def _bayesian_diffuse_to_children(self, kc, dynamic=False):
children = self.link_model.get_children(kc)
if children:
for child in children:
child_parents = self.link_model.get_parents(child)
child_parents_truthtable = truthtable(len(child_parents))
m_pba_truthtable = [
np.prod([
self.get_mastering_probability(child_parents[i])
if child_parents_truthtable[j][i] else 1 -
self.get_mastering_probability(child_parents[i])
for i in range(len(child_parents))
]) for j in range(len(child_parents_truthtable))
]
cond_pbas = self.link_model.get_links(
child)['from_parents'].probability_vector
if dynamic:
alpha = 0.8
prior_child_pba = self.get_mastering_probability(child)
child_m_pba = alpha * (
prior_child_pba + (1 - prior_child_pba) * np.dot(m_pba_truthtable, cond_pbas)) \
+ (1 - alpha) * np.dot(m_pba_truthtable, cond_pbas)
else:
child_m_pba = np.dot(m_pba_truthtable, cond_pbas)
self.set_mastering_probability(child, child_m_pba)
self._bayesian_diffuse_to_children(child)
def _bayesian_diffuse_to_parents(self, kc, dynamic=False):
parents = self.link_model.get_parents(kc)
if parents:
for parent in parents:
parent_children = self.link_model.get_children(parent)
parent_children_truthtable = truthtable(len(parent_children))
m_pba_truthtable = [
np.prod([
self.get_mastering_probability(parent_children[i])
if parent_children_truthtable[j][i] else 1 -
self.get_mastering_probability(parent_children[i])
for i in range(len(parent_children))
]) for j in range(len(parent_children_truthtable))
]
cond_pbas = self.link_model.get_links(
parent)['from_children'].probability_vector
if dynamic:
alpha = 0.8
prior_parent_pba = self.get_mastering_probability(parent)
parent_m_pba = alpha * (
prior_parent_pba + (1 - prior_parent_pba) * np.dot(m_pba_truthtable, cond_pbas)) \
+ (1 - alpha) * np.dot(m_pba_truthtable, cond_pbas)
else:
parent_m_pba = np.dot(m_pba_truthtable, cond_pbas)
self.set_mastering_probability(parent, parent_m_pba)
self._bayesian_diffuse_to_parents(parent, dynamic)
def get_pc_vec(self):
pc_vec = []
conds = truthtable(len(self.parents) + len(self.children))
for cond in conds:
cond_p_prob = self.p_vec[bool_list_to_int(
cond[:len(self.parents)])] if any(self.p_vec) else 1.
cond_c_prob = self.c_vec[bool_list_to_int(
cond[len(self.parents):])] if any(self.c_vec) else 1.
print(self.p_vec, self.c_vec)
print(cond, cond_p_prob, cond_c_prob)
pc_vec.append(cond_p_prob * cond_c_prob)
return pc_vec
def set_children_conditioned_probability_vector(self,
vec=None,
step_by_step=False):
"""
Set the conditional probability vector of KnowledgeComponent given the mastering of its children.
:param vec: the conditional probability vector knowing the states of children given by converting the index of
the probability vector in binary (e.g. for B, C children of A, vec is 4 length : 1st index corresponds to
B is False, C is False ; 3rd index is B True, C False)
:param step_by_step: if the form of linked_knowledge_component is unknown, it is possible to set conditional
probabilities value by value.
:return: None, only changes the value of self.probability_vector
"""
if not step_by_step:
assert len(vec) == 2**len(self.linked_knowledge_components), "Length of the probability vector " \
"should be 2^n."
self.probability_vector = vec
else:
combinations = truthtable(len(self.linked_knowledge_components))
for comb in combinations:
self.probability_vector[bool_list_to_int(comb)] = input(
f"Probability {self.knowledge_component.name} knowing "
f"{[child.name for child in self.linked_knowledge_components]} value {comb} is :"
)