def test_ln_likelihood(self):
test_rel = relationship.Relationship(utils.straight_line, TEST_X,
TEST_Y)
expected_lnl = -3
actual_lnl = optimize.ln_likelihood([1., 0.], test_rel.function,
test_rel.x, test_rel.y)
assert_almost_equal(actual_lnl, expected_lnl, decimal=0)
def bayesian_information_criteria(self):
"""
Calculate the `Bayesian information criteria`_ for the relationship.
Returns:
:py:attr:`float`: Bayesian information criteria.
.. _Bayesian information criteria: https://en.wikipedia.org/wiki/Bayesian_information_criterion
"""
self.max_likelihood('diff_evo')
l_hat = optimize.ln_likelihood(self.variable_medians, self.function, self.abscissa, self.ordinate)
return np.log(self.x.size) * self.len_parameters - 2 * l_hat
def ln_probability(variables, function, abscissa, ordinate, priors):
"""
Determine the natural log probability for a given set of variables, by
summing the prior and likelihood.
Args:
variables (:py:attr:`array_like`): Variables for the function evaluation.
function (:py:attr:`callable`): The function to be evaluated.
abscissa (:py:attr:`array_like`): The abscissa values.
ordinate (:py:attr:`array_like`): The ordinate values.
unaccounted_uncertainty (:py:attr:`bool`): Should an unaccounted uncertainty parameter be considered.
prior_function (:py:attr:`callable`, optional): The function to populated some prior distributions. Default is :func:`~uravu.relationship.Relationship.prior()`.
Returns:
:py:attr:`float`: Negative natural log-probability between model and data, considering priors.
"""
log_prior = 0
for i, var in enumerate(variables):
log_prior += priors[i].logpdf(var)
lnl = optimize.ln_likelihood(variables, function, abscissa, ordinate)
return log_prior + lnl