def __init__(
self,
gid: str,
nodes: Set[NumCatRVariable],
edges: Set[Edge],
factors: Optional[Set[Factor]] = None,
data={},
):
"""!
\brief constructor for a generic Probabilistic Graphical Model
The generic model that extends the #Graph definition by adding a new
set, called set of factors.
Most of the parameters are documented in #Graph.
"""
super().__init__(gid=gid, data=data, nodes=nodes, edges=edges)
if factors is None:
fs: Set[Factor] = set()
for e in BaseGraphOps.edges(self):
estart = e.start()
eend = e.end()
sdata = estart.data()
edata = eend.data()
evidences = set()
if "evidence" in sdata:
evidences.add((estart.id(), sdata["evidence"]))
if "evidence" in edata:
evidences.add((eend.id(), edata["evidence"]))
f = Factor(gid=str(uuid4()), scope_vars=set([estart, eend]))
if len(evidences) != 0:
f = f.reduced_by_value(evidences)
fs.add(f)
self.Fs = fs
else:
self.Fs = factors
def from_undigraph(cls, udi: UndiGraph):
"""!
\brief Make a markov network from undirected graph
\throws ValueError If nodes are not an instance of a random variable,
we raise a value error.
Unless it is specified we assume that edges indicate a joint
distribution
\code{.py}
>>> idata = {"A": {"outcome-values": [True, False]}}
>>> a = NumCatRVariable(
>>> node_id="a",
>>> input_data=idata["A"],
>>> distribution=lambda x: 0.01 if x else 0.99,
>>> )
>>> b = NumCatRVariable(
>>> node_id="b", input_data=idata["B"], distribution=lambda x: 0.5
>>> )
>>> d = NumCatRVariable(
>>> node_id="d",
>>> input_data=idata["A"],
>>> distribution=lambda x: 0.7468 if x else 0.2532,
>>> )
>>> c = NumCatRVariable(
>>> node_id="c",
>>> input_data=idata["A"],
>>> distribution=lambda x: 0.7312 if x else 0.2688,
>>> )
>>> ab = Edge(
>>> "ab", start_node=a, end_node=b, edge_type=EdgeType.UNDIRECTED
>>> )
>>> ad = Edge(
>>> "ad", start_node=a, end_node=d, edge_type=EdgeType.UNDIRECTED
>>> )
>>> bc = Edge(
>>> "bc", start_node=b, end_node=c, edge_type=EdgeType.UNDIRECTED
>>> )
>>> dc = Edge(
>>> "dc", start_node=d, end_node=c, edge_type=EdgeType.UNDIRECTED
>>> )
>>> ugraph = UndiGraph(
>>> "ug1",
>>> data={"m": "f"},
>>> nodes=set([a, b, c, d]),
>>> edges=set([ab, ad, bc, dc]),
>>> )
>>> markov = MarkovNetwork.from_undigraph(udi=ugraph)
\endcode
"""
for n in BaseGraphOps.nodes(udi):
if not isinstance(n, RandomVariable):
raise ValueError(
"Nodes are not an instance of random variable")
fs: Set[Factor] = set()
maximal_cliques = udi.find_maximal_cliques()
for clique in maximal_cliques:
evidences = set()
for n in clique:
edata = n.data()
if "evidence" in edata:
evidences.add((n.id(), edata["evidence"]))
f = Factor(gid=str(uuid4()), scope_vars=clique)
if len(evidences) != 0:
f = f.reduced_by_value(evidences)
fs.add(f)
return MarkovNetwork(
gid=str(uuid4()),
nodes=BaseGraphOps.nodes(udi),
edges=BaseGraphOps.edges(udi),
factors=fs,
)
