def log_current_conditional_probability(self):
assert(self.current_value > 0)
shape = Node.parent_node_value(self.shape)
scale = Node.parent_node_value(self.scale)
p = stats.invgamma.pdf(self.current_value, a=shape, scale=scale)
log_p = (0 if p == 0 else math.log(p))
_log.debug("p({}={}) = {}".format(self.display_name, self.current_value, p))
return log_p
def log_current_conditional_probability(self):
assert(self.current_value > 0)
alpha = Node.parent_node_value(self.alpha)
beta = Node.parent_node_value(self.beta)
p = stats.beta.pdf(self.current_value, a=alpha, b=beta)
log_p = (0 if p == 0 else math.log(p))
_log.debug("p({}={}) = {}".format(self.display_name, self.current_value, p))
return log_p
def log_current_conditional_probability(self):
"""
Return probability given current values of 'mean' and 'var'.
(If 'mean' and 'var' are parent nodes, get their current_value.)
"""
mean = Node.parent_node_value(self.mean)
var = Node.parent_node_value(self.var)
p = stats.norm.pdf(self.current_value, mean, math.sqrt(var))
_log.debug(" p = {}".format(p))
log_p = (0 if p == 0 else math.log(p))
_log.debug("p({}={}) = {}".format(self.display_name, self.current_value, p))
return log_p
def log_current_conditional_probability(self):
assert(self.current_value > 0)
rate = Node.parent_node_value(self.rate)
p = stats.poisson.pmf(self.current_value, mu=rate)
log_p = (0 if p == 0 else math.log(p))
_log.debug("p({}={}) = {}".format(self.display_name, self.current_value, p))
return log_p
def log_current_conditional_probability(self):
"""
For Bernoulli/Binomial, sample directly instead of trying to use Metropolis.
"""
p = Node.parent_node_value(self.p)
sample = 1 if random.random() <= p else 0
log_sample = (0 if sample == 0 else math.log(sample))
_log.debug("p({}={}) = {}".format(self.display_name, self.current_value, sample))
return log_sample
