def _resample_sigma(self, stats):
ysq, yxT, xxT, n = stats
AAT = np.array([np.outer(a,a) for a in self.A])
alpha = self.alpha_0 + n / 2.0
beta = self.beta_0
beta += 0.5 * ysq
beta += -1.0 * np.sum(yxT * self.A, axis=1)
beta += 0.5 * np.sum(AAT * xxT, axis=(1,2))
self.sigmasq_flat = np.reshape(sample_invgamma(alpha, beta), (self.D_out,))
def _resample_eta(self, datas):
N, B = self.N, self.B
alpha = self.a_0
beta = self.b_0
for data in datas:
X, y = self.extract_data(data)
T = X.shape[0]
alpha += T / 2.0
beta += np.sum((y-self.mean(X))**2)
self.eta = sample_invgamma(alpha, beta)
def _resample_eta(self, datas):
N, B = self.N, self.B
alpha = self.a_0
beta = self.b_0
for data in datas:
X, y = self.extract_data(data)
T = X.shape[0]
alpha += T / 2.0
beta += np.sum((y - self.mean(X))**2)
self.eta = sample_invgamma(alpha, beta)
def __init__(self, N, B, a_0=2.0, b_0=2.0, eta=None, **kwargs):
super(SparseGaussianRegression, self).__init__(N, B, **kwargs)
# Initialize the noise model
assert np.isscalar(a_0) and a_0 > 0
assert np.isscalar(b_0) and a_0 > 0
self.a_0, self.b_0 = a_0, b_0
if eta is not None:
assert np.isscalar(eta) and eta > 0
self.eta = eta
else:
# Sample eta from its inverse gamma prior
self.eta = sample_invgamma(self.a_0, self.b_0)
def __init__(self, N, B,
a_0=2.0, b_0=2.0, eta=None,
**kwargs):
super(SparseGaussianRegression, self).__init__(N, B, **kwargs)
# Initialize the noise model
assert np.isscalar(a_0) and a_0 > 0
assert np.isscalar(b_0) and a_0 > 0
self.a_0, self.b_0 = a_0, b_0
if eta is not None:
assert np.isscalar(eta) and eta > 0
self.eta = eta
else:
# Sample eta from its inverse gamma prior
self.eta = sample_invgamma(self.a_0, self.b_0)
def resample_from_mf(self):
for d in range(self.D_out):
self.A[d] = sample_gaussian(J=self.mf_J_A[d], h=self.mf_h_A[d])
self.sigmasq_flat = sample_invgamma(self.mf_alpha,
self.mf_beta) * np.ones(self.D_out)
def resample_from_mf(self):
for d in range(self.D_out):
self.A[d] = sample_gaussian(J=self.mf_J_A[d], h=self.mf_h_A[d])
self.sigmasq_flat = sample_invgamma(self.mf_alpha, self.mf_beta) * np.ones(self.D_out)
