def test_loglike_against_gaussian(self):
mu = np.random.randn(3)
A = mu[:, None]
sigma = np.random.randn(3, 3)
sigma = sigma.dot(sigma.T)
data = np.random.randn(25, mu.shape[0])
d1 = distributions.Gaussian(mu=mu, sigma=sigma)
likes1 = d1.log_likelihood(data)
d2 = self.distribution_class(A=A, sigma=sigma)
likes2 = d2.log_likelihood(
np.hstack((np.ones((data.shape[0], 1)), data)))
assert np.allclose(likes1, likes2)
plt.interactive(True)
from pybasicbayes import models, distributions
GENERATE_DATA = True
###########################
# generate or load data #
###########################
alpha_0 = 5.0
obs_hypparams = dict(mu_0=np.zeros(2), sigma_0=np.eye(2), kappa_0=0.05, nu_0=5)
priormodel = models.Mixture(
alpha_0=alpha_0,
components=[distributions.Gaussian(**obs_hypparams) for itr in range(30)])
data, _ = priormodel.generate(100)
del priormodel
plt.figure()
plt.plot(data[:, 0], data[:, 1], 'kx')
plt.title('data')
raw_input() # pause for effect
###############
# inference #
###############
posteriormodel = models.Mixture(
alpha_0=alpha_0,
def test_empirical_bayes(self):
data = np.random.randn(50, 2)
distributions.Gaussian().empirical_bayes(data).hypparams