def test_log_likelihood2(self):
data = np.random.randn(1000, 600)
mu = np.random.randn(600)
sigmas = np.random.uniform(1, 2, size=600)
d = distributions.DiagonalGaussian(mu=mu, sigmas=sigmas)
pdf1 = d.log_likelihood(data)
import scipy.stats as stats
pdf2 = stats.norm.logpdf(data, loc=mu, scale=np.sqrt(sigmas)).sum(1)
assert np.allclose(pdf1, pdf2)
np.seterr(invalid='raise')
from matplotlib import pyplot as plt
import copy
from pybasicbayes import models, distributions
from pybasicbayes.util.text import progprint_xrange
alpha_0 = 5.0
obs_hypparams = dict(mu_0=np.zeros(2),
alphas_0=2 * np.ones(2),
betas_0=np.ones(2),
nus_0=0.1 * np.ones(2))
priormodel = models.Mixture(alpha_0=alpha_0,
components=[
distributions.DiagonalGaussian(**obs_hypparams)
for itr in range(30)
])
data, _ = priormodel.generate(500)
data2, _ = priormodel.generate(500)
del priormodel
posteriormodel = models.Mixture(
alpha_0=alpha_0,
components=[
distributions.DiagonalGaussian(**obs_hypparams) for itr in range(30)
])
posteriormodel.add_data(data)