def six_node_graph():
Path('data').mkdir(exist_ok=True)
rng = RandomState(1)
graph = GraphModel()
rv_dim = 2
# init varnode
for _ in range(6):
potential = np.ones([rv_dim])
varnode = VarNode(rv_dim, potential)
graph.add_varnode(varnode)
# init factornode
edges = [
[0, 1],
[1, 2],
[1, 3],
[3, 4],
[3, 5]
]
for item in edges:
potential = rng.normal(size=[rv_dim, rv_dim])
factorname = [f"VarNode_{data:03d}" for data in item]
factornode = FactorNode(factorname, np.exp(potential))
graph.add_factornode(factornode)
graph.plot(f"data/six_node_graph.png")
return graph
def add_factor_different(self, name_list, is_conv=False):
if is_conv:
factor_potential = diagonal_potential_conv(self.node_dim,
self.node_dim, self.rng)
else:
factor_potential = diagonal_potential_different(
self.node_dim, self.node_dim, self.rng)
factornode = FactorNode(name_list, factor_potential)
self.graph.add_factornode(factornode)
return factornode
def setUp(self):
self.graph = GraphModel(coef_policy=bp_policy)
# init varnode
potential = np.ones(2)
self.varnode_1 = VarNode(2, potential)
self.graph.add_varnode(self.varnode_1)
varnode_name = [node.name for node in [self.varnode_1]]
factor_potential = np.array(
[3, 2]
)
self.factornode_1 = FactorNode(varnode_name, factor_potential)
self.graph.add_factornode(self.factornode_1)
factor_potential_2 = np.array(
[1, 4]
)
self.factornode_2 = FactorNode(varnode_name, factor_potential_2)
self.graph.add_factornode(self.factornode_2)
def three_node_tree():
graph = two_node_tree()
varnode = VarNode(2)
graph.add_varnode(varnode)
connect_var = ['VarNode_001', 'VarNode_002']
factor_potential = np.array([
[6, 14],
[1, 4]
])
factornode = FactorNode(connect_var, factor_potential)
graph.add_factornode(factornode)
return graph
def two_node_tree():
graph = GraphModel()
# init varnode
for _ in range(2):
varnode = VarNode(2)
graph.add_varnode(varnode)
connect_var = ['VarNode_000', 'VarNode_001']
factor_potential = np.array([
[3, 2],
[3, 17]
])
factornode = FactorNode(connect_var, factor_potential)
graph.add_factornode(factornode)
return graph
def add_factor(self, name_list, is_conv=False):
"""add factor to hmm graph
:param name_list: connected node
:type name_list: list
:param is_conv: is emit or not , defaults to False-- transition
:type is_conv: bool, optional
:return: FactorNode
:rtype: cbp.FactorNode
"""
if is_conv: # emit
potential = self.simulator.get_emission_potential()
else:
potential = self.simulator.get_transition_potential()
factornode = FactorNode(name_list, potential)
self.graph.add_factornode(factornode)
return factornode
def add_factor(self, name_list, is_conv=False):
factor_potential = identity_potential(self.node_dim, self.node_dim,
self.rng)
factornode = FactorNode(name_list, factor_potential)
self.graph.add_factornode(factornode)
return factornode