def test_load_from_json():
dg = BayesianNetwork()
with open('one_parent_net.json', 'r') as f:
json_ = json.load(f)
dg.load_graph_from_json(json_)
assert len(dg) == 2
parents = dg.get_parents('parent')
assert len(parents) == 0
parents = dg.get_parents('child')
assert len(parents) == 1
factor = dg.get_value('parent', 'factor')
f = Factor(scope=np.array([0]),
card=np.array([2]),
val=np.array([0.57, 0.43]))
assert is_equal(f, factor)
factor = dg.get_value('child', 'factor')
f = Factor(scope=np.array([1, 0]),
card=np.array([2, 2]),
val=np.array([0.40, 0.60, 0.13, 0.87]))
assert is_equal(f, factor)
def test_another_marg():
f1 = Factor(scope=np.array([0]),
card=np.array([2], dtype=np.int32),
val=np.array([0.57, 0.43]))
f2 = Factor(scope=np.array([1, 0]),
card=np.array([2, 2], dtype=np.int32),
val=np.array([0.4, 0.6, 0.13, 0.87]))
f = get_product(f1, f2)
f_marg = get_marg(f, 1)
assert np.array_equal(f_marg.scope, np.array([0]))
assert np.array_equal(f_marg.card, np.array([2]))
assert np.allclose(f_marg.val, np.array([0.57, 0.43]))
def f1():
'''
f1
[1 phi]
--------
0 0.11
1 0.89
'''
return Factor(scope=np.array([1]),
card=np.array([2]),
val=np.array([0.11, 0.89]))
def f4():
'''
f4
[4 5 phi]
------------
[0 0 0.39]
[1 0 0.61]
[0 1 0.06]
[1 1 0.94]
'''
return Factor(scope=np.array([4, 5]),
card=np.array([2, 2]),
val=np.array([0.39, 0.61, 0.06, 0.94]))
def f3():
'''
f3
[3 2 phi]
------------
[0 0 0.39]
[1 0 0.61]
[0 1 0.06]
[1 1 0.94]
'''
return Factor(scope=np.array([3, 2]),
card=np.array([2, 2]),
val=np.array([0.39, 0.61, 0.06, 0.94]))
def f2():
'''
f2
[2 1 phi]
------------
[0 0 0.59]
[1 0 0.41]
[0 1 0.22]
[1 1 0.78]
'''
return Factor(scope=np.array([2, 1]),
card=np.array([2, 2]),
val=np.array([0.59, 0.41, 0.22, 0.78]))
def load_graph_from_json(self, json_):
self._dg = nx.DiGraph()
for node_name, value in json_.iteritems():
# Have we seen this node name before?
if node_name not in self._to_index:
self._to_index[node_name] = self._index
self._to_name[self._index] = node_name
self._index += 1
for p in value['parents']:
if p not in self._to_index:
self._to_index[p] = self._index
self._to_name[self._index] = p
self._index += 1
p_indx = [self._to_index[p] for p in value['parents']]
c_indx = self._to_index[node_name]
scope = np.array([c_indx] + p_indx)
card = np.ones(scope.shape[0], dtype=np.int32) * 2
N = np.prod(card)
val = np.empty((N,))
for v in value['values']:
A = np.array(v['states'])
A = A.reshape((-1, A.shape[0]))
i = assign_to_indx(A, card)[0]
val[i] = v['value']
# What are the parents?
f = Factor(scope=scope,
card=card,
val=val)
f = log(f)
var_scope = set([node_name] + value['parents'])
self.add_node(node_name=node_name,
attr_dict={'factor': f,
'scope': var_scope})
self.add_parents(node_name, value['parents'])
def test_is_equal_scope_out_of_order(f2):
'''
f2
[2 1 phi]
------------
[0 0 0.59]
[1 0 0.41]
[0 1 0.22]
[1 1 0.78]
f2_
[1 2 phi]
------------
[0 0 0.59]
[1 0 0.22]
[0 1 0.41]
[1 1 0.78]
'''
f2_ = Factor(scope=np.array([1, 2]),
card=np.array([2, 2]),
val=np.array([0.59, 0.22, 0.41, 0.78]))
assert is_equal(f2, f2_)
assert is_equal(f2_, f2)