def ask(self, query, bayes_net):
evidence_vars = dict(query.evidence)
evidence_vars[query.query_variable] = True
true_probability = self._enumerate_all(bayes_net, bayes_net.get_variables(), evidence_vars)
evidence_vars[query.query_variable] = False
false_probability = self._enumerate_all(bayes_net, bayes_net.get_variables(), evidence_vars)
return normalize((true_probability, false_probability))
def rejection_sample(self, x, evidence, number_of_samples, randomizer=StandardRandomizer()):
true_values = 0
false_values = 0
for i in range(0, number_of_samples):
sample = self.get_prior_sample(randomizer)
if self._consistent(sample, evidence):
query_value = sample[x]
if query_value:
true_values += 1
else:
false_values += 1
return normalize([true_values, false_values])
def ask(self, query, probability_distribution):
"""
:param query (Query):
:param probability_distribution (ProbabilityDistribution):
:return (float, float):
"""
h = dict(query.evidence)
h[query.query_variable] = True
true_distr = probability_distribution.probability_of(h)
h[query.query_variable] = False
false_distr = probability_distribution.probability_of(h)
return normalize((true_distr, false_distr))
def likelihood_weighting(self, var, evidence, number_of_samples, randomizer=StandardRandomizer()):
true_probability = 0
false_probability = 0
for i in range(number_of_samples):
x = {}
w = 1
for node in self._get_variable_nodes():
if evidence.get(node.variable) != None:
w *= node.probability_of(x)
x[node.variable] = evidence[node.variable]
else:
x[node.variable] = node.is_true_for(randomizer.next_double(), x)
query_value = x[var]
if query_value:
true_probability += w
else:
false_probability += w
return normalize([true_probability, false_probability])
def mcmc_ask(self, var, evidence, number_of_values, randomizer=StandardRandomizer()):
true_values = 0
false_values = 0
non_evidence_variables = self._non_evidence_variables(evidence)
event = self._create_random_event(non_evidence_variables, evidence, randomizer)
for i in range(number_of_values):
for non_evidence_var in non_evidence_variables:
node = self._get_node_of(non_evidence_var)
markov_blanket = self._create_markov_blanket(node)
mb = self._create_mb_values(markov_blanket, event)
true_probability, false_probability = self.rejection_sample(node.variable, mb, 100, randomizer)
event[node.variable] = self._truth_value(true_probability, randomizer)
query_value = event[var]
if query_value:
true_values += 1
else:
false_values += 1
return normalize((true_values, false_values))
