def estimate(self):
"""
Estimates the `DAG` structure that fits best to the given data set,
according to the scoring method supplied in the constructor.
Exhaustively searches through all models. Only estimates network structure, no parametrization.
Returns
-------
model: `DAG` instance
A `DAG` with maximal score.
Examples
--------
>>> import pandas as pd
>>> import numpy as np
>>> from pgmpy.estimators import ExhaustiveSearch
>>> # create random data sample with 3 variables, where B and C are identical:
>>> data = pd.DataFrame(np.random.randint(0, 5, size=(5000, 2)), columns=list('AB'))
>>> data['C'] = data['B']
>>> est = ExhaustiveSearch(data)
>>> best_model = est.estimate()
>>> best_model
<pgmpy.base.DAG.DAG object at 0x7f695c535470>
>>> best_model.edges()
[('B', 'C')]
"""
best_dag = max(self.all_dags(), key=self.scoring_method.score)
best_model = DAG()
best_model.add_nodes_from(sorted(best_dag.nodes()))
best_model.add_edges_from(sorted(best_dag.edges()))
return best_model
def estimate(self,
tabu_length=100,
max_indegree=2,
black_list=None,
epsilon=1e-4,
max_iter=1e6,
show_progress=True):
# We will be using K2Score for this model
score = K2Score(data=self.data)
# Model gets the score for a node and its parents
# This is used on every iteration for all possible changes
# This is greddy and picks the best available option
score_fn = score.local_score
# Initialize a Starting DAG
# PGMPY made a DAG class that adds some functionality to nx.DiGrpah
start_dag = DAG()
start_dag.add_nodes_from(self.variables)
# Set the edges we do not want to have in the graph
if black_list is None:
black_list = set()
else:
black_list = set(black_list)
# Just change Maxindegree to a certain number when doing the model
# I think this is to keep track of the changes we already made to the model
tabu_list = deque(maxlen=tabu_length)
# Initialize a current model
current_model = start_dag
if show_progress:
iteration = trange(int(max_iter))
else:
iteration = range(int(max_iter))
for _ in iteration:
# Get the best operations based on K2 score with self._legal_operations
best_operation, best_score_change = max(self._legal_operations(
model=current_model,
score=score_fn,
tabu_list=tabu_list,
max_indegree=max_indegree,
black_list=black_list,
),
key=lambda t: t[1])
if best_score_change < epsilon:
break
elif best_operation[0] == '+':
current_model.add_edge(*best_operation[1])
tabu_list.append(("-", best_operation[1]))
elif best_operation[0] == '-':
current_model.remove_edge(*best_operation[1])
tabu_list.append(("+", best_operation[1]))
elif best_operation[0] == 'flip':
X, Y = best_operation[1]
current_model.remove_edge(X, Y)
current_model.add_edge(Y, X)
tabu_list.append(best_operation)
return current_model
def estimate(
self, start=None, tabu_length=0, max_indegree=None, epsilon=1e-4, max_iter=1e6
):
"""
Performs local hill climb search to estimates the `DAG` structure
that has optimal score, according to the scoring method supplied in the constructor.
Starts at model `start` and proceeds by step-by-step network modifications
until a local maximum is reached. Only estimates network structure, no parametrization.
Parameters
----------
start: DAG instance
The starting point for the local search. By default a completely disconnected network is used.
tabu_length: int
If provided, the last `tabu_length` graph modifications cannot be reversed
during the search procedure. This serves to enforce a wider exploration
of the search space. Default value: 100.
max_indegree: int or None
If provided and unequal None, the procedure only searches among models
where all nodes have at most `max_indegree` parents. Defaults to None.
epsilon: float (default: 1e-4)
Defines the exit condition. If the improvement in score is less than `epsilon`,
the learned model is returned.
max_iter: int (default: 1e6)
The maximum number of iterations allowed. Returns the learned model when the
number of iterations is greater than `max_iter`.
Returns
-------
model: `DAG` instance
A `DAG` at a (local) score maximum.
Examples
--------
>>> import pandas as pd
>>> import numpy as np
>>> from pgmpy.estimators import HillClimbSearch, BicScore
>>> # create data sample with 9 random variables:
... data = pd.DataFrame(np.random.randint(0, 5, size=(5000, 9)), columns=list('ABCDEFGHI'))
>>> # add 10th dependent variable
... data['J'] = data['A'] * data['B']
>>> est = HillClimbSearch(data, scoring_method=BicScore(data))
>>> best_model = est.estimate()
>>> sorted(best_model.nodes())
['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J']
>>> best_model.edges()
[('B', 'J'), ('A', 'J')]
>>> # search a model with restriction on the number of parents:
>>> est.estimate(max_indegree=1).edges()
[('J', 'A'), ('B', 'J')]
"""
nodes = self.state_names.keys()
if start is None:
start = DAG()
start.add_nodes_from(nodes)
elif not isinstance(start, DAG) or not set(start.nodes()) == set(nodes):
raise ValueError(
"'start' should be a DAG with the same variables as the data set, or 'None'."
)
tabu_list = []
current_model = start
iter_no = 0
while iter_no <= max_iter:
iter_no += 1
best_score_delta = 0
best_operation = None
for operation, score_delta in self._legal_operations(
current_model, tabu_list, max_indegree
):
if score_delta > best_score_delta:
best_operation = operation
best_score_delta = score_delta
if best_operation is None or best_score_delta < epsilon:
break
elif best_operation[0] == "+":
current_model.add_edge(*best_operation[1])
tabu_list = ([("-", best_operation[1])] + tabu_list)[:tabu_length]
elif best_operation[0] == "-":
current_model.remove_edge(*best_operation[1])
tabu_list = ([("+", best_operation[1])] + tabu_list)[:tabu_length]
elif best_operation[0] == "flip":
X, Y = best_operation[1]
current_model.remove_edge(X, Y)
current_model.add_edge(Y, X)
tabu_list = ([best_operation] + tabu_list)[:tabu_length]
return current_model
class TestDAGCreation(unittest.TestCase):
def setUp(self):
self.graph = DAG()
def test_class_init_without_data(self):
self.assertIsInstance(self.graph, DAG)
def test_class_init_with_data_string(self):
self.graph = DAG([("a", "b"), ("b", "c")])
self.assertListEqual(sorted(self.graph.nodes()), ["a", "b", "c"])
self.assertListEqual(
hf.recursive_sorted(self.graph.edges()), [["a", "b"], ["b", "c"]]
)
def test_add_node_string(self):
self.graph.add_node("a")
self.assertListEqual(list(self.graph.nodes()), ["a"])
def test_add_node_nonstring(self):
self.graph.add_node(1)
def test_add_nodes_from_string(self):
self.graph.add_nodes_from(["a", "b", "c", "d"])
self.assertListEqual(sorted(self.graph.nodes()), ["a", "b", "c", "d"])
def test_add_nodes_from_non_string(self):
self.graph.add_nodes_from([1, 2, 3, 4])
def test_add_node_weight(self):
self.graph.add_node("weighted_a", 0.3)
self.assertEqual(self.graph.nodes["weighted_a"]["weight"], 0.3)
def test_add_nodes_from_weight(self):
self.graph.add_nodes_from(["weighted_b", "weighted_c"], [0.5, 0.6])
self.assertEqual(self.graph.nodes["weighted_b"]["weight"], 0.5)
self.assertEqual(self.graph.nodes["weighted_c"]["weight"], 0.6)
self.graph.add_nodes_from(["e", "f"])
self.assertEqual(self.graph.nodes["e"]["weight"], None)
self.assertEqual(self.graph.nodes["f"]["weight"], None)
def test_add_edge_string(self):
self.graph.add_edge("d", "e")
self.assertListEqual(sorted(self.graph.nodes()), ["d", "e"])
self.assertListEqual(list(self.graph.edges()), [("d", "e")])
self.graph.add_nodes_from(["a", "b", "c"])
self.graph.add_edge("a", "b")
self.assertListEqual(
hf.recursive_sorted(self.graph.edges()), [["a", "b"], ["d", "e"]]
)
def test_add_edge_nonstring(self):
self.graph.add_edge(1, 2)
def test_add_edges_from_string(self):
self.graph.add_edges_from([("a", "b"), ("b", "c")])
self.assertListEqual(sorted(self.graph.nodes()), ["a", "b", "c"])
self.assertListEqual(
hf.recursive_sorted(self.graph.edges()), [["a", "b"], ["b", "c"]]
)
self.graph.add_nodes_from(["d", "e", "f"])
self.graph.add_edges_from([("d", "e"), ("e", "f")])
self.assertListEqual(sorted(self.graph.nodes()), ["a", "b", "c", "d", "e", "f"])
self.assertListEqual(
hf.recursive_sorted(self.graph.edges()),
hf.recursive_sorted([("a", "b"), ("b", "c"), ("d", "e"), ("e", "f")]),
)
def test_add_edges_from_nonstring(self):
self.graph.add_edges_from([(1, 2), (2, 3)])
def test_add_edge_weight(self):
self.graph.add_edge("a", "b", weight=0.3)
if nx.__version__.startswith("1"):
self.assertEqual(self.graph.edge["a"]["b"]["weight"], 0.3)
else:
self.assertEqual(self.graph.adj["a"]["b"]["weight"], 0.3)
def test_add_edges_from_weight(self):
self.graph.add_edges_from([("b", "c"), ("c", "d")], weights=[0.5, 0.6])
if nx.__version__.startswith("1"):
self.assertEqual(self.graph.edge["b"]["c"]["weight"], 0.5)
self.assertEqual(self.graph.edge["c"]["d"]["weight"], 0.6)
self.graph.add_edges_from([("e", "f")])
self.assertEqual(self.graph.edge["e"]["f"]["weight"], None)
else:
self.assertEqual(self.graph.adj["b"]["c"]["weight"], 0.5)
self.assertEqual(self.graph.adj["c"]["d"]["weight"], 0.6)
self.graph.add_edges_from([("e", "f")])
self.assertEqual(self.graph.adj["e"]["f"]["weight"], None)
def test_update_node_parents_bm_constructor(self):
self.graph = DAG([("a", "b"), ("b", "c")])
self.assertListEqual(list(self.graph.predecessors("a")), [])
self.assertListEqual(list(self.graph.predecessors("b")), ["a"])
self.assertListEqual(list(self.graph.predecessors("c")), ["b"])
def test_update_node_parents(self):
self.graph.add_nodes_from(["a", "b", "c"])
self.graph.add_edges_from([("a", "b"), ("b", "c")])
self.assertListEqual(list(self.graph.predecessors("a")), [])
self.assertListEqual(list(self.graph.predecessors("b")), ["a"])
self.assertListEqual(list(self.graph.predecessors("c")), ["b"])
def test_get_leaves(self):
self.graph.add_edges_from(
[("A", "B"), ("B", "C"), ("B", "D"), ("D", "E"), ("D", "F"), ("A", "G")]
)
self.assertEqual(sorted(self.graph.get_leaves()), sorted(["C", "G", "E", "F"]))
def test_get_roots(self):
self.graph.add_edges_from(
[("A", "B"), ("B", "C"), ("B", "D"), ("D", "E"), ("D", "F"), ("A", "G")]
)
self.assertEqual(["A"], self.graph.get_roots())
self.graph.add_edge("H", "G")
self.assertEqual(sorted(["A", "H"]), sorted(self.graph.get_roots()))
def test_init_with_cycle(self):
self.assertRaises(ValueError, DAG, [("a", "a")])
self.assertRaises(ValueError, DAG, [("a", "b"), ("b", "a")])
self.assertRaises(ValueError, DAG, [("a", "b"), ("b", "c"), ("c", "a")])
def tearDown(self):
del self.graph
def pdag_to_dag(pdag):
"""Completes a PDAG to a DAG, without adding v-structures, if such a
completion exists. If no faithful extension is possible, some fully
oriented DAG that corresponds to the PDAG is returned and a warning is
generated. This is a static method.
Parameters
----------
pdag: DAG
A directed acyclic graph pattern, consisting in (acyclic) directed edges
as well as "undirected" edges, represented as both-way edges between
nodes.
Returns
-------
dag: DAG
A faithful orientation of pdag, if one exists. Otherwise any
fully orientated DAG/BayesianModel with the structure of pdag.
References
----------
[1] Chickering, Learning Equivalence Classes of Bayesian-Network Structures,
2002; See page 454 (last paragraph) for the algorithm pdag_to_dag
http://www.jmlr.org/papers/volume2/chickering02a/chickering02a.pdf
[2] Dor & Tarsi, A simple algorithm to construct a consistent extension
of a partially oriented graph, 1992,
http://ftp.cs.ucla.edu/pub/stat_ser/r185-dor-tarsi.pdf
Examples
--------
>>> import pandas as pd
>>> import numpy as np
>>> from pgmpy.base import DAG
>>> from pgmpy.estimators import ConstraintBasedEstimator
>>> data = pd.DataFrame(np.random.randint(0, 4, size=(5000, 3)), columns=list('ABD'))
>>> data['C'] = data['A'] - data['B']
>>> data['D'] += data['A']
>>> c = ConstraintBasedEstimator(data)
>>> pdag = c.skeleton_to_pdag(*c.estimate_skeleton())
>>> pdag.edges()
[('B', 'C'), ('D', 'A'), ('A', 'D'), ('A', 'C')]
>>> c.pdag_to_dag(pdag).edges()
[('B', 'C'), ('A', 'D'), ('A', 'C')]
>>> # pdag_to_dag is static:
... pdag1 = DAG([('A', 'B'), ('C', 'B'), ('C', 'D'), ('D', 'C'), ('D', 'A'), ('A', 'D')])
>>> ConstraintBasedEstimator.pdag_to_dag(pdag1).edges()
[('D', 'C'), ('C', 'B'), ('A', 'B'), ('A', 'D')]
>>> # example of a pdag with no faithful extension:
... pdag2 = DAG([('A', 'B'), ('A', 'C'), ('B', 'C'), ('C', 'B')])
>>> ConstraintBasedEstimator.pdag_to_dag(pdag2).edges()
UserWarning: PDAG has no faithful extension (= no oriented DAG with the same v-structures as PDAG).
Remaining undirected PDAG edges oriented arbitrarily.
[('B', 'C'), ('A', 'B'), ('A', 'C')]
"""
pdag = pdag.copy()
dag = DAG()
dag.add_nodes_from(pdag.nodes())
# add already directed edges of pdag to dag
for X, Y in pdag.edges():
if not pdag.has_edge(Y, X):
dag.add_edge(X, Y)
while pdag.number_of_nodes() > 0:
# find node with (1) no directed outgoing edges and
# (2) the set of undirected neighbors is either empty or
# undirected neighbors + parents of X are a clique
found = False
for X in pdag.nodes():
directed_outgoing_edges = set(pdag.successors(X)) - set(
pdag.predecessors(X))
undirected_neighbors = set(pdag.successors(X)) & set(
pdag.predecessors(X))
neighbors_are_clique = all((pdag.has_edge(Y, Z)
for Z in pdag.predecessors(X)
for Y in undirected_neighbors
if not Y == Z))
if not directed_outgoing_edges and (not undirected_neighbors
or neighbors_are_clique):
found = True
# add all edges of X as outgoing edges to dag
for Y in pdag.predecessors(X):
dag.add_edge(Y, X)
pdag.remove_node(X)
break
if not found:
warn(
"PDAG has no faithful extension (= no oriented DAG with the "
+
"same v-structures as PDAG). Remaining undirected PDAG edges "
+ "oriented arbitrarily.")
for X, Y in pdag.edges():
if not dag.has_edge(Y, X):
try:
dag.add_edge(X, Y)
except ValueError:
pass
break
return dag