def check_model_gradients(model):
model = set(model)
# find the markov blanket
children = set([])
for s in model:
for s2 in s.extended_children:
if isinstance(s2, Stochastic) and s2.observed:
children.add(s2)
# self.markov_blanket is a list, because we want self.stochastics to have the chance to
# raise ZeroProbability exceptions before self.children.
markov_blanket = list(model) + list(children)
gradients = utils.logp_gradient_of_set(model)
for variable in model:
analytic_gradient = gradients[variable]
numeric_gradient = get_numeric_gradient(markov_blanket, variable)
assert_array_almost_equal(
numeric_gradient, analytic_gradient, 3,
"analytic gradient for model " + str(model) +
" with respect to variable " + str(variable) + " is not correct.")
def check_model_gradients(model):
model = set(model)
# find the markov blanket
children = set([])
for s in model:
for s2 in s.extended_children:
if isinstance(s2, Stochastic) and s2.observed:
children.add(s2)
# self.markov_blanket is a list, because we want self.stochastics to have the chance to
# raise ZeroProbability exceptions before self.children.
markov_blanket = list(model) + list(children)
gradients = utils.logp_gradient_of_set(model)
for variable in model:
analytic_gradient = gradients[variable]
numeric_gradient = get_numeric_gradient(markov_blanket, variable)
assert_array_almost_equal(numeric_gradient, analytic_gradient, 3,
"analytic gradient for model " + str(model) +
" with respect to variable " + str(variable) +
" is not correct.")
def check_gradients(stochastic):
stochastics = find_variable_set(stochastic)
gradients = utils.logp_gradient_of_set(stochastics, stochastics)
for s, analytic_gradient in six.iteritems(gradients):
numeric_gradient = get_numeric_gradient(stochastics, s)
assert_array_almost_equal(
numeric_gradient, analytic_gradient, 3,
"analytic gradient for " + str(stochastic) +
" with respect to parameter " + str(s) + " is not correct.")
def check_gradients(stochastic):
stochastics = find_variable_set(stochastic)
gradients = utils.logp_gradient_of_set(stochastics, stochastics)
for s, analytic_gradient in six.iteritems(gradients):
numeric_gradient = get_numeric_gradient(stochastics, s)
assert_array_almost_equal(numeric_gradient, analytic_gradient, 3,
"analytic gradient for " + str(stochastic) +
" with respect to parameter " + str(s) +
" is not correct.")
