.00109989, 0.29357064])
P_mle = tmatrix(C, reversible=True, mu=pi)
N = 1000
sampler = TransitionMatrixSampler(C, reversible=True, mu=pi, n_steps=10)
sample = np.zeros((N, 3, 3))
for i in range(N):
s = sampler.sample()
np.testing.assert_equal(s.dtype, dtype)
sample[i, :, :] = s
mean = np.mean(sample, axis=0)
std = np.std(sample, axis=0)
# Check if sample mean and MLE agree within the sample standard deviation
np.testing.assert_(np.all(np.abs(mean - P_mle) <= std))
def test_reversible(dtype):
C = 1.0 * np.array([[7048, 6, 0], [6, 2, 3], [0, 3, 2933]]).astype(dtype)
P_mle = tmatrix(C, reversible=True)
N = 1000
sampler = TransitionMatrixSampler(C, reversible=True)
sample = np.zeros((N, 3, 3))
for i in range(N):
s = sampler.sample()
np.testing.assert_equal(s.dtype, dtype)
sample[i, :, :] = s
mean = np.mean(sample, axis=0)
std = np.std(sample, axis=0)
# Check if sample mean and MLE agree within the sample standard deviation
diff = np.abs(mean - P_mle)
np.testing.assert_(np.all(diff <= std))
def __init__(self, C, P0=None, seed: int = -1):
from deeptime.markov.tools.estimation import transition_matrix as tmatrix
from deeptime.markov.tools.analysis import stationary_distribution
from .._mle_bindings import RevSampler32, RevSampler64, RevSampler128
if C.dtype not in (np.float32, np.float64, np.longdouble):
dtype = np.float64
else:
dtype = C.dtype
self.C = C.astype(dtype)
"""Set up initial state of the chain"""
if P0 is None:
# only do a few iterations to get close to the MLE and suppress not converged warning
P0 = tmatrix(self.C,
reversible=True,
maxiter=100,
warn_not_converged=False)
assert P0.dtype == self.C.dtype
else:
P0 = P0.astype(self.C.dtype)
pi0 = stationary_distribution(P0).astype(self.C.dtype)
V0 = pi0[:, np.newaxis] * P0
self.V = V0
self.c = self.C.sum(axis=1)
"""Check for valid input"""
self.check_input()
"""Get nonzero indices"""
self.I, self.J = np.where((self.C + self.C.T) > 0.0)
"""Init Vsampler"""
if self.C.dtype == np.float32:
self._sampler = RevSampler32(seed)
elif self.C.dtype == np.float64:
self._sampler = RevSampler64(seed)
elif self.C.dtype == np.longdouble:
self._sampler = RevSampler128(seed)
else:
raise ValueError(f"Unknown dtype {self.C.dtype}")