def to_bn(self, use_mst=False):
""" Given the undirected graph, return the directed model corresponding to the minimal I-map. The directed model is
placed in the directed attribute.
Remarks: This method supports models which are not connected.
:param use_mst (boolean): To direct a prelearned model in the directed attribute, direct_mst=False. To obtain
the directed model from the maximum spanning tree learned by self.mst, use direct_mst=True.
"""
# Generate connected components
if use_mst:
edges = list(self.maxtree.edges())
vertices = list(self.maxtree.nodes)
mm = MarkovModel()
mm.add_nodes_from(vertices)
mm.add_edges_from(edges)
bm = mm.to_bayesian_model()
self.directed = bm
else:
connected = list(nx.algorithms.components.connected_components(self.undirected))
# Hold the edges of each connected component as a list(list())
connected_comps = list()
# If the graph is not completely connected
if len(connected) > 1:
self.directed = BayesianModel()
for comp in connected:
# If the connected component is not a single vertex
if len(comp) > 1:
edges_to_add = list()
temp = MarkovModel()
for vert1 in comp:
neighbours = list(self.undirected.neighbors(vert1))
for vert2 in neighbours:
if not ((vert1, vert2) in edges_to_add) and not ((vert2, vert1) in edges_to_add):
edges_to_add.append((vert1, vert2))
temp.add_nodes_from(comp)
temp.add_edges_from(edges_to_add)
temp_bn = temp.to_bayesian_model()
connected_comps.append(temp_bn)
else:
self.directed.add_nodes_from(comp)
for bn in connected_comps:
self.directed.add_nodes_from(list(bn.nodes))
self.directed.add_edges_from(list(bn.edges))
else:
# If the graph is completely connected, just add all edges to markov model
edges = list(self.undirected.edges())
vertices = list(self.undirected.nodes)
mm = MarkovModel()
mm.add_nodes_from(vertices)
mm.add_edges_from(edges)
bm = mm.to_bayesian_model()
self.directed = bm
def setUp(self):
self.bayesian = BayesianModel([('a', 'b'), ('b', 'c'), ('c', 'd'),
('d', 'e')])
a_cpd = TabularCPD('a', 2, [[0.4, 0.6]])
b_cpd = TabularCPD('b',
2, [[0.2, 0.4], [0.3, 0.4]],
evidence='a',
evidence_card=[2])
c_cpd = TabularCPD('c',
2, [[0.1, 0.2], [0.3, 0.4]],
evidence='b',
evidence_card=[2])
d_cpd = TabularCPD('d',
2, [[0.4, 0.3], [0.2, 0.1]],
evidence='c',
evidence_card=[2])
e_cpd = TabularCPD('e',
2, [[0.3, 0.2], [0.4, 0.1]],
evidence='d',
evidence_card=[2])
self.bayesian.add_cpds(a_cpd, b_cpd, c_cpd, d_cpd, e_cpd)
self.markov = MarkovModel([('a', 'b'), ('b', 'd'), ('a', 'c'),
('c', 'd')])
factor_1 = Factor(['a', 'b'], [2, 2], np.array([100, 1, 1, 100]))
factor_2 = Factor(['a', 'c'], [2, 2], np.array([40, 30, 100, 20]))
factor_3 = Factor(['b', 'd'], [2, 2], np.array([1, 100, 100, 1]))
factor_4 = Factor(['c', 'd'], [2, 2], np.array([60, 60, 40, 40]))
self.markov.add_factors(factor_1, factor_2, factor_3, factor_4)
def setUp(self):
# A test Bayesian model
diff_cpd = TabularCPD('diff', 2, [[0.6], [0.4]])
intel_cpd = TabularCPD('intel', 2, [[0.7], [0.3]])
grade_cpd = TabularCPD('grade',
3,
[[0.3, 0.05, 0.9, 0.5], [0.4, 0.25, 0.08, 0.3],
[0.3, 0.7, 0.02, 0.2]],
evidence=['diff', 'intel'],
evidence_card=[2, 2])
self.bayesian_model = BayesianModel()
self.bayesian_model.add_nodes_from(['diff', 'intel', 'grade'])
self.bayesian_model.add_edges_from([('diff', 'grade'),
('intel', 'grade')])
self.bayesian_model.add_cpds(diff_cpd, intel_cpd, grade_cpd)
# A test Markov model
self.markov_model = MarkovModel([('A', 'B'), ('C', 'B'), ('B', 'D')])
factor_ab = DiscreteFactor(['A', 'B'], [2, 3], [1, 2, 3, 4, 5, 6])
factor_cb = DiscreteFactor(['C', 'B'], [4, 3],
[3, 1, 4, 5, 7, 8, 1, 3, 10, 4, 5, 6])
factor_bd = DiscreteFactor(['B', 'D'], [3, 2], [5, 7, 2, 1, 9, 3])
self.markov_model.add_factors(factor_ab, factor_cb, factor_bd)
self.gibbs = GibbsSampling(self.bayesian_model)
def test_find_MAP():
print '-' * 80
G = MarkovModel()
G.add_nodes_from(['x1', 'x2', 'x3'])
G.add_edges_from([('x1', 'x2'), ('x1', 'x3')])
phi = [
DiscreteFactor(['x2', 'x1'],
cardinality=[2, 2],
values=np.array([[1.0 / 1, 1.0 / 2], [1.0 / 3,
1.0 / 4]])),
DiscreteFactor(['x3', 'x1'],
cardinality=[2, 2],
values=np.array([[1.0 / 1, 1.0 / 2], [1.0 / 3,
1.0 / 4]]))
]
# DiscreteFactor(['x1'], cardinality=[2],
# values=np.array([2,2]))]
G.add_factors(*phi)
print "nodes:", G.nodes()
bp = BeliefPropagation(G)
bp.max_calibrate()
# bp.calibrate()
clique_beliefs = bp.get_clique_beliefs()
print clique_beliefs
print clique_beliefs[('x1', 'x2')]
print clique_beliefs[('x1', 'x3')]
# print 'partition function should be', np.sum(clique_beliefs[('x1', 'x3')].values)
phi_query = bp._query(['x1', 'x2', 'x3'], operation='maximize')
# phi_query = bp._query(['x1', 'x2', 'x3'], operation='marginalize')
print phi_query
sleep(52)
def setUp(self):
# It is just a moralised version of the above Bayesian network so all the results are same. Only factors
# are under consideration for inference so this should be fine.
self.markov_model = MarkovModel([('A', 'J'), ('R', 'J'), ('J', 'Q'),
('J', 'L'), ('G', 'L'), ('A', 'R'),
('J', 'G')])
factor_a = TabularCPD('A', 2, values=[[0.2], [0.8]]).to_factor()
factor_r = TabularCPD('R', 2, values=[[0.4], [0.6]]).to_factor()
factor_j = TabularCPD('J',
2,
values=[[0.9, 0.6, 0.7, 0.1],
[0.1, 0.4, 0.3, 0.9]],
evidence=['A', 'R'],
evidence_card=[2, 2]).to_factor()
factor_q = TabularCPD('Q',
2,
values=[[0.9, 0.2], [0.1, 0.8]],
evidence=['J'],
evidence_card=[2]).to_factor()
factor_l = TabularCPD('L',
2,
values=[[0.9, 0.45, 0.8, 0.1],
[0.1, 0.55, 0.2, 0.9]],
evidence=['J', 'G'],
evidence_card=[2, 2]).to_factor()
factor_g = TabularCPD('G', 2, [[0.6], [0.4]]).to_factor()
self.markov_model.add_factors(factor_a, factor_r, factor_j, factor_q,
factor_l, factor_g)
self.markov_inference = VariableElimination(self.markov_model)
def __init__(self, theta, seed=None):
if seed is not None:
np.random.seed(seed)
# graph
self.G = MarkovModel()
for _, row in theta.iterrows():
# unary
if row["j"]==row["k"]:
self.G.add_node(str(int(row["j"])))
theta_jj = row["value"]
self.G.add_factors(DiscreteFactor([str(int(row["j"]))], [2],
np.exp([-theta_jj,theta_jj])))
# pairwise
elif row["value"]!=0:
self.G.add_edge(str(int(row["j"])), str(int(row["k"])))
theta_jk = row["value"]
self.G.add_factors(DiscreteFactor([str(int(row["j"])), str(int(row["k"]))],
[2, 2], np.exp([theta_jk, -theta_jk, -theta_jk, theta_jk])))
self.G.check_model()
self.infer = BeliefPropagation(self.G)
self.infer.calibrate()
def setUp(self):
self.bayesian = BayesianModel([("a", "b"), ("b", "c"), ("c", "d"),
("d", "e")])
a_cpd = TabularCPD("a", 2, [[0.4, 0.6]])
b_cpd = TabularCPD("b",
2, [[0.2, 0.4], [0.8, 0.6]],
evidence=["a"],
evidence_card=[2])
c_cpd = TabularCPD("c",
2, [[0.1, 0.2], [0.9, 0.8]],
evidence=["b"],
evidence_card=[2])
d_cpd = TabularCPD("d",
2, [[0.4, 0.3], [0.6, 0.7]],
evidence=["c"],
evidence_card=[2])
e_cpd = TabularCPD("e",
2, [[0.3, 0.2], [0.7, 0.8]],
evidence=["d"],
evidence_card=[2])
self.bayesian.add_cpds(a_cpd, b_cpd, c_cpd, d_cpd, e_cpd)
self.markov = MarkovModel([("a", "b"), ("b", "d"), ("a", "c"),
("c", "d")])
factor_1 = DiscreteFactor(["a", "b"], [2, 2],
np.array([100, 1, 1, 100]))
factor_2 = DiscreteFactor(["a", "c"], [2, 2],
np.array([40, 30, 100, 20]))
factor_3 = DiscreteFactor(["b", "d"], [2, 2],
np.array([1, 100, 100, 1]))
factor_4 = DiscreteFactor(["c", "d"], [2, 2],
np.array([60, 60, 40, 40]))
self.markov.add_factors(factor_1, factor_2, factor_3, factor_4)
def setUp(self):
# A test Bayesian model
diff_cpd = TabularCPD("diff", 2, [[0.6], [0.4]])
intel_cpd = TabularCPD("intel", 2, [[0.7], [0.3]])
grade_cpd = TabularCPD(
"grade",
3,
[[0.3, 0.05, 0.9, 0.5], [0.4, 0.25, 0.08, 0.3],
[0.3, 0.7, 0.02, 0.2]],
evidence=["diff", "intel"],
evidence_card=[2, 2],
)
self.bayesian_model = BayesianModel()
self.bayesian_model.add_nodes_from(["diff", "intel", "grade"])
self.bayesian_model.add_edges_from([("diff", "grade"),
("intel", "grade")])
self.bayesian_model.add_cpds(diff_cpd, intel_cpd, grade_cpd)
# A test Markov model
self.markov_model = MarkovModel([("A", "B"), ("C", "B"), ("B", "D")])
factor_ab = DiscreteFactor(["A", "B"], [2, 3], [1, 2, 3, 4, 5, 6])
factor_cb = DiscreteFactor(["C", "B"], [4, 3],
[3, 1, 4, 5, 7, 8, 1, 3, 10, 4, 5, 6])
factor_bd = DiscreteFactor(["B", "D"], [3, 2], [5, 7, 2, 1, 9, 3])
self.markov_model.add_factors(factor_ab, factor_cb, factor_bd)
self.gibbs = GibbsSampling(self.bayesian_model)
def generate_random_fully_connected_MRF(N):
'''
Create an MRF with N binary nodes that are all part of the same clique, which
has 2^N random values
'''
G = MarkovModel()
#create an N nodes
node_names = ['x%d' % i for i in range(N)]
G.add_nodes_from(node_names)
#add an edge between each node and its 4 neighbors, except when the
#node is on the grid border and has fewer than 4 neighbors
edges = []
for i in range(N):
for j in range(i + 1, N):
edges.append(('x%d' % i, 'x%d' % j))
assert (len(edges) == .5 * N**2)
G.add_edges_from(edges)
single_factor = DiscreteFactor(
edges,
cardinality=[2 for i in range(N)],
values=np.random.rand(*[2 for i in range(N)]))
G.add_factors(single_factor)
return G
def test_is_iequivalent(self):
G = BayesianModel([('x', 'y'), ('z', 'y'), ('x', 'z'), ('w', 'y')])
self.assertRaises(TypeError, G.is_iequivalent, MarkovModel())
G1 = BayesianModel([('V', 'W'), ('W', 'X'), ('X', 'Y'), ('Z', 'Y')])
G2 = BayesianModel([('W', 'V'), ('X', 'W'), ('X', 'Y'), ('Z', 'Y')])
self.assertTrue(G1.is_iequivalent(G2))
G3 = BayesianModel([('W', 'V'), ('W', 'X'), ('Y', 'X'), ('Z', 'Y')])
self.assertFalse(G3.is_iequivalent(G2))
def test_is_iequivalent(self):
G = BayesianModel([("x", "y"), ("z", "y"), ("x", "z"), ("w", "y")])
self.assertRaises(TypeError, G.is_iequivalent, MarkovModel())
G1 = BayesianModel([("V", "W"), ("W", "X"), ("X", "Y"), ("Z", "Y")])
G2 = BayesianModel([("W", "V"), ("X", "W"), ("X", "Y"), ("Z", "Y")])
self.assertTrue(G1.is_iequivalent(G2))
G3 = BayesianModel([("W", "V"), ("W", "X"), ("Y", "X"), ("Z", "Y")])
self.assertFalse(G3.is_iequivalent(G2))
def build_graph(self):
mm = MarkovModel()
mm.add_nodes_from(self.dataset.dataframe.columns)
graph_structure_name = list(
map(lambda tuple: (self.columns[tuple[0]], self.columns[tuple[1]]),
self.graph_structure_index))
mm.add_edges_from(graph_structure_name)
self.model = mm.to_bayesian_model()
def setUp(self):
self.maxDiff = None
edges = [['family-out', 'dog-out'],
['bowel-problem', 'dog-out'],
['family-out', 'light-on'],
['dog-out', 'hear-bark']]
cpds = {'bowel-problem': np.array([[0.01],
[0.99]]),
'dog-out': np.array([[0.99, 0.01, 0.97, 0.03],
[0.9, 0.1, 0.3, 0.7]]),
'family-out': np.array([[0.15],
[0.85]]),
'hear-bark': np.array([[0.7, 0.3],
[0.01, 0.99]]),
'light-on': np.array([[0.6, 0.4],
[0.05, 0.95]])}
states = {'bowel-problem': ['true', 'false'],
'dog-out': ['true', 'false'],
'family-out': ['true', 'false'],
'hear-bark': ['true', 'false'],
'light-on': ['true', 'false']}
parents = {'bowel-problem': [],
'dog-out': ['bowel-problem', 'family-out'],
'family-out': [],
'hear-bark': ['dog-out'],
'light-on': ['family-out']}
self.bayesmodel = BayesianModel(edges)
tabular_cpds = []
for var, values in cpds.items():
cpd = TabularCPD(var, len(states[var]), values,
evidence=parents[var],
evidence_card=[len(states[evidence_var])
for evidence_var in parents[var]])
tabular_cpds.append(cpd)
self.bayesmodel.add_cpds(*tabular_cpds)
self.bayeswriter = UAIWriter(self.bayesmodel)
edges = {('var_0', 'var_1'), ('var_0', 'var_2'), ('var_1', 'var_2')}
self.markovmodel = MarkovModel(edges)
tables = [(['var_0', 'var_1'],
['4.000', '2.400', '1.000', '0.000']),
(['var_0', 'var_1', 'var_2'],
['2.2500', '3.2500', '3.7500', '0.0000', '0.0000', '10.0000',
'1.8750', '4.0000', '3.3330', '2.0000', '2.0000', '3.4000'])]
domain = {'var_1': '2', 'var_2': '3', 'var_0': '2'}
factors = []
for table in tables:
variables = table[0]
cardinality = [int(domain[var]) for var in variables]
values = list(map(float, table[1]))
factor = DiscreteFactor(variables, cardinality, values)
factors.append(factor)
self.markovmodel.add_factors(*factors)
self.markovwriter = UAIWriter(self.markovmodel)
def markovmodel_test():
"""
>>> from ibeis.algo.hots.pgm_ext import * # NOQA
"""
from pgmpy.models import MarkovModel
from pgmpy.factors import Factor
markovmodel = MarkovModel([('A', 'B'), ('B', 'C'), ('C', 'D'), ('D', 'A')])
factor_a_b = Factor(variables=['A', 'B'], cardinality=[2, 2], values=[100, 5, 5, 100])
factor_b_c = Factor(variables=['B', 'C'], cardinality=[2, 2], values=[100, 3, 2, 4])
factor_c_d = Factor(variables=['C', 'D'], cardinality=[2, 2], values=[3, 5, 1, 6])
factor_d_a = Factor(variables=['D', 'A'], cardinality=[2, 2], values=[6, 2, 56, 2])
markovmodel.add_factors(factor_a_b, factor_b_c, factor_c_d, factor_d_a)
pgm_viz.show_markov_model(markovmodel)
pgm_viz.show_junction_tree(markovmodel)
def __init__(self, pnh, gh):
'''
Constructor
'''
extractor = pnh.get_data_extractor()
self.best_model = BayesianModel()
self.training_instances = ""
self.device_considered = pnh.get_device()
self.priority_considered = pnh.get_priority()
self.markov = MarkovModel()
self.general_handler = gh
self.variables_names = extractor.get_variable_names()
self.rankedDevices = extractor.get_ranked_devices()
self.data = pnh.get_dataframe()
self.file_writer = pnh.get_file_writer()
self.file_suffix = pnh.get_file_suffix()
def get_model(self):
"""
Returns an instance of Bayesian Model or Markov Model.
Varibles are in the pattern var_0, var_1, var_2 where var_0 is
0th index variable, var_1 is 1st index variable.
Return
------
model: an instance of Bayesian or Markov Model.
Examples
--------
>>> reader = UAIReader('TestUAI.uai')
>>> reader.get_model()
"""
if self.network_type == 'BAYES':
model = BayesianModel()
model.add_nodes_from(self.variables)
model.add_edges_from(self.edges)
tabular_cpds = []
for cpd in self.tables:
child_var = cpd[0]
states = int(self.domain[child_var])
arr = list(map(float, cpd[1]))
values = np.array(arr)
values = values.reshape(states, values.size // states)
tabular_cpds.append(TabularCPD(child_var, states, values))
model.add_cpds(*tabular_cpds)
return model
elif self.network_type == 'MARKOV':
model = MarkovModel(self.edges)
factors = []
for table in self.tables:
variables = table[0]
cardinality = [int(self.domain[var]) for var in variables]
value = list(map(float, table[1]))
factor = DiscreteFactor(variables=variables,
cardinality=cardinality,
values=value)
factors.append(factor)
model.add_factors(*factors)
return model
def setUp(self):
self.bayesian_model = BayesianModel([('A', 'J'), ('R', 'J'),
('J', 'Q'), ('J', 'L'),
('G', 'L')])
cpd_a = TabularCPD('A', 2, [[0.2], [0.8]])
cpd_r = TabularCPD('R', 2, [[0.4], [0.6]])
cpd_j = TabularCPD('J', 2,
[[0.9, 0.6, 0.7, 0.1], [0.1, 0.4, 0.3, 0.9]],
['R', 'A'], [2, 2])
cpd_q = TabularCPD('Q', 2, [[0.9, 0.2], [0.1, 0.8]], ['J'], [2])
cpd_l = TabularCPD('L', 2,
[[0.9, 0.45, 0.8, 0.1], [0.1, 0.55, 0.2, 0.9]],
['G', 'J'], [2, 2])
cpd_g = TabularCPD('G', 2, [[0.6], [0.4]])
self.bayesian_model.add_cpds(cpd_a, cpd_g, cpd_j, cpd_l, cpd_q, cpd_r)
self.sampling_inference = BayesianModelSampling(self.bayesian_model)
self.markov_model = MarkovModel()
def setUp(self):
self.bayesian_model = BayesianModel([("A", "J"), ("R", "J"),
("J", "Q"), ("J", "L"),
("G", "L")])
cpd_a = TabularCPD("A", 2, [[0.2], [0.8]])
cpd_r = TabularCPD("R", 2, [[0.4], [0.6]])
cpd_j = TabularCPD("J", 2,
[[0.9, 0.6, 0.7, 0.1], [0.1, 0.4, 0.3, 0.9]],
["R", "A"], [2, 2])
cpd_q = TabularCPD("Q", 2, [[0.9, 0.2], [0.1, 0.8]], ["J"], [2])
cpd_l = TabularCPD("L", 2,
[[0.9, 0.45, 0.8, 0.1], [0.1, 0.55, 0.2, 0.9]],
["G", "J"], [2, 2])
cpd_g = TabularCPD("G", 2, [[0.6], [0.4]])
self.bayesian_model.add_cpds(cpd_a, cpd_g, cpd_j, cpd_l, cpd_q, cpd_r)
self.sampling_inference = BayesianModelSampling(self.bayesian_model)
self.markov_model = MarkovModel()
def setUp(self):
# It is just a moralised version of the above Bayesian network so all the results are same. Only factors
# are under consideration for inference so this should be fine.
self.markov_model = MarkovModel([
("A", "J"),
("R", "J"),
("J", "Q"),
("J", "L"),
("G", "L"),
("A", "R"),
("J", "G"),
])
factor_a = TabularCPD("A", 2, values=[[0.2], [0.8]]).to_factor()
factor_r = TabularCPD("R", 2, values=[[0.4], [0.6]]).to_factor()
factor_j = TabularCPD(
"J",
2,
values=[[0.9, 0.6, 0.7, 0.1], [0.1, 0.4, 0.3, 0.9]],
evidence=["A", "R"],
evidence_card=[2, 2],
).to_factor()
factor_q = TabularCPD("Q",
2,
values=[[0.9, 0.2], [0.1, 0.8]],
evidence=["J"],
evidence_card=[2]).to_factor()
factor_l = TabularCPD(
"L",
2,
values=[[0.9, 0.45, 0.8, 0.1], [0.1, 0.55, 0.2, 0.9]],
evidence=["J", "G"],
evidence_card=[2, 2],
).to_factor()
factor_g = TabularCPD("G", 2, [[0.6], [0.4]]).to_factor()
self.markov_model.add_factors(factor_a, factor_r, factor_j, factor_q,
factor_l, factor_g)
self.markov_inference = VariableElimination(self.markov_model)
def to_markov_model(self):
"""
Converts bayesian model to markov model. The markov model created would
be the moral graph of the bayesian model.
Examples
--------
>>> from pgmpy.models import BayesianModel
>>> G = BayesianModel([('diff', 'grade'), ('intel', 'grade'),
... ('intel', 'SAT'), ('grade', 'letter')])
>>> mm = G.to_markov_model()
>>> mm.nodes()
['diff', 'grade', 'intel', 'SAT', 'letter']
>>> mm.edges()
[('diff', 'intel'), ('diff', 'grade'), ('intel', 'grade'),
('intel', 'SAT'), ('grade', 'letter')]
"""
from pgmpy.models import MarkovModel
moral_graph = self.moralize()
mm = MarkovModel(moral_graph.edges())
mm.add_factors(*[cpd.to_factor() for cpd in self.cpds])
return mm
def get_model(self, threshold=0.95):
if isinstance(self.model, BayesianModel):
nodes = self.data.columns
self.model.add_nodes_from(nodes)
edges = []
for u, v in combinations(nodes, 2):
f_exp = self.data.groupby([u, v]).size().values
u_f_obs = self.data.ix[:, u].value_counts().values
v_f_obs = self.data.ix[:, v].value_counts().values
if stats.chisquare(
f_obs=[i * j for i in u_f_obs for j in v_f_obs],
f_exp=f_exp).pvalue < threshold:
edges.append((u, v))
self.model.add_edges_from(edges)
return nodes, edges
elif isinstance(self.model, MarkovModel):
nodes = self.data.columns
for node in nodes:
self.node_card[node] = self.data.ix[:, node].unique().size
self.model.add_nodes_from(nodes)
all_possible_edges = list(combinations(nodes, 2))
optimum = 1000000
optimal_edges = None
for edge_comb in chain(*[
combinations(all_possible_edges, r)
for r in range(1,
len(all_possible_edges) + 1)
]):
self.model = MarkovModel(edge_comb)
factors, score = self.get_parameters(score=True)
if score < optimum:
optimum = score
optimal_edges = edge_comb
return nodes, edge_comb
def to_markov_model(self):
"""
Converts the factor graph into markov model.
A markov model contains nodes as random variables and edge between
two nodes imply interaction between them.
Examples
--------
>>> from pgmpy.models import FactorGraph
>>> from pgmpy.factors import Factor
>>> G = FactorGraph()
>>> G.add_nodes_from(['a', 'b', 'c'])
>>> G.add_nodes_from(['phi1', 'phi2'])
>>> G.add_edges_from([('a', 'phi1'), ('b', 'phi1'),
... ('b', 'phi2'), ('c', 'phi2')])
>>> phi1 = Factor(['a', 'b'], [2, 2], np.random.rand(4))
>>> phi2 = Factor(['b', 'c'], [2, 2], np.random.rand(4))
>>> G.add_factors(phi1, phi2)
>>> mm = G.to_markov_model()
"""
from pgmpy.models import MarkovModel
mm = MarkovModel()
variable_nodes = self.get_variable_nodes()
if len(set(self.nodes()) - set(variable_nodes)) != len(self.factors):
raise ValueError(
'Factors not associated with all the factor nodes.')
mm.add_nodes_from(variable_nodes)
for factor in self.factors:
scope = factor.scope()
mm.add_edges_from(itertools.combinations(scope, 2))
mm.add_factors(factor)
return mm
import numpy as np
import pandas as pd
from pgmpy.models import MarkovModel
from pgmpy.estimators import PseudoMomentMatchingEstimator
# Generating some random data
raw_data = np.random.randint(low=0, high=2, size=(100, 4))
raw_data
data = pd.DataFrame(raw_data, columns=['A', 'B', 'C', 'D'])
data
# Diamond shaped Markov Model as stated in Fig. 6.1
markov_model = MarkovModel([('A', 'B'), ('B', 'C'), ('C', 'D'), ('D', 'A')])
markov_model.fit(data, estimator=PseudoMomentMatchingEstimator)
factors = coin_model.get_factors()
factors
def test_class_init_with_data_nonstring(self):
self.g = MarkovModel([(1, 2), (2, 3)])
import numpy as np
import pandas as pd
from pgmpy.models import MarkovModel
from pgmpy.estimators import MaximumLikelihoodEstimator
# Generating some random data
raw_data = np.random.randint(low=0, high=2, size=(100, 2))
raw_data
data = pd.DataFrame(raw_data, columns=['A', 'B'])
data
# Markov Model as stated in Fig. 6.5
markov_model = MarkovModel([('A', 'B')])
markov_model.fit(data, estimator=MaximumLikelihoodEstimator)
factors = coin_model.get_factors()
print(factors[0])
import numpy as np import pandas as pd from pgmpy.models import MarkovModel from pgmpy.estimators import MaximumLikelihoodEstimator # Generating random data raw_data = np.random.randint(low=0, high=2, size=(1000, 2)) data = pd.DataFrame(raw_data, columns=['X', 'Y']) model = MarkovModel() model.fit(data, estimator=MaximumLikelihoodEstimator) model.get_factors() model.nodes() model.edges()
# Computing the cluster belief according the formula stated
# above
cluster_belief = psi * messages_prod_factor
# As cluster belief represents a probability distribution in
# this case, thus it should be normalized
cluster_belief.normalize()
return cluster_belief
phi_a_b = Factor(['a', 'b'], [2, 2], [10, 0.1, 0.1, 10])
phi_a_c = Factor(['a', 'c'], [2, 2], [5, 0.2, 0.2, 5])
phi_c_d = Factor(['c', 'd'], [2, 2], [0.5, 1, 20, 2.5])
phi_d_b = Factor(['d', 'b'], [2, 2], [5, 0.2, 0.2, 5])
# Cluster 1 is a MarkovModel A--B
cluster_1 = MarkovModel([('a', 'b')])
# Adding factors
cluster_1.add_factors(phi_a_b)
# Cluster 2 is a MarkovModel A--C--D--B
cluster_2 = MarkovModel([('a', 'c'), ('c', 'd'), ('d', 'b')])
# Adding factors
cluster_2.add_factors(phi_a_c, phi_c_d, phi_d_b)
# Message passed from cluster 1 -> 2 should the M-Projection of psi1
# as the sepset of cluster 1 and 2 is A, B thus there is no need to
# marginalize psi1
delta_1_2 = compute_message(cluster_1, cluster_2)
def main():
# maximum size of segments grid cell
map_size = 50
# minimum number of pixels to consider segment
pixels_thresh = 500
# number of image in the database
num_images = 70
# counter for incorrectly classified segments
num_incorrect = 0
for idx in range(num_images):
print('\n\nImage ', idx)
# read image from disk
image = cv2.imread('export/image_%04d.jpg' % idx)
# read ground truth classes
labels_gt_im = cv2.imread('export/labels_%04d.png' % idx,
cv2.IMREAD_ANYDEPTH).astype(np.int)
labels_gt_im[labels_gt_im == 65535] = -1
# read a priori probabilities from classifier
prob_im = cv2.imread('export/prob_%04d.png' % idx,
cv2.IMREAD_ANYDEPTH) / 65535.0
# read image division into segments
segments_im = cv2.imread('export/segments_%04d.png' % idx,
cv2.IMREAD_ANYDEPTH)
# display image
cv2.imshow('original image', image)
cv2.waitKey(1)
# mean R, G, B values for each segment
mean_rgb = np.zeros([map_size, map_size, 3], dtype=np.float)
# number of pixels for each segment
num_pixels = np.zeros([map_size, map_size], dtype=np.uint)
# a priori probabilities for each segment
prob = np.zeros([map_size, map_size, 2], dtype=np.float)
# ground truth classes for each segment
labels_gt = -1 * np.ones([map_size, map_size], dtype=np.int)
for y in range(map_size):
for x in range(map_size):
# segment identifier
cur_seg = y * map_size + x
# indices of pixels belonging to the segment
cur_pixels = np.nonzero(segments_im == cur_seg)
if len(cur_pixels[0]) > 0:
num_pixels[y, x] = len(cur_pixels[0])
# flip because opencv stores images as BGR by default
mean_rgb[y,
x, :] = np.flip(np.mean(image[cur_pixels],
axis=0))
# average results from the classifier - should not be necessary
prob[y, x, 0] = np.mean(prob_im[cur_pixels])
prob[y, x, 1] = 1.0 - prob[y, x, 0]
# labeling boundaries don't have to be aligned with segments boundaties
# count which label is the most common in this segment
labels_unique, count = np.unique(labels_gt_im[cur_pixels],
return_counts=True)
labels_gt[y, x] = labels_unique[np.argmax(count)]
pass
# build model
# PUT YOUR CODE HERE
nodes = []
# if segment size is bigger than pixels_thresh, add current segment to node list
for y in range(map_size):
for x in range(map_size):
if num_pixels[y, x] > pixels_thresh:
nodes.append("s_" + str(y) + "_" + str(x))
factors_unary = []
# if segment size is bigger than pixels_thresh, add unary factor of segment to list
for y in range(map_size):
for x in range(map_size):
if num_pixels[y, x] > pixels_thresh:
# print(prob[y, x] * 0.9/1 + 0.05)
cur_f = create_factor(["s_" + str(y) + "_" + str(x)],
[[0, 1]], [0.945212], [unary_feat],
[prob[y, x]])
factors_unary.append(cur_f)
factors_pairway = []
edges_pairway = []
# if segment size is bigger than pixels_thresh and his neighbour too, add pairway factor to list
for y in range(map_size - 1):
for x in range(map_size - 1):
if num_pixels[y, x] > pixels_thresh:
# check neighnour to the right
if num_pixels[y, x + 1] > pixels_thresh:
cur_f = create_factor([
"s_" + str(y) + "_" + str(x),
"s_" + str(y) + "_" + str(x + 1)
], [[0, 1], [0, 1]], [1.86891, 1.07741, 1.89271], [
pairway_feat_red, pairway_feat_green,
pairway_feat_blue
], [mean_rgb[y, x], mean_rgb[y, x + 1]])
factors_pairway.append(cur_f)
edges_pairway.append(
("s_" + str(y) + "_" + str(x),
"s_" + str(y) + "_" + str(x + 1)))
# check neighbour under
if num_pixels[y + 1, x] > pixels_thresh:
cur_f = create_factor([
"s_" + str(y) + "_" + str(x),
"s_" + str(y + 1) + "_" + str(x)
], [[0, 1], [0, 1]], [1.86891, 1.07741, 1.89271], [
pairway_feat_red, pairway_feat_green,
pairway_feat_blue
], [mean_rgb[y, x], mean_rgb[y + 1, x]])
factors_pairway.append(cur_f)
edges_pairway.append(
("s_" + str(y) + "_" + str(x),
"s_" + str(y + 1) + "_" + str(x)))
# use Markov model, because MPLP doesn't support factor graphs
G = MarkovModel()
G.add_nodes_from(nodes) # add nodes
G.add_factors(*factors_unary) # add unary factors
G.add_factors(*factors_pairway) # add pairway factors
G.add_edges_from(edges_pairway) # add edges
# check if everything is ok
print("Check model: ", G.check_model())
# ------------------
# inferred image pixels
labels_infer = -1 * np.ones([map_size, map_size], dtype=np.int)
# read results of the inference
# PUT YOUR CODE HERE
mplp_infer = Mplp(G)
q = mplp_infer.map_query()
segment_cnt = 0
for y in range(map_size):
for x in range(map_size):
if num_pixels[y, x] > pixels_thresh:
val = q["s_" + str(y) + "_" + str(x)]
rv = None
if val == factors_unary[segment_cnt].values[0]:
rv = 0
else:
rv = 1
labels_infer[y, x] = rv
segment_cnt += 1
# ------------------
labels_class = -1 * np.ones([map_size, map_size], dtype=np.int)
for y in range(map_size - 1):
for x in range(map_size - 1):
if num_pixels[y, x] > pixels_thresh:
# label as the class with the highest probability
if prob[y, x, 0] >= 0.5:
labels_class[y, x] = 0
else:
labels_class[y, x] = 1
# count correct pixels
cnt_corr = np.sum(
np.logical_and(labels_gt == labels_infer, labels_gt != -1))
print('Accuracy = ', cnt_corr / np.sum(labels_gt != -1))
num_incorrect += np.sum(labels_gt != -1) - cnt_corr
# transfer results for segments onto image
labels_infer_im = -1 * np.ones_like(labels_gt_im)
labels_class_im = -1 * np.ones_like(labels_gt_im)
for y in range(map_size - 1):
for x in range(map_size - 1):
if labels_infer[y, x] >= 0:
cur_seg = y * map_size + x
cur_pixels = np.nonzero(segments_im == cur_seg)
labels_infer_im[cur_pixels] = labels_infer[y, x]
labels_class_im[cur_pixels] = labels_class[y, x]
# class 0 - green, class 1 - red in BGR
colors = np.array([[0, 255, 0], [0, 0, 255]], dtype=np.uint8)
image_infer = image.copy()
image_class = image.copy()
image_gt = image.copy()
# color pixels according to label
for l in range(2):
image_infer[labels_infer_im == l] = colors[l]
image_class[labels_class_im == l] = colors[l]
image_gt[labels_gt_im == l] = colors[l]
# show inferred, classified, and gt image by blending the original image with the colored one
image_infer_vis = (0.75 * image + 0.25 * image_infer).astype(np.uint8)
cv2.imshow('inferred', image_infer_vis)
image_class_vis = (0.75 * image + 0.25 * image_class).astype(np.uint8)
cv2.imshow('classified', image_class_vis)
image_gt_vis = (0.75 * image + 0.25 * image_gt).astype(np.uint8)
cv2.imshow('ground truth', image_gt_vis)
# uncomment to stop after each image
# cv2.waitKey()
cv2.waitKey(10)
print('Incorrectly inferred ', num_incorrect, ' segments')
def setUp(self):
self.graph = MarkovModel()
# THIS CODE HAS TO BE RUN ON PYTHON 2
# Otherwise, you will get wrong results
from pgmpy.models import MarkovModel
from pgmpy.factors.discrete import DiscreteFactor
from pgmpy.inference import BeliefPropagation
import numpy as np
# Construct a graph
PGM = MarkovModel()
PGM.add_nodes_from(['w1', 'w2', 'w3'])
PGM.add_edges_from([('w1', 'w2'), ('w2', 'w3')])
tr_matrix = np.array([1, 10, 3, 2, 1, 5, 3, 3, 2])
tr_matrix = np.array([1, 2, 3, 10, 1, 3, 3, 5, 2]).reshape(3, 3).T.reshape(-1)
phi = [DiscreteFactor(edge, [3, 3], tr_matrix) for edge in PGM.edges()]
print(phi[0])
print(phi[1])
PGM.add_factors(*phi)
# Calculate partition funtion
Z = PGM.get_partition_function()
print('The partition function is:', Z)
# Calibrate the click
belief_propagation = BeliefPropagation(PGM)
belief_propagation.calibrate()
# Output calibration result, which you should get
query = belief_propagation.query(variables=['w2'])
print('After calibration you should get the following mu(S):\n', query * Z)
